Estimating method and system of state of charge of battery

Abstract

The present disclosure provides an estimating method of state of charge of a battery including: obtaining a training set, the training set including a plurality of samples, each sample including a plurality of environmental factors, a plurality of vehicle parameters, a plurality of battery metrics and a plurality of heating circuit datasets; training an estimating model by the training set with using the environmental factors, the vehicle parameters, the battery metrics and the heating circuit datasets as input, optimizing the estimating model by optimal conditions and modifying the learning rate by a modifying condition; and installing the optimized estimating model to an electric vehicle, the estimating model generating an SOC estimation based on a plurality of real-time environmental factors, a plurality of real-time vehicle parameters, a plurality of real-time battery metrics and a plurality of real-time heating circuit datasets of the electric vehicle as moving. Therefore, the estimation accuracy of SOC is increased.

Description

Technical Field

[0001] This invention relates to a prediction method and system, and more particularly to a method and system for predicting the state of charge of a battery. Prior Technology

[0002] In recent years, in order to reduce air pollution caused by exhaust emissions from gasoline and diesel vehicles, zero-emission vehicles such as electric vehicles have developed rapidly and are gradually replacing gasoline and diesel vehicles. However, for electric vehicles, the design of the Battery Management System (BGM) is very important. It can monitor the battery voltage, temperature, and state of health (SOH), and predict and manage the state of charge (SOC).

[0003] Although there are many existing methods for predicting battery charging status using deep learning, the prediction models they establish are too simplistic and do not take into account the environment, vehicle driving parameters and the health status of the battery itself. They also place too much emphasis on the nonlinearity of the battery and time-varying parameters, which leads to prediction errors.

[0004] In view of this, improving the accuracy of battery charging state prediction methods and systems has become a goal pursued by relevant industry players. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides a method and system for predicting the state of charge of a battery. Through the configuration of the method and system, the accuracy of battery state of charge prediction can be improved.

[0006] According to an embodiment of the present invention, a method for predicting the state of charge of a battery is provided, comprising obtaining a training set containing complex samples, each sample containing complex environmental factors, complex vehicle parameters, complex battery indices, and complex heating information; training a prediction model using the training set, wherein the aforementioned complex environmental factors, complex vehicle parameters, complex battery indices, and complex heating information are used as inputs; and optimizing the prediction model. The optimization formula at a time step tn is as follows: ; ;and ; This represents a model parameter at time step tn. This represents a model parameter at time step t(n+1). This represents the mean square error at time step tn. Let represent an intermediate variable at time step t from time step 1 to time step tn. The learning rate at time step tn is expressed by the following formula: ; Let represent a learning rate at time step t(n-1). Let denot be a mean square error at time step tn-1, where tn is a positive integer, t is a variable and a positive integer between 1 and tn, and α1 and α2 are acceleration and deceleration coefficients, respectively; and let α1 and α2 be the acceleration and deceleration coefficients, respectively; and let α1 and α2 be the acceleration and deceleration coefficients, respectively; and let α1 and α2 be the acceleration and deceleration coefficients, respectively; and let α1 and α2 be the acceleration and deceleration coefficients, respectively; and let α2 ...

[0007] According to the battery charging state prediction method of the aforementioned embodiments, the prediction model may include three hidden layers.

[0008] According to the battery charging state prediction method of the aforementioned embodiments, the three hidden layers may contain 64, 128 and 64 neurons respectively.

[0009] According to the battery charging state prediction method of the aforementioned embodiments, each hidden layer may use a ReLU excitation function.

[0010] According to the battery charging state prediction method of the aforementioned embodiments, the initial learning rate of the prediction model can be set to 0.001.

[0011] According to the battery charging state prediction method of the aforementioned embodiments, the aforementioned complex environmental factors, the aforementioned complex vehicle parameters, the aforementioned complex battery indicators and the aforementioned complex heating information can be obtained from real vehicle driving records.

