TWI741632B - Prediction method for temperature coefficient of resistance, compensation method for current measurement and the device thereof on an intelligent current sharing module of batteries - Google Patents

Abstract

The present invention provides a prediction method for temperature coefficient of resistance, compensation method for current measurement and the device thereof on an intelligent current sharing module of batteries, including the steps of: finding the temperature coefficient of resistance through an intelligent correction mechanism of artificial neural network to avoid interferences and noises; smoothing the data; fitting curves by intelligent algorithm of artificial neural network; figuring out the weighting and bias between the output layer, hidden layer and output layer neurons by using back propagation; and lastly, recording all the weighting and bias on a MCU of a microprocessor to compensate for electric current. Therefore, the present invention can make up for the insufficiency of traditional quadratic curve fitting algorithm and predict the non-linear phenomena of the height of copper composition’s resistance. In addition, the accuracy of current measurement can be within the range of ±0.1% so that it can accurately and easily compensate for the non-linear phenomena of the height of copper composition’s resistance.

Description

Resistance temperature coefficient prediction method, current measurement compensation correction method and device of battery intelligent shunt module

The present invention provides a method for predicting the temperature coefficient of resistance of a battery intelligent shunt module, a method for current measurement compensation and correction, and a device thereof, in particular, a method that uses a neural network to predict the temperature coefficient of resistance and can calculate and compensate the measured current .

Press, the shunt is mainly a resistor that can pass a large current, so that when the current passes, the voltage can be measured by measuring its voltage, and then converted into a current, so as to measure the large current; and the copper alloy shunt, It will thermally expand and contract with the change of the external environment temperature, which will affect its own resistance value, so it will cause the deviation of the measured current calculation, as shown in the following mathematical formula 1: [Mathematical formula 1] I=V /(R+△R)

Among them, I is the current; V is the voltage; R is the original resistance value of the copper alloy; △R is the variation of resistance with the external environment and temperature; and its variation has very complex nonlinear variables, as shown in Figure 1. As shown, it is obvious that the measured current cannot accurately represent the total current of the original output. If the traditional quadratic curve fitting is used, it cannot overcome the complex nonlinear characteristics.

As far as current technology is concerned, the current measurement error accuracy of the copper alloy shunt developed is within the range of 2~5%, and the error range is still high, so the accuracy is still insufficient.

In view of this, our inventors have devoted themselves to further research on the measurement of shunt current, and proceeded to develop and improve, hoping to develop a better invention to solve the above problems, and after continuous testing and modification, the present invention come out.

The purpose of the present invention is to solve the aforementioned problems. In order to achieve the above objectives, our inventors provide a method for predicting the temperature coefficient of resistance of a battery intelligent shunt module. The steps include: (a) Calculating a dependent output current And the temperature coefficient of resistance of the sensed current; (b) construct a type of neural network, and use an environmental influence factor as the input layer; (c) define a predictive output value and an actual value that depend on the output layer of the neural network (D) Perform a forward pass algorithm like neural network, substituting the predicted output value and actual value into and calculate the loss function; (e) Use the backward pass algorithm to calculate the weight and offset of this type of neural network The partial derivative of the value is used as the iterative update amount; (f) The weight and offset value of the iterative update are substituted into this type of neural network for recursion; (g) Steps (d) to (f) are repeated until the loss function Approaching a specific value; and (h) the output layer of the neural network is trained and defined as the predicted temperature coefficient of resistance.

According to the above-mentioned method for predicting the resistance temperature coefficient of the battery intelligent shunt module, this type of neural network is created by an excitation function.

，且該特定值為0；其中，

係該輸出值，而Y為該實際值。 According to the above-mentioned method for predicting the resistance temperature coefficient of the battery intelligent shunt module, the loss function is

, And the specific value is 0; where,

Is the output value, and Y is the actual value.

The present invention also provides a current measurement compensation correction method for a battery intelligent shunt module, which includes the resistance predicted by the method for predicting the temperature coefficient of resistance of the battery intelligent shunt module according to any one of claims 1 to 3 For the temperature coefficient, the steps include: (i) obtaining the output current from the predicted temperature coefficient of resistance according to step (a).

The present invention also provides a battery intelligent shunt module, which includes: a processing unit equipped with the resistance temperature coefficient prediction method of the battery intelligent shunt module as described in any one of claims 1 to 3 after training The weight and offset value are calculated by calculating the output current based on the sensed current by the current measurement compensation correction method of the battery smart shunt module as described in claim 1.

