JP4609883B2 - Fully charged capacity calculation device for power storage device for vehicle - Google Patents

Description

  The present invention relates to a full charge capacity calculation device for a vehicle power storage device using a neural network.

  In a secondary battery for a vehicle (also referred to as a power storage device), the charging / discharging state varies in a very wide range due to fluctuations in the running state, and it is difficult to accurately detect the full charge capacity of the secondary battery. It was. For this reason, in order to prevent overcharge and overdischarge, the usable charge / discharge range of the secondary battery has to be set narrow, and there has been a strong demand for improvement from the viewpoint of effective use of the battery.

As a solution to this problem, Patent Documents 1 and 2 below propose a neural network type battery state calculation method for calculating the full charge capacity and life of a secondary battery using a neural network (also referred to as a neural network). Yes.
Japanese Patent Laid-Open No. 2003-24971 JP-A-9-243716

  However, the determination accuracy based on the life and full charge capacity using the neural network type battery state detection technology according to Patent Documents 1 and 2 described above is a neural network that can flexibly change the calculation function of the battery characteristics due to the deterioration of each battery. Despite the use of arithmetic, it was not practical enough.

  Of course, it is highly possible that the calculation accuracy of the full charge capacity can be improved by performing a large-scale neural network calculation by putting all the electrical state quantities of the secondary battery into the neural network. However, the circuit scale and computation scale of a neural network computing device that can be mounted on the vehicle have strong restrictions in terms of cost, power consumption, and computation speed, and it is almost impossible to mount such a large-scale neural network computing device on the vehicle. .

  In other words, since the full charge capacity of the secondary battery has a correlation with the history of the electric battery state quantity such as voltage and current, the above history is input to a neural network that has learned the relationship between the history and the full charge capacity in advance. It is considered that calculating the full charge capacity of the power storage device for the vehicle by performing a neural network calculation is very effective in calculating the full charge capacity of the power storage device for the vehicle where the fluctuation of the battery state amount is large. Therefore, it is inevitable to improve the detection accuracy while suppressing the increase in the scale of the neural network operation.

  The present invention has been made in view of the above problems, and provides a full charge capacity calculation device for a power storage device for a vehicle that realizes an improvement in full charge capacity detection accuracy while suppressing an increase in the scale of neural network calculation. It is aimed.

A full charge capacity calculation device for a power storage device for a vehicle according to the present invention that solves the above-mentioned problem is a voltage / current history detection means for detecting and outputting a voltage history and a current history of a predetermined time immediately before a chargeable / dischargeable battery , A full charge capacity calculation device for a power storage device for a vehicle, comprising: a calculation means for performing a neural network calculation of the full charge capacity of the battery using the voltage history and the current history as input parameters. Based on the approximate expression obtained by the method of least squares, the open circuit voltage and the internal resistance at the time of discharging a predetermined discharge amount from the fully charged state of the battery are calculated, and in addition to the voltage history and the current history, the open circuit voltage and the internal resistance are calculated. As an input parameter, a neural network calculation is performed on the full charge capacity of the battery as an output parameter.

  That is, according to the present invention, in addition to a group of secondary battery voltage / current pairs (that is, voltage history and current history) sampled at a predetermined timing in the immediately preceding predetermined period, the open circuit voltage and the internal resistance during a predetermined capacity discharge are small. It has been found that the full charge capacity detection accuracy that can withstand practical use can be obtained from the data. That is, according to the test results, it was found that the neural network calculation accuracy of the full charge capacity can be remarkably improved only by adding a small amount of data such as the open circuit voltage and the internal resistance at the time of discharging the predetermined capacity to the voltage history and current history of the battery. .

  Therefore, according to the present invention, the above-described conventional full charge capacity calculation method can be performed with high accuracy compared with the conventional full charge capacity calculation method that does not use the neural network calculation within a calculation time required for practical use by a small-scale neural network calculation. Compared to the neural network calculation, the full charge capacity can be detected with a small calculation scale and high accuracy. As a result, the usage capacity range can be expanded without fear of overcharge and overdischarge of the on-vehicle battery, and as a result, the required discharge capacity range can be covered by a battery with a much smaller capacity than before, The on-board battery space can be reduced and the vehicle weight can be reduced.

  Note that the voltage (that is, the terminal voltage) and the current history (charging / discharging current) constituting the voltage history are voltage / current pairs sampled at the same time as described above. The open circuit voltage at the time of discharging from the fully charged state to the predetermined capacity is a state in which 0 to 30% of the initial fully charged capacity is discharged from the fully charged state, more preferably 2 to 20%, more preferably 3 It can be in a state of being discharged by 10%.

