Battery SOH estimation model construction method, application and system

Abstract

The invention provides a battery SOH estimation model construction method, application and system. The improved BP neural network model construction method comprises the following steps: initializing a weight and a threshold value of a BP neural network by adopting a normal distribution initialization method; designing activation functions, dividing the activation functions into an activation function g1 (x) from an input layer to a hidden layer and an activation function g2 (x) from the hidden layer to an output layer, activating forward propagation, and obtaining expected values of output and loss functions of each layer; calculating an error term of an output layer and an error term of a hidden layer according to the loss function; and updating a weight and a bias item in the BP neural network. The improved BP neural network model is faster in operation speed, can realize global optimization, and is suitable for estimating the SOH of the battery, so that the SOH of the battery can be estimated directly by measuring the open-circuit voltage and the electrochemical impedance spectrum of the battery, the process is simple and convenient, multiple charging and discharging cycles of the battery are avoided, and energy and time are saved.

Description

Technical Field

[0001] The present invention relates to the technical field of battery detection, and in particular to a battery SOH estimation model construction method, application and system. Background Art

[0002] Batteries are widely used in the fields of electricity, national defense, transportation, communications, energy storage, etc. Batteries are in a floating charge state under normal conditions. When power is lost unexpectedly, the battery will provide emergency power supply for loads such as control circuits. However, due to the lack of effective maintenance and performance diagnosis methods, many battery packs are far from reaching their rated life in actual use, and often have insufficient power supply capacity, affecting the stable operation of power stations.

[0003] The main reason for the early failure of battery packs is the presence of degraded cells in the battery pack. Degraded cells in the battery pack will lead to inconsistencies among the cells, which will lead to the "barrel effect", and eventually the battery pack will not work properly due to the early failure of individual cells. In addition, the degraded battery will first affect the cells near it, and then gradually spread to eventually cause the entire battery pack to not work properly, seriously affecting the power supply safety of key equipment and emergency situations. Given the importance of maintaining the stability of the power station system, it is particularly important to quickly and accurately detect the health status of the battery and detect degraded batteries in a timely manner.

[0004] However, most current power grid systems obtain the actual maximum energy storage capacity of batteries through nuclear charge and discharge every two months, which is time-consuming and consumes a lot of resources. Therefore, it is necessary to find a more convenient and fast method to evaluate the health status (State of Health, SOH) of batteries, effectively improve the utilization efficiency of batteries, and thus maintain the stability of power station systems. Summary of the invention

[0005] The technical solution of the present invention is used to solve the problems of excessively long time and high resource consumption when detecting the health status of a battery in the prior art.

[0006] The present invention solves the above technical problems through the following technical means:

[0007] The battery SOH estimation model construction method uses the open circuit voltage value and the phase angle value of the characteristic frequency point as the input of the improved BP neural network model, and the health state estimation value of the battery is the model output, thereby obtaining the battery SOH estimation value; the improved BP neural network model construction method is as follows:

[0008] S1. Initialize the weights and thresholds of the BP neural network using the normal distribution initialization method;

[0009] S2. Design the activation function, which is divided into the activation function g1(x) from the input layer to the hidden layer and the activation function g2(x) from the hidden layer to the output layer. The formulas are as follows:

[0010]

[0011] Among them, the parameter Ψ is a positive number, which can be freely selected according to the input data value, as long as it is ensured that the input data |x| ≤ Ψ;

[0012] S3. Activate the forward propagation to obtain the expected values of the outputs of each layer and the loss function Among them, y represents the true value, represents the predicted value;

[0013] S4. Calculate the error terms of the output layer and the hidden layer according to the loss function;

[0014] S5. Update the weights and bias terms in the BP neural network, including the update of the output layer parameters and the update of the hidden layer parameters;

[0015] S6. Use the adaptive rate learning algorithm, that is, when the training error increases, the learning rate is decreased, and when the training error decreases, the learning rate is increased. The adaptive learning rate formula is as follows:

[0016]

[0017] Among them, ΔE is the change in the error function E, a > 1, 0 < b < 1, and k is an appropriate positive constant.