[0012] According to another embodiment of the present invention, a battery state of charge (SOC) prediction system is provided, which applies the battery SOC prediction method of the aforementioned embodiments, and includes an environment detection device, a vehicle parameter acquisition device, a battery index acquisition device, a heating information acquisition device, and a processor. The environment detection device is used to detect the environment in which the electric vehicle is driving, to obtain the aforementioned multiple real-time environmental factors. The vehicle parameter acquisition device is used to obtain the aforementioned multiple real-time vehicle parameters of the electric vehicle while it is driving. The battery index acquisition device is used to obtain the aforementioned multiple real-time battery indexes of a battery of the electric vehicle while it is driving. The heating information acquisition device is used to obtain the aforementioned multiple real-time heating information of a heating device of the electric vehicle while it is driving. The processor is signal-connected to the environment detection device, the vehicle parameter acquisition device, the battery index acquisition device, and the heating information acquisition device, and has a prediction model installed on it. The prediction model takes into account the aforementioned multiple real-time environmental factors, the aforementioned multiple real-time vehicle parameters, the aforementioned multiple real-time battery indexes, and the aforementioned multiple real-time heating information, and outputs a SOC prediction value.

[0013] The battery state of charge prediction system according to the aforementioned embodiments may include a prediction model containing three hidden layers.

[0014] According to the battery charging state prediction system of the aforementioned embodiments, the three hidden layers may contain 64, 128 and 64 neurons respectively.

[0015] According to the battery state of charge prediction system of the aforementioned embodiments, each hidden layer may use a ReLU excitation function. Simple Explanation of the Diagram

[0016] Figure 1 illustrates a block flowchart of a battery charging state estimation method according to an embodiment of the present invention; Figure 2 illustrates a schematic diagram of the architecture of a battery state of charge prediction system according to another embodiment of the present invention; Figure 3 illustrates a comparison between the predicted and actual SOC values ​​of an experimental example of the present invention; and Figure 4 illustrates the error between the estimated SOC value and the actual SOC value of the experimental example of the present invention. Implementation

[0017] The embodiments of the present invention will now be described with reference to the drawings. For clarity, many practical details will be described in the following description. However, the reader should understand that these practical details are not intended to limit the invention. That is, in some embodiments of the invention, these practical details are not essential. Furthermore, for the sake of simplicity, some conventional structures and elements will be shown in the drawings in a simple schematic manner; and repeated elements may be represented by the same or similar designations.

[0018] Furthermore, the terms "first," "second," and "third" used in this article are merely used to describe different components or parts, and do not impose any restrictions on the components / parts themselves. Therefore, the first component / part can also be referred to as the second component / part. Moreover, the combinations of components / parts / mechanisms / modules in this article are not combinations generally known, conventional, or conventional in this field. Whether the components / parts / mechanisms / modules themselves are conventional cannot be used to determine whether their combination relationships are easily accomplished by those with ordinary knowledge in the technical field.

[0019] Please refer to Figure 1, which illustrates a block flowchart of a battery state of charge estimation method 100 according to an embodiment of the present invention. The battery state of charge estimation method 100 includes steps S110, S120, and S130.

[0020] Step S110 is used to obtain a training set, which contains a complex number of samples, each of which contains a complex number of environmental factors, complex number of vehicle parameters, complex number of battery indicators and complex number of heating information.

[0021] Step S120 trains a prediction model with a training set, where environmental factors, vehicle parameters, battery indicators and heating information are used as inputs, and optimizes the prediction model. The optimization formulas at the nth time step are shown in equations (1) to (3).

[0022] (1).

[0023] (2).

[0024] (3).

[0025] This represents the model parameters at time step tn. This represents the model parameters at time step t(n+1). This represents the mean square error at time step tn. Let represent an intermediate variable at time step t from time step 1 to time step tn. Let tn represent the learning rate at time step tn, and its adjustment formula is shown in equation (4).

[0026] (4).

[0027] Let represent the learning rate at time step t(n-1). Let denot represent the mean square error at time step tn-1, where tn is a positive integer, t is a variable and a positive integer between 1 and tn, α1 and α2 are the acceleration and deceleration coefficients, respectively, and α1 is the coefficient used for dynamic control of the learning rate increase and decrease.