According to the above-mentioned battery intelligent shunt module device, it further includes a shunt, and the shunt is a copper alloy shunt.

Based on the above description and settings, it is obvious that the present invention mainly has the following advantages and effects, which are detailed as follows: 1. The present invention can make up for the shortcomings of traditional quadratic curve fitting algorithms and is used to predict the resistance height of copper alloys. The non-linear phenomenon of copper alloy resistance, and the accuracy of current measurement error can be within ±0.1%, which can accurately and easily compensate the highly non-linear phenomenon of copper alloy resistance.

S001~S008: steps

X: Input layer

H: hidden layer

:輸出層

: Output layer

W1: first neuron weight

B1: first neuron offset value

W2: second neuron weight

B2: The second neuron offset value

f(x): excitation function

The first figure is the experimental result of measuring current versus temperature of the conventional copper alloy shunt with environment and temperature variation.

Figure 2 is a flowchart of the present invention.

Figure 3 is a schematic diagram of the architecture of the neural network of the present invention.

Figure 4 is a graph showing the experimental results of the present invention using a neural network such as five neurons.

Figure 5 is a graph showing the experimental results of the present invention using a neural network such as ten neurons.

Regarding the technical means of our inventors, several preferred embodiments are described in detail below in conjunction with the drawings, so as to provide a thorough understanding and approval of the present invention.

Please refer to Figure 2 first. The present invention is a method for predicting the temperature coefficient of resistance of a battery intelligent shunt module. The architecture of the execution steps can include a current load supply, a voltage supply, a power amplifier, and a high-efficiency CAN data communication. Module and constant temperature and humidity box, the steps performed by the present invention include:

S001: Calculate a resistance temperature coefficient that depends on the output current and the sensed current; in one embodiment, the relationship is shown in Mathematical Formula 2: [Mathematical Formula 2] I _All = I _Sensor × α

Among them, I _All is the original output current, and I _Sensor is the sensing current measured by the shunt, and α is the temperature coefficient of resistance; Mathematical formula 2 can be simplified as shown in Mathematical formula 3:

The output current of the power supply and the sensed current measured by the shunt can be obtained before calibration, and the temperature coefficient of resistance α under the current environment and temperature can be obtained by substituting them into Mathematical formula 3.

S002: Construct a type of neural network, which can be a conventional AI single-layer neural network intelligent calculation architecture, and its general calculation formula is shown in the following mathematical formula 4:

In this embodiment, the AI single-layer neural network intelligent calculation architecture used is shown in Figure 3, and its formula can be expressed as the following mathematical formula 5:

為輸出層；W_n係各個神經元的權重，其為二維矩陣；B_n為各神經元之偏移值，其係為一向量；f_n係第n次輸出時之激勵函數，於本實施例中，係採用Sigmoid(x)=1/(1+e^-x)為激勵函數，如第4圖所示，藉可對數值進行非線性處理；藉此，係可藉由將一環境影響因數作為輸入層輸入類神經網路； Among them, X is the input layer; H _n represents the nth hidden layer;

Is the output layer; W _n is the weight of each neuron, which is a two-dimensional matrix; B _n is the offset value of each neuron, which is a vector; f _n is the activation function at the nth output, in this In the embodiment, Sigmoid(x)=1/(1+e ^-x ) is used as the excitation function. As shown in Figure 4, the numerical value can be processed nonlinearly; thus, the environment can be The influence factor is used as the input layer input neural network;

S003: In order to find the weight and offset value of each neuron in a single neural layer, the present invention defines a loss function e dependent on the predicted output value of the neural network output layer and an actual value; the loss function is In an embodiment, the system may be as shown in the following mathematical formula 6:

係該輸出值，而Y為該實際值。 in,

Is the output value, and Y is the actual value.

及實際值Y代入數學式6並計算該損失函數e。 S004: Perform a forward pass algorithm like neural network to predict the output value

And the actual value Y is substituted into Mathematical formula 6 and the loss function e is calculated.

S005: Backpropagation can then be used to calculate _{the partial derivative of the weight W n} and the offset value B _n of this type of neural network as the iterative update amount, as shown in the following mathematical formulas 7 and 8:

Among them, η is the learning efficiency.

S006: Substitute the weights and offsets of the iterative update into this type of neural network (Math. 5) for recursion.