[0018] Furthermore, the specific step S4 is as follows: The error term of the output layer is to calculate the gradient value or partial derivative of the loss function with respect to the output layer. According to the chain rule, there is:

[0019]

[0020] The error term of the hidden layer is to calculate the gradient value or partial derivative of the loss function with respect to the hidden layer. According to the chain rule, there is:

[0021]

[0022] Furthermore, the update formula for the output layer parameters in the step S5 is:

[0023]

[0024] The update formula for the hidden layer parameters is:

[0025]

[0026] Among them, the parameter η represents the learning rate, k = 1, 2, …, n represents the number of updates or iterations. When k = 1, it represents the first update, and so on.

[0027] Furthermore, the learning rate η is usually represented by a fixed constant. The improved method for the learning rate η is as follows: Use the adaptive rate learning algorithm, that is, when the training error increases, the learning rate is decreased; when the training error decreases, the learning rate is increased. The improved adaptive learning rate formula is as follows:

[0028]

[0029] Among them, ΔE is the change in the error function E, a > 1, 0 < b < 1, and k is an appropriate positive constant, all of which can be adjusted according to the actual operation of the model. The improved adaptive learning rate can enable the model to move from a local optimum to a global optimum solution.

[0030] The present invention also provides a method for estimating the SOH of a battery, including the following steps:

[0031] S10. Measure the open-circuit voltage of the battery immediately after it is taken out of the floating charge state;

[0032] S20. Measure the impedance spectrum data of the battery; extract the phase angle data at a fixed frequency point from the obtained impedance spectrum data;

[0033] S30. Use the open-circuit voltage value and the phase angle value at the fixed frequency point as the inputs of the battery SOH estimation model constructed by using any one of the methods described in claims 1 to 4, and the estimated value of the health state of the battery is the output of the model, thereby obtaining the SOH estimated value of the battery.

[0034] Furthermore, the open-circuit voltage of the battery described in step S10 is the voltage difference between the positive and negative terminals of the battery.

[0035] Furthermore, the phase angle data at the fixed frequency point described in step S20 is obtained by processing and calculating the impedance spectrum data. The calculation formula for the phase angle data at the fixed frequency point is where Z' is the real part data extracted from the impedance spectrum data, and Z” is the imaginary part data extracted from the impedance spectrum data. The dimensions of both are Ω·cm 2 。

[0036] Furthermore, to select the best fixed frequency point, the Pearson correlation coefficient is used to calculate the correlation between the battery SOH value and the phase angle data at each frequency point respectively, and the frequency point with the highest correlation is selected as the fixed frequency point. The calculation formula for the Pearson correlation coefficient is as follows:

[0037]

[0038] where X and Y are two samples to be estimated.

[0039] The present invention also provides a battery SOH estimation system, comprising:

[0040] Open circuit voltage measurement module: measures the open circuit voltage of the battery just after it leaves the floating charge state;

[0041] Phase angle data measurement module: measures the impedance spectrum data of the battery; extracts the phase angle data of a fixed frequency point from the acquired impedance spectrum data;

[0042] Estimation module: using the open circuit voltage value and the phase angle value at a fixed frequency point as inputs of a battery SOH estimation model constructed by the method described in any one of claims 1 to 4, and the battery health status estimation value is the model output, thereby obtaining the battery SOH estimation value.

[0043] Furthermore, in order to select the best fixed frequency point, the Pearson correlation coefficient is used to calculate the correlation between the battery SOH value and the phase angle data at each frequency point, and the frequency point with the highest correlation is selected as the fixed frequency point. The calculation formula of the Pearson correlation coefficient is as follows:

[0044]

[0045] Among them, X and Y are two samples to be estimated.

[0046] The advantages of the present invention are:

[0047] The present invention collects the open circuit voltage data of the battery and extracts specific parameters from the impedance spectrum data, uses the obtained characteristic parameters to train an improved BP neural network model suitable for battery parameters, establishes a connection between the open circuit voltage, the phase angle value data at a fixed frequency point of the impedance spectrum and the battery health state, and the improved BP neural network model based on the battery-related data runs faster and can achieve global optimization, which is suitable for battery health state estimation, so that the battery SOH can be directly estimated by measuring the open circuit voltage and electrochemical impedance spectrum of the battery. The process is simple and convenient, avoids multiple charge and discharge cycles of the battery, and saves energy and time. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of a battery SOH estimation method according to Embodiment 1 of the present invention;