[0028] Step S130 involves installing the optimized prediction model onto an electric vehicle. The prediction model generates a SOC estimate based on multiple real-time environmental factors, multiple real-time vehicle parameters, multiple real-time battery indicators, and multiple real-time heating information during the electric vehicle's operation.

[0029] Therefore, the prediction model can receive environmental factors, vehicle parameters, battery specifications, and heating information as inputs during training. It adjusts the learning rate based on the mean squared error of the previous and current time steps to update the model parameters for the next time step, effectively optimizing the prediction model. Thus, after being installed in an electric vehicle, the prediction model can output a SOC (State of Charge) estimate based on real-time environmental factors, real-time vehicle parameters, real-time battery specifications, and real-time heating information, thereby improving the accuracy of the prediction.

[0030] In step S110, the environmental factors, vehicle parameters, battery parameters, and heating information in the training set can be obtained from the actual driving records of the electric vehicle. Environmental factors include temperature and altitude; vehicle parameters include speed and throttle; battery parameters include voltage, current, and SOC; and heating information includes interior temperature and heating power. This data can be obtained, for example, from environmental detection devices, vehicle parameter acquisition devices, battery parameter acquisition devices, and heating information acquisition devices on the electric vehicle.

[0031] The prediction model in step S120 can be, for example, a model of a recurrent neural network (RNN), and the formulas are as shown in equations (5) to (6).

[0032] (5).

[0033] (6).

[0034] ht represents a hidden state of the prediction model at time step t, ht-1 represents a hidden state of the prediction model at time step (t-1), xt represents an input value of the prediction model at time step t, yt represents a SOC prediction value of the prediction model at time step t, Whh, Wzh, and Why are all weight matrices, bh and by are bias terms, f and Let be the activation function. Therefore, the input value at time step t is related to the input value at the previous time step (i.e., time step t-1).

[0035] In this embodiment, the prediction model may include three hidden layers, which may contain 64, 128, and 64 neurons respectively, and each hidden layer may use a ReLU activation function, i.e., f and All of them can be ReLU activation functions, but are not limited to this.

[0036] During training, the initial learning rate of the prediction model can be set to 0.001, and the normalization strength can be set to 0.1, but this is not a limitation. The loss function (objective function) of the prediction model can be as shown in equation (7).

[0037] (7).

[0038] N is the number of time steps, which is a positive integer. Therefore, the prediction model can execute from time step 1 to time step N. E is the cumulative mean square error, which is the cumulative mean square error from time step 1 to time step N. ŷ t is the actual SOC value. When executing to time step tn, the mean square error can be used as a reference. To adjust the model parameters at time step t(n+1). As shown in equations (1) to (3), tn can be any value from 1 to N, that is, the model parameters for the next time step are adjusted according to the mean square error of the current time step, and the effect of time delay is taken into account. It should be noted that the model parameters here may include the weight matrices Whh, Wzh and Why in equations (5) and (6).

[0039] During training, a training epoch can be set, model parameters can be adjusted, and the loss function (mean squared error in this invention) can be minimized to optimize the prediction model. During training, the learning rate will affect the convergence result of the prediction model, so the adjustment of the learning rate is very important. In this embodiment, the learning rate is adjusted using equation (4). If the derivative signs of specific weights are the same, the learning rate can be appropriately increased to reduce the number of repetitions. If the prediction model converges to the concave part of the error surface, the derivative signs of specific weights may change in continuous operation, and the learning rate needs to be appropriately reduced to avoid weight oscillation.

[0040] The training set can be input into a computer, and the prediction model can also be built and trained by the computer. Once the prediction model is trained, it can be installed on the electric vehicle, specifically on the processor of the electric vehicle. Therefore, step S130 allows the prediction model to be installed on the processor of the electric vehicle. In this way, real-time environmental factors, real-time vehicle parameters, real-time battery indicators, and real-time heating information obtained when the electric vehicle is in motion can be input into the prediction model to generate a SOC prediction value.

[0041] Please refer to Figure 2, which illustrates a schematic diagram of the architecture of a battery state of charge prediction system 200 according to another embodiment of the present invention. The battery state of charge prediction system 200 may include an environmental detection device 210, a vehicle parameter acquisition device 220, a battery index acquisition device 230, a heating information acquisition device 240, and a processor 250.