S007: Repeat the aforementioned steps S004 to S006 until the loss function e approaches a specific value. In this embodiment, the specific value is 0.

趨近於實際值Y，亦即使損失函數e趨近於0。 In this way, the output of each neuron in each layer of neural network is calculated according to the same rule. The input layer is firstly inner product with the weight, then the offset value is added, and finally the output is converted through the excitation function; and training refers to Use backward transfer calculation iteratively to optimize the weight W _n and the offset value B _n to make the output value

Approaching the actual value Y , even if the loss function e approaches 0.

S008: As mentioned above, by using the environmental influence factor as the input layer X, the output layer of this type of neural network is defined as the predicted temperature coefficient of resistance after training, so that the resistance can be obtained more accurately Temperature coefficient α.

Accordingly, in the current measurement compensation correction method of a battery intelligent shunt module, the resistance temperature coefficient α obtained after the training can be substituted into Mathematical Formula 3, and the calculated current can be calculated based on the measured sensing current. The correct output current.

Thereby, the present invention can be applied to a battery intelligent shunt module device, which includes a shunt, which can be a copper alloy shunt, and has a processing unit, which can be a chip of a microprocessor (MCU), and apply the weight and offset value of this type of neural network to the processing unit through storage, programming, etc., so that the processing unit can execute and predict the temperature coefficient of resistance α, and as mentioned above, It is substituted into Mathematical formula 3 to estimate the output current.

As a result, as shown in Figures 5 and 6, it is the result of verification using neural networks such as 5 neurons and 10 neurons when the present invention is applied to the shunt. It can be seen that 10 The accuracy of a neural network such as a neuron is relatively high, and it can be proved that the present invention can accurately fit and compensate the highly nonlinear phenomenon of copper alloy resistance, and can obtain the correct output current.

In summary, the technical means disclosed in the present invention can effectively solve the conventional problems and achieve the expected purpose and effect. It has not been seen in the publications, has not been used publicly, and has long-term progress before the application. The patent law claims that the invention is correct. Yan filed an application in accordance with the law and prayed for the detailed examination and grant of the invention patent.

However, the above are only a few preferred embodiments of the present invention, and should not be used to limit the scope of implementation of the present invention, that is, all equivalent changes and modifications made in accordance with the scope of the patent application of the present invention and the content of the description of the invention are all It should still fall within the scope of the patent for this invention.

S001~S008: steps

Claims (6)

A method for predicting the temperature coefficient of resistance of a battery intelligent shunt module. The steps include: (a) calculating a resistance temperature coefficient dependent on the total output current of a power supply and the sensing current measured by a shunt; (b) ) Construct a type of neural network, and use an environmental influence factor as the input layer; (c) Define a loss function that depends on the predicted output value of the neural network output layer and an actual value; (d) Perform a neural network The forward pass algorithm, which substitutes the predicted output value and the actual value and calculates the loss function; (e) uses the backward pass algorithm to calculate the partial derivative of the weight and offset value of this type of neural network as the iterative update amount; ( f) Substitute the weights and offsets of the iterative update into this type of neural network for recursion; (g) Repeat steps (d) to (f) until the loss function approaches a specific value; and (h) After training, the output layer of this type of neural network is defined as the predicted temperature coefficient of resistance. The method for predicting the temperature coefficient of resistance of a battery intelligent shunt module according to claim 1, wherein the neural network of this type is created by an excitation function. The method for predicting the temperature coefficient of resistance of a battery intelligent shunt module as described in claim 1, wherein the loss function is

, And the specific value is 0; where,

Is the output value, and Y is the actual value. A current measurement compensation correction method for a battery intelligent shunt module, which includes the temperature coefficient of resistance predicted by the method for predicting the temperature coefficient of resistance of the battery intelligent shunt module according to any one of claims 1 to 3, which The steps include: (i) obtaining the output current from the predicted temperature coefficient of resistance according to step (a). A battery intelligent shunt module device, comprising: a processing unit, which is provided with the weights and biases of the resistance temperature coefficient prediction method of the battery intelligent shunt module as described in any one of claims 1 to 3 after training The value is shifted, and the output current is calculated according to the sensed current by the current measurement compensation correction method of the battery intelligent shunt module as described in claim 1. The battery intelligent shunt module according to claim 5 further includes a shunt, and the shunt is a copper alloy shunt.

Images