[0049] Figure 2 This is an open circuit voltage distribution diagram of the battery SOH estimation method according to the first embodiment of the present invention;

[0050] Figure 3 This is an algorithm flow chart of the BP neural network model used in the battery SOH estimation method of the first embodiment of the present invention;

[0051] Figure 4It is a phase angle-frequency curve diagram of the battery SOH estimation method according to the first embodiment of the present invention;

[0052] Figure 5 This is a training and testing flow chart of the improved BP neural network model used in the battery SOH estimation method of the first embodiment of the present invention;

[0053] Figure 6 This is a battery SOH estimation effect diagram of a training set of a battery SOH estimation method according to Embodiment 1 of the present invention;

[0054] Figure 7 This is a battery SOH estimation effect diagram of a test set of the battery SOH estimation method according to the first embodiment of the present invention. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] Embodiment 1

[0057] This embodiment describes an improved BP neural network model for battery SOH estimation.

[0058] The improved BP neural network optimizes the activation function and adaptive rate in the commonly used BP neural network model, speeds up the model's running iteration speed, and enables the model to move from local optimization to global optimization and obtain the global optimal solution.

[0059] Specifically, the construction method of the improved BP neural network model is as follows:

[0060] Step 1: Use normal distribution to initialize the weights and bias terms in the network, denoted as ω (0) , b1 (0) , v (0) , b2 (0) , normal distribution, that is, random initialization, samples from a Gaussian distribution with a mean of 0 and a standard deviation of 1, and uses some smaller values ​​to initialize the parameter ω to avoid overfitting problems from the beginning;

[0061] Step 2: Set the activation function of the BP network. The commonly used activation function is the sigmoid activation function. However, this patent redesigns the activation function based on the characteristic parameter values ​​obtained in steps 1 and 4, and divides the activation function into the activation function g1(x) from the input layer to the hidden layer and the activation function g2(x) from the hidden layer to the output layer. The formulas are:

[0062]

[0063] Among them, the parameter Ψ is a positive number and can be freely selected according to the input data value, and it only needs to ensure that the input data |x|≤Ψ. In this embodiment, the characteristic parameter values ​​of the input model need not be greater than 3, so Ψ=3 can be taken. The design of this activation function simplifies the function model, reduces the amount of calculation, makes the BP neural network model iteration faster, and can set the parameters of the activation function according to the input parameter value, which can improve the accuracy of the prediction.

[0064] So the formula from the input layer to the hidden layer is net1 = ω T x+b1, h=g1(net1), the formula from hidden layer to output layer is Where x is the data value of the input layer, h is the value of the input data transformed to the hidden layer, and the model function is

[0065] Step 3: Activate forward propagation to obtain the expected value of each layer output and loss function Among them, y represents the true value, Represents the predicted value.

[0066] Step 4: Calculate the error term of the output layer and the error term of the hidden layer based on the loss function. The error term of the output layer is to calculate the gradient value or partial derivative of the loss function with respect to the output layer. According to the chain rule:

[0067]

[0068] The error term of the hidden layer is to calculate the gradient value or partial derivative of the loss function with respect to the hidden layer. According to the chain rule, we have:

[0069]

[0070] Step 5: Update the weights and bias items in the neural network, including output layer parameter updates and hidden layer parameter updates. The output layer parameter update formula is:

[0071]

[0072] The hidden layer parameter update formula is:

[0073]

[0074] Among them, the parameter η represents the learning rate, k = 1, 2, …, n represents the number of updates or iterations. When k = 1, it represents the first update, and so on.

[0075] Step 6: The learning rate η is usually represented by a fixed constant. In this patent, the learning rate is improved by using an adaptive rate learning algorithm, that is, when the training error increases, the learning rate is decreased, and when the training error decreases, the learning rate is increased. The improved adaptive learning rate formula adopted in this patent is as follows:

[0076]

[0077] Among them, ΔE is the change in the error function E, a > 1, 0 < b < 1, and k is an appropriate positive constant. In this embodiment, a = 1.05, b = 0.7, and k = 0.2 are selected. The improved adaptive learning rate can enable the model to move from a local optimum to a global optimum solution.

[0078] S7: Repeat steps S2 to S6 until the value of the loss function converges to a smaller value or the preset number of iterations is reached.