[0042] An environmental detection device 210 is used to detect the environment in which an electric vehicle is driving, to obtain multiple real-time environmental factors. A vehicle parameter acquisition device 220 is used to obtain multiple real-time vehicle parameters of the electric vehicle while it is driving. A battery indicator acquisition device 230 is used to obtain multiple real-time battery indicators of the electric vehicle's battery while it is driving. A heating information acquisition device 240 is used to obtain multiple real-time heating information of the electric vehicle's heating device while it is driving. A processor 250 is signal-connected to the environmental detection device 210, the vehicle parameter acquisition device 220, the battery indicator acquisition device 230, and the heating information acquisition device 240, and includes a prediction model 251. The prediction model 251 takes the aforementioned multiple real-time environmental factors, the aforementioned multiple real-time vehicle parameters, the aforementioned multiple real-time battery indicators, and the aforementioned multiple real-time heating information as input, and outputs a SOC (State of Charge) prediction value. The prediction model 251 can be optimized using the battery state of charge prediction method 100 and installed in the electric vehicle.

[0043] The environmental detection device 210 may include, for example, a camera, radar, lidar, positioning elements, and an external temperature sensor. The positioning elements may be, for example, a Global Positioning System (GPS). The vehicle parameter acquisition device 220 may include, for example, a power drive module and a braking module. The power drive module may include an engine, transmission, driveshaft, and differential. The braking module may include brake discs and brake pads. The battery indicator acquisition device 230 may include voltage and current measuring devices on the battery. The heating information acquisition device 240 may include an internal temperature sensor. However, it is not limited to the above. Any device / equipment that can obtain real-time environmental factors, real-time vehicle parameters, real-time battery indicators, and real-time heating information is acceptable. The processor 250 may include, for example, a central processing unit (CPU), a microprocessor, a digital signal processor (DSP), a programmable controller, a programmable logic device (PLD), or other similar devices or combinations thereof.

[0044] In this way, real-time data at the moment of driving can be obtained through the environmental detection device 210, vehicle parameter acquisition device 220, battery index acquisition device 230 and heating information acquisition device 240, and can be input into the prediction model 251, so as to obtain the real-time SOC prediction value.

[0045] <Experimental Example>

[0046] The experimental examples of this invention use 72 real-world driving records of a BMW i3 electric vehicle (60 Ah battery capacity) in summer and winter as the training set, and a portion of these records as the test set. The SOC values ​​in the test set are the actual SOC values. The real-world driving records can include environmental factors, vehicle parameters, battery specifications, and heating information, totaling 24 features such as altitude, motor torque, longitudinal acceleration, braking signal, battery voltage, battery current, battery temperature, and battery charge. All 24 features can be used as inputs, and in this invention, minimum redundancy maximum correlation (mRMR) can be used for feature selection to find the features with the highest correlation.

[0047] During training, the training set can be split into 5 parts for cross-validation, with an initial learning rate of 0.001, a training cycle of more than 1000, a batch size of 32, and the loss function minimized to optimize the prediction model.

[0048] Please refer to Figures 3 and 4. Figure 3 shows a comparison between the predicted and actual SOC values ​​of the experimental examples of this invention, and Figure 4 shows the error between the predicted and actual SOC values ​​of the experimental examples of this invention. As can be seen from Figures 3 and 4, the predicted SOC value output by the prediction model of this invention is close to the actual SOC value, which can accurately predict the battery charging state with a very high accuracy.