  However, in the present invention, at least the voltage history and current history, the open circuit voltage at the time of predetermined capacity discharge, and the internal resistance are used, but another state quantity may be used as an input parameter for the neural network calculation. However, in terms of the gist of the present invention, it is practically important to prevent an increase in the amount of operation of the neural network by adding this input parameter, and the addition should be made within a range where the increase in the operation scale does not exceed 50%.

  Furthermore, in the present invention, a voltage history and current history in a predetermined period immediately before, that is, a group of sampled voltage / current pairs are stored for neural network calculation. The group of stored voltage / current pairs can be obtained only by processing by the least square method, and the circuit scale, particularly the data storage amount, can be saved.

  In a preferred aspect, the calculating means calculates a battery deterioration level (= the full charge capacity / the full charge capacity as a ratio between the full charge capacity obtained by the current neural network calculation and an initial value of the full charge capacity of the battery stored in advance. (Initial full charge capacity)) is calculated. Thereby, for example, the battery life and the replacement time can be determined based on the degree of battery deterioration.

  A neural network calculation method for a power storage device for a vehicle according to the present invention will be specifically described with reference to the accompanying drawings.

(overall structure)
A neural network calculation method for the power storage device for a vehicle according to the first embodiment will be described below. First, the circuit configuration of the apparatus will be described with reference to the block diagram shown in FIG.

  101 is an in-vehicle power storage device (hereinafter also referred to as a battery), 102 is an in-vehicle generator that charges the in-vehicle power storage device, 103 is an electric device that forms an in-vehicle electric load fed from the in-vehicle power storage device 101, and 104 is an in-vehicle power storage device 101 A current sensor for detecting the charging / discharging current of the battery, 105 is a storage battery state detection device which is an electronic circuit device for detecting the state of the in-vehicle power storage device 101, and 106 stores the voltage and current of the input battery as a voltage history and a current history. And a buffer unit 107 that calculates and outputs the current value of the open circuit voltage and / or the current value of the internal resistance, and 107 outputs various input signals input from the buffer unit 106 and a correction signal generator 109 described later as a neural network. The neural network unit that calculates and inputs the full charge capacity by calculation, 108 is read from the neural network unit 107, etc. A generator control device 109 for controlling the amount of power generated by the in-vehicle generator 102 based on the signal, 109 calculates an open circuit voltage and internal resistance at the time of predetermined capacity discharge as calibration data, which will be described later, and inputs data to the neural network unit 107 Is a correction signal generation unit that outputs as

  That is, in this embodiment, the storage battery state detection device 105 calculates the open circuit voltage and internal resistance at the time of predetermined capacity discharge, in addition to the buffer unit 106 and the neural network unit 107, and inputs the calibration input to the neural network unit 107. It has a feature that it has a correction signal generator 109 that outputs data. In this embodiment, the buffer unit 106, the neural network unit 107, and the correction signal generation unit 109 are realized by software calculation by a microcomputer device, but may be configured by a dedicated hardware circuit.

  However, in FIG. 1, the correction signal generation unit 109 is described as a circuit to which the voltage of the in-vehicle power storage device 101 and the current of the current sensor are input, but actually, the correction signal generation unit 109 is the buffer unit 106. Similarly to the neural network unit 107, it is constituted by microcomputer software, and calculates a group of voltage / current pairs held in the RAM or register of the microcomputer to calculate an open circuit voltage and an internal resistance at a predetermined capacity discharge.

(Buffer 106)
The buffer unit 106 is a pre-signal processing circuit of the neural network unit 107, and simultaneously samples the voltage of the in-vehicle power storage device 101 and the current from the current sensor 104 at regular intervals to obtain a battery voltage history and a current history. The voltage and current at each time point are stored in parallel and output to the neural network unit 107 in parallel. In order to reduce the numerical limit of the input cell of the neural network unit 107 and the calculation burden, the voltage / current sampling data constituting the battery voltage history and current history is composed of data up to a predetermined time point retroactive from the present time. .

(Correction signal generator 109)
The correction signal generation unit 109 calculates the open circuit voltage and internal resistance at the time of predetermined capacity discharge from full charge, and outputs the open circuit voltage and internal resistance at the time of predetermined capacity discharge to the neural network unit 107 as calibration data in the neural network calculation. To do. The correction signal generator 109 is illustrated in the flowchart of FIG.