[0079] Embodiment 2

[0080] As Figure 1 shown, the battery SOH estimation method according to the embodiment of the present invention includes the following steps:

[0081] Step 1: Measure the open-circuit voltage of the battery

[0082] Disconnect the battery from the floating charge state, and use a multimeter to measure the voltage difference between the positive and negative terminals of the battery, which is the open-circuit voltage of the battery. The open-circuit voltage is used as one of the characteristic parameters.

[0083] Step 2: Measure the electrochemical impedance spectrum of the battery

[0084] After the measurement of the open-circuit voltage is completed, measure the electrochemical impedance spectrum of the battery. The constant current mode is adopted, and the impedance spectrum measurement frequency is set to 0.01 Hz to 1000 Hz.

[0085] Step 3: Process the impedance spectrum data and extract the phase angle data at different frequency points

[0086] The phase angle data at a fixed frequency point is obtained by processing and calculating the impedance spectrum data. The calculation formula for the phase angle data at a fixed frequency point is where Z' is the real part data extracted from the impedance spectrum data, Z” is the imaginary part data extracted from the impedance spectrum data, and the dimension of both is Ω·cm 2 .

[0087] Step 4: Determine the characteristic frequency point and extract the phase angle data at the characteristic frequency point as another characteristic parameter

[0088] The Pearson correlation coefficient is used to calculate the correlation between the phase angle data and the battery SOH value at different frequency points, and the frequency point with the best correlation is selected as the characteristic frequency point. The calculation formula of the Pearson correlation coefficient is as follows:

[0089]

[0090] Among them, X and Y are two samples to be estimated.

[0091] Step 5: Using the open circuit voltage value and the phase angle value at the fixed frequency point as the model as the input of the improved BP neural network constructed in Example 1, the battery health state estimation value model outputs, thereby obtaining the battery SOH estimation value.

[0092] like Figure 1 As shown, the open circuit voltage of multiple batteries measured in advance, the phase angle data of the characteristic frequency points of the impedance spectrum and the SOH data are used to train the improved BP neural network model, so that it can achieve the optimal solution for parameter estimation. The trained improved BP neural network model is then used to estimate the SOH of the lead-acid battery. By inputting the characteristic parameters, the SOH estimation value of the battery can be directly obtained.

[0093] Test verification

[0094] After the battery is out of floating charge, measure the open circuit voltage of the battery and draw a scatter plot as shown below: Figure 2 As shown, the open circuit voltage and health status of the battery show a good linear relationship. After measuring the impedance spectrum data, a curve diagram of the relationship between frequency and phase angle is drawn as shown in Figure 4 As shown, the Pearson correlation coefficient of the phase angle value and the battery SOH value at different frequencies is calculated. The results show that the phase angle value at 59.6Hz has the highest correlation with SOH, so the open circuit voltage value and the phase angle value at 59.6Hz of the impedance spectrum are selected as characteristic parameters. The battery to be tested is divided into a 70% training set and a 30% test set. It should be noted that the training set is the part of the data sample used to train the improved BP neural network model, and the independent test set is the part of the data sample without the test set.

[0095] Input the characteristic parameters of the training set into the improved BP neural network model and train the improved BP neural network model. For the training set effect, refer to Figure 6 , the root mean square error between the predicted value and the true value RMSE = 0.0072524, then the characteristic parameters of the test set are input into the improved BP neural network model, and the model's battery SOH estimation effect on the test set is shown in Figure 7The root mean square error (RMSE) between the predicted value and the true value is 0.009731, which fully proves the accuracy of the prediction of the improved BP neural network model of the present invention.

[0096] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a battery SOH estimation model, using the open circuit voltage value and the phase angle value of the characteristic frequency point as the input of the improved BP neural network model, and the health status estimation value of the battery as the model output, thereby obtaining the SOH estimation value of the battery; characterized in that: The method for constructing an improved BP neural network model is as follows: S1. Initialize the weights and thresholds of the BP neural network using the normal distribution initialization method; S2. Design activation functions, which are divided into the activation function g1(x) from the input layer to the hidden layer and the activation function g2(x) from the hidden layer to the output layer. The formulas are as follows: Among them, the parameter Ψ is a positive number, which can be freely selected according to the input data value, as long as the input data |x| ≤ Ψ is ensured; S3. Activate forward propagation to obtain the expected value of each layer output and loss function Among them, y represents the true value, represents the predicted value; S4. Calculate the error terms of the output layer and the hidden layer according to the loss function; S5. Update the weights and bias terms in the BP neural network, including output layer parameter update and hidden layer parameter update; S6. Use the adaptive learning rate algorithm, that is, when the training error increases, the learning rate is decreased, and when the training error decreases, the learning rate is increased. The adaptive learning rate formula is as follows: Among them, ΔE is the change in the error function E, a > 1, 0 < b < 1, and k is an appropriate positive constant.