[0049] Table 1 is a comparison table of the performance metrics of the experimental example and a comparative example of the present invention. The comparative example uses robust regression to predict the state of charge of the battery and compares the results using different performance metrics, including root mean square error (RMSE), R-squared (or coefficient of determination), mean squared error (MSE), and mean absolute error (MAE). Table 1 Experimental Example Comparative example Summer data Winter data Summer data Winter data Training results RMSE 0.8362 1.9254 2.0241 5.6452 R-squared 1.00 0.99 0.98 0.92 MSE 0.6992 3.707 4.0968 31.868 MAE 0.6172 1.4218 1.5161 3.8883 Test Results RMSE 0.8425 2.355 2.0147 5.6396 R-squared 1.00 0.99 0.98 0.92 MSE 0.7098 4.9973 4.0589 31.805 MAE 0.6267 1.6709 1.5144 3.8847

[0050] As shown in Table 1, the experimental examples of this invention exhibit high accuracy, with the difference being more pronounced in winter data. This further demonstrates that the battery state-of-charge (SOC) prediction method and system of this invention can improve the accuracy of SOC prediction. Furthermore, the experimental examples of this invention identified temperature, voltage, current, and route complexity as the most correlated features with SOC, while the effects of speed and altitude were relatively mild. In contrast, the comparative examples showed a higher correlation with voltage, a milder influence from temperature and current, and lower influences from speed, altitude, and route complexity. Therefore, the prediction model of this invention can effectively improve prediction accuracy when using multiple types of information as input.

[0051] The invention has been disclosed above with reference to embodiments, but it is not intended to limit the invention. Anyone skilled in the art can make various modifications and refinements without departing from the spirit and scope of the invention. Therefore, the scope of protection of the invention shall be determined by the appended claims.

[0052] 100: Battery state of charge prediction method 200: Battery State of Charge Prediction System 210: Environmental Detection Device 220: Vehicle parameter acquisition device 230: Battery performance measurement device 240: Heating Information Acquisition Device 250: Processor 251: Prediction Model S110, S120, S130: Steps

Claims

1. A method for predicting the state of charge of a battery, comprising: obtaining a training set containing a plurality of samples, each of the samples containing a plurality of environmental factors, a plurality of vehicle parameters, a plurality of battery indices, and a plurality of heating information; training a prediction model using the training set, wherein the environmental factors, the vehicle parameters, the battery indices, and the heating information are used as inputs, and optimizing the prediction model, wherein the optimization formula at a time step tn is as follows: ; ; and ; represent a model parameter at the time step tn, represent a model parameter at the time step tn+1, represent a mean square error at the time step tn, represent an intermediate variable at time step t from time step 1 to time step tn, and represent a learning rate at the time step tn, the adjustment formula of which is as follows: ; represent a learning rate at time step tn-1, represent the learning rate at time step tn-1, represent the learning rate at time step tn+ ... The mean square error at time step n-1, where tn is a positive integer, t is a variable and a positive integer between 1 and tn, α1 and α2 are acceleration and deceleration coefficients, respectively; and the optimized prediction model is installed on an electric vehicle, which generates a SOC prediction based on complex real-time environmental factors, complex real-time vehicle parameters, complex real-time battery indicators and complex real-time heating information during the operation of the electric vehicle.

2. The battery state of charge prediction method as described in claim 1, wherein the prediction model comprises three hidden layers.

3. The battery state of charge prediction method as described in claim 2, wherein the three hidden layers contain 64, 128 and 64 neurons, respectively.

4. The battery state of charge prediction method as described in claim 3, wherein each of the three hidden layers uses a ReLU activation function.

5. The battery state of charge prediction method as described in claim 4, wherein an initial learning rate of the prediction model is set to 0.

001.

6. The battery state of charge prediction method as described in claim 1, wherein the environmental factors, vehicle parameters, battery specifications and heating information are derived from actual vehicle driving records.

7. A battery state of charge (SOC) prediction system, employing the battery SOC prediction method as described in claim 1, and comprising: an environment detection device for detecting an environment in which the electric vehicle is driven, and obtaining real-time environmental factors of the environment; a vehicle parameter acquisition device for obtaining real-time vehicle parameters of the electric vehicle while driving; a battery index acquisition device for obtaining real-time battery indexes of a battery of the electric vehicle while driving; a heating information acquisition device for obtaining real-time heating information of a heating device of the electric vehicle while driving; and a processor, signal-connected to the environment detection device, the vehicle parameter acquisition device, the battery index acquisition device, and the heating information acquisition device, and having the prediction model installed thereon, the prediction model taking into input the real-time environmental factors, the real-time vehicle parameters, the real-