  The correction signal generator 109 is started by starting running (step 601), and detects the battery current and terminal voltage (step 602). A full charge determination (described later) is performed for the detected current / terminal voltage (step 603), and if it is full charge, the subsequent charge / discharge current integration is started (step 604), and the integrated current value (Ah) is It is determined whether or not the predetermined discharge amount has been reached (step 605), and when it is reached, the open circuit voltage at this time is calculated (step 606) and rewritten as the open circuit voltage at the time of predetermined capacity discharge (step 607). The internal resistance of the battery at the time is calculated (step 608), and this is rewritten as the internal resistance at the time of discharging a predetermined capacity (step 609).

The full charge determination described in step 603 will be described in more detail with reference to FIG. The full charge determination is stored in advance as a predetermined area in the two-dimensional space of the battery voltage and current. When the input current / voltage characteristics enter the predetermined area (see FIG. 3), it is determined that the battery is fully charged.

  The calculation for obtaining the open circuit voltage and the internal resistance at the time of discharging the predetermined capacity from the full charge described in Steps 606 and 608 will be described in more detail with reference to FIG.

  An approximate expression indicating the relationship between voltage and current is obtained by a least square method from a predetermined number of voltage / current pairs input during a predetermined period immediately before the discharge of a predetermined capacity from full charge, and the open circuit voltage ( Battery voltage when the current is considered to be 0, which is also referred to as open-circuit voltage), and the internal resistance is obtained as the slope of this approximate expression, and these are used as the open circuit voltage and internal resistance during the above-mentioned predetermined capacity discharge. . In order to improve the accuracy of the above linear approximation, it is preferable to obtain the polarization state of the battery from the past current information and express it as a polarization index, and to select data in which this polarization index is within a predetermined range. Since the creation of a linear approximation formula using this kind of least square method and the extraction of the open circuit voltage and internal resistance using this linear approximation formula are known matters, further explanation is omitted.

(Neural network unit 107)
The neural network unit 107 is schematically shown in FIG. However, as described above, since the neural network unit 107 is actually configured by software operations sequentially performed at a predetermined operation interval, the circuit configuration illustrated in FIG. 5 is merely functional.

  Although the neural network unit 107 for calculating the full charge capacity shown in FIG. 5 has a form of learning by a three-layer feedforward type error back propagation method, it is not limited to this form. The input layer 201 includes a predetermined number of input cells. Each input cell has voltage history data Vi and current history data Ii from the buffer unit 106, and an open circuit voltage Vo and internal resistance r at the time of predetermined capacity discharge as calibration data input from the correction signal generation unit 109. The data is output to all the calculation cells in the intermediate layer 202. In this embodiment, the voltage history data Vi and the current history data Ii are each composed of five points of data sampled at a constant interval, but are not limited thereto. For example, the voltage or current may be data that is more than a predetermined amount away from other data.

  Each calculation cell of the intermediate layer 202 performs a neural network calculation described later on each input data input from each input cell of the input layer 201, and outputs a full charge capacity as a calculation result to the output cell of the output layer 203. The output layer 203 outputs the full charge capacity to the outside.

If the input data of the jth cell of the input layer 201 of the neural network unit 107 is INj, and the coupling coefficient of the jth cell of the input layer 201 and the kth cell of the intermediate layer 202 is Wjk, the input data to the kth cell of the intermediate layer The input signal is
INPUTk (t) ＝ Σ (Wjk * INj) (j = 1 to 2m + 3)
It becomes. The output signal from the kth cell of the intermediate layer is
OUTk (t) = f (x) = f (INPUTk (t) + b)
It is represented by b is a constant. f (INPUTk (t) + b) is a non-linear function called a sigmoid function with INPUTk (t) + b as an input variable.
f (INPUTk (t) + b) = 1 / (1 + exp (− (INPUTk (t) + b)))
Is a function defined by If the coupling coefficient between the kth cell of the intermediate layer 202 and the cell of the output layer 203 is Wk, the input signal to the output layer is
INPUTo (t) = Σ Wk * OUTk (t)
k = 1 to Q
It is represented by Q is the number of cells in the intermediate layer 202. The output signal at time t is
OUT (t) = L * INPUTo (t)
It becomes. L is a linear constant.

In this specification, the learning process means that each cell has a minimum error between a final output OUT (t) at time t and a teacher signal (that is, a true value tar (t)) measured in advance. Is to optimize the coupling coefficient between. The output OUT (t) is an output parameter to be output by the output layer 203, and here is the full charge capacity at the time t.

  Next, a method for updating each coupling coefficient will be described.

The update of the coupling coefficient Wk between the kth cell in the intermediate layer and the cell in the output layer is
Wk = Wk + △ Wk
Done in Here, ΔWk is defined as follows.