2. The method for constructing a battery SOH estimation model according to claim 1, characterized in that: The specific step S4 is: The error term of the output layer is to calculate the gradient value or partial derivative of the loss function with respect to the output layer. According to the chain rule, there is: The error term of the hidden layer is to calculate the gradient value or partial derivative of the loss function with respect to the hidden layer. According to the chain rule, there is:

3. The method for constructing a battery SOH estimation model according to claim 1, characterized in that: The output layer parameter update formula in the step S5 is: The hidden layer parameter update formula is: Among them, the parameter η represents the learning rate, k = 1, 2,..., n represents the update times or iteration times, k = 1 represents the first update, and so on.

4. The method for constructing a battery SOH estimation model according to claim 3, characterized in that: The learning rate η is usually represented by a fixed constant. The improvement method of the learning rate η is: Use the adaptive learning rate algorithm, that is, when the training error increases, the learning rate is decreased, and when the training error decreases, the learning rate is increased. The improved adaptive learning rate formula adopted is as follows: Among them, ΔE is the change in the error function E, a > 1, 0 < b < 1, and k is an appropriate positive constant, all of which can be adjusted according to the actual operation of the model. The improved adaptive learning rate can enable the model to move from local optimum to global optimum solution.

5. A battery SOH estimation method, characterized in that: It includes the following steps: S10. Measure the open-circuit voltage of the battery just after it is taken out of the floating charge state; S20. Measure the impedance spectrum data of the battery; Extract the phase angle data at fixed frequency points from the obtained impedance spectrum data; S30. Use the open-circuit voltage value and the phase angle value at the fixed frequency point as the input of the battery SOH estimation model constructed by any one of claims 1 to 4, and the estimated value of the battery health state is the output of the model, so as to obtain the SOH estimated value of the battery.

6. The battery SOH estimation method according to claim 5, characterized in that: The open-circuit voltage of the battery described in step S10 is the voltage difference between the positive and negative terminals of the battery.

7. The battery SOH estimation method according to claim 5, characterized in that: The phase angle data at the fixed frequency point in step S20 is obtained by processing and calculating the impedance spectrum data. The calculation formula for the phase angle data at the fixed frequency point is: Where Z' is the real part data extracted from the impedance spectrum data, and Z" is the imaginary part data extracted from the impedance spectrum data. Both dimensions are Ω·cm 2 .

8. The battery SOH estimation method according to claim 5, characterized in that: To select the best fixed frequency point, calculate the correlation between the battery SOH value and the phase angle data at each frequency point using the Pearson correlation coefficient, and select the frequency point with the highest correlation as the fixed frequency point. The calculation formula of the Pearson correlation coefficient is as follows: Among them, X and Y are two samples to be estimated.

9. A battery SOH estimation system, characterized in that: It includes: Open-circuit voltage measurement module: Measure the open-circuit voltage of the battery just after it is taken out of the floating charge state; Phase angle data measurement module: Measure the impedance spectrum data of the battery; Extract the phase angle data at fixed frequency points from the obtained impedance spectrum data; Estimation module: using the open circuit voltage value and the phase angle value at a fixed frequency point as inputs of a battery SOH estimation model constructed by the method described in any one of claims 1 to 4, and the battery health status estimation value is the model output, thereby obtaining the battery SOH estimation value.

10. The battery SOH estimation system according to claim 9, characterized in that: In order to select the best fixed frequency point, the Pearson correlation coefficient is used to calculate the correlation between the battery SOH value and the phase angle data at each frequency point, and the frequency point with the highest correlation is selected as the fixed frequency point. The calculation formula of the Pearson correlation coefficient is as follows: Among them, X and Y are two samples to be estimated.

Images