△ Wk = −η * ∂Ek / ∂Wk η; Constant
= η * [OUT (t) − tar (t)] * [∂OUT (t) / ∂Wk]
= η * [OUT (t) − tar (t)] * L * [∂INPUTo (t) / ∂Wk]
= η * L * [OUT (t) − tar (t)] * OUTk (t)
It is represented by Ek is an amount representing an error between teacher data and network output, and is defined by the following equation.

Ek = [OUT (t)-tar (t)] x [OUT (t)-tar (t)] / 2
Next, a rule for updating the coupling coefficient Wjk between the kth cell in the intermediate layer 202 and the jth cell in the input layer 201 will be described. The update of the coupling coefficient Wjk is realized by the following equation.

Wjk = Wjk + △ Wjk
Here, ΔWjk is defined as follows.

△ Wjk =-η * ∂Ek / ∂Wjk
=-Η * [∂Ek / ∂INPUTk (t)] * [∂INPUTk (t) / ∂Wjk]
=-Η * [∂Ek / ∂OUTk (t)] * [∂OUTk (t) / ∂INPUTk (t)] * INj
=-Η * [∂Ek / ∂OUT (t)] * [∂OUT (t) / ∂INPUTo] *
[∂INPUTo / OUTk (t)] * f '(INPUTk (t) + b) * INj
= −η * (OUT (t) −tar (t)) * L * Wk * f ′ (INPUTk (t) + b) * INj
= −η * L * Wk * INj * ( OUT (t) −tar (t)) * f ′ (INPUTk (t) + b)
Here, f ′ (INPUTk (t) + b) is a differential value of the transfer function f.

The output OUT (t), that is, the full charge capacity at time t is calculated again with the new coupling coefficients Wk and Wjk updated in this way, and the coupling coefficient is continuously updated until the error function Ek becomes a predetermined minute value or less. The process of updating the coupling coefficient so that the error function Ek becomes a predetermined value or less is the learning process.

  A flowchart of the learning process will be described with reference to FIG. However, the neural network unit 107 outputs the current full charge capacity.

First, an appropriate initial value of each coupling coefficient of the neural network unit 107 is set (step 302). This may be determined appropriately using, for example, a random number. Next, a predetermined input signal for learning is individually input to each cell of the input layer 201 of the neural network unit 107 (step 303), and this input signal is subjected to neural network calculation using the initial value of the coupling coefficient described above. Thus, the full charge capacity as an output parameter is calculated (step 304).

  Next, the error function Ek is calculated by the above method (step 305), and it is determined whether or not this error function is smaller than a predetermined minute value th (step 306). If the error function Ek is larger than a predetermined minute value th, the update amount ΔW of each coupling coefficient defined in the learning process is calculated (step 307), and each coupling coefficient is updated (step 308).

Next, the learning input signal described above is input again to each cell of the input layer 201 to calculate the full charge capacity as an output parameter. Next, the miscalculation function Ek is evaluated, and if it is below the minute value th, it is determined that the learning has been completed (step 309), and this learning process is terminated. If the error function Ek is not below the minute value, the coupling coefficient is updated again to calculate the full charge capacity , the error function Ek is evaluated, and this process is repeated until the error function Ek falls below the minute value.

  Therefore, if a typical charge / discharge pattern as described above is learned for some battery types in the neural network unit 107 before shipment of the product, the learning result can be written in the neural network unit 107. Thus, it is possible to sequentially calculate the full charge capacity of the on-vehicle storage battery that is running.

  When the full charge is not determined or when the open circuit voltage at the predetermined capacity discharge is not detected from the full charge, the value obtained previously as the open circuit voltage at the predetermined capacity discharge is held. Further, if the open circuit voltage and the internal resistance at the time of discharging the predetermined capacity change, the full charge capacity can be accurately detected according to the deterioration of the battery by updating and holding it.

(Test results)
Test results obtained by measuring the full charge capacity of the test battery in the learned neural network unit 107 will be described below.

  Nine vehicle lead-acid batteries with various degradation states and an initial full charge capacity of 27 Ah were used as test products. The current full charge capacity of each battery was calculated from the accumulated current value after discharging from the fully charged state until the terminal voltage became 10.5V under the discharge condition of 0.2C. These test batteries were connected to the above-described neural network unit 107 mounted on the vehicle, and the vehicle was run under the 10.15 mode driving condition to perform a neural network operation, and the full charge capacity was calculated through the neural network operation. However, the open circuit voltage and the internal resistance at the time of discharging the predetermined capacity described above were the open circuit voltage and the internal resistance at the time when 5.0 Ah was discharged from the fully charged state. As described above, the voltage history and the current history are constituted by five voltage / current pairs sampled immediately before at a predetermined interval. The average value of the detection error of the full charge capacity of each test battery obtained from the time of 5 Ah discharge to the end of travel after obtaining full charge determination during travel is shown below.

Test item 1 Full charge capacity 18.2Ah
Detection error 2.3Ah
Test product 2 Fully charged capacity 21.8Ah
Detection error 0.6Ah
Test item 3 Fully charged capacity 10.5Ah
Detection error 0.6Ah
Test item 4 Full charge capacity 10.0Ah
Detection error 0.1Ah
Test product 5 Fully charged capacity 18.3 Ah
Detection error 2.1Ah
Test item 6 Fully charged capacity 21.2Ah
Detection error 1.2Ah
Test item 7 Full charge capacity 24.3Ah
Detection error 3.4Ah
Test article 8 Full charge capacity 27.6Ah
Detection error 0.2Ah
Test item 9 Fully charged capacity 25.1Ah
Detection error 3.3Ah
Next, in the above-described neural network unit 107, although the voltage history and current history are used as input parameters, the full charge capacity calculation result when the open circuit voltage and internal resistance at the time of predetermined capacity discharge as calibration data are not used is as follows: It describes. The test conditions are the same as above.

Test item 1 Full charge capacity 18.2Ah
Detection error 3.9Ah
Test product 2 Fully charged capacity 21.8Ah
Detection error 2.8Ah
Test item 3 Fully charged capacity 10.5Ah
Detection error 5.4Ah
Test item 4 Full charge capacity 10.0Ah
Detection error 5.7Ah
Test product 5 Fully charged capacity 18.3 Ah
Detection error 4.4Ah
Test item 6 Fully charged capacity 21.2Ah
Detection error 3.4Ah
Test item 7 Full charge capacity 24.3Ah
Detection error 1.7Ah
Test article 8 Full charge capacity 27.6Ah
Detection error 2.8Ah
Test item 9 Fully charged capacity 25.1Ah
Detection error 2.7Ah
FIG. 7 illustrates detection errors with respect to the actual full charge capacity, depending on whether or not the open circuit voltage and the internal resistance at the time of discharging the predetermined capacity are used as calibration data. It has been found that the calculation accuracy of the fully charged capacity of a deteriorated battery can be significantly improved by only slightly increasing the input data from 10 to 12 points. In addition, when the full charge capacity obtained by the above-described neural network calculation becomes less than a predetermined ratio with respect to the initial full charge capacity stored in advance, it can be determined that the battery has reached the end of its life and is ready for replacement.

It is a block diagram which shows the circuit structure of the apparatus of an Example. It is a flowchart which shows the calculation method of the open circuit voltage and internal resistance at the time of predetermined capacity discharge from full charge in driving | running | working of an Example. It is a figure which shows the full charge area | region for the full charge determination of an Example. It is a figure which shows the example of the approximate expression for obtaining the open circuit voltage and internal resistance at the time of predetermined capacity | capacitance discharge from the full charge of an Example. It is a block diagram which shows the structure of the neural network part for full charge capacity | capacitance detection. It is a flowchart of the neural network part of FIG. It is a figure which shows the comparison result of the detection error of a full charge capacity by the case where the open circuit voltage and internal resistance at the time of predetermined capacity discharge are used, and the case where it does not use.

Explanation of symbols

101 on-vehicle power storage device 102 on-vehicle generator 104 current sensor 105 storage battery state detection device (calculation means)
106 Buffer unit 107 Neural network unit (Neural network unit)
108 generator control device 109 correction signal generation unit 201 input layer 202 intermediate layer 203 output layer

Claims (2)

Voltage / current history detection means for detecting and outputting voltage history and current history for a predetermined time immediately before a chargeable / dischargeable battery, and neural network calculation of the full charge capacity of the battery using the voltage history and current history as input parameters A full charge capacity computing device for a power storage device for a vehicle comprising computing means for
The computing means is
Based on an approximate expression obtained by the least square method from the voltage history and the current history, the open circuit voltage and the internal resistance at the time of discharging a predetermined discharge amount from the fully charged state of the battery are calculated,
A full charge capacity calculation device for a power storage device for a vehicle, wherein the open circuit voltage and internal resistance are used as input parameters in addition to the voltage history and current history, and a full charge capacity of the battery as an output parameter is calculated by a neural network . The full charge capacity computing device for a power storage device for a vehicle according to claim 1,
The computing means is
Calculate the battery deterioration level (= full charge capacity / (initial full charge capacity)) as a ratio between the full charge capacity obtained by the neural network calculation this time and the initial value of the full charge capacity of the battery stored in advance. A full charge capacity computing device for a power storage device for a vehicle.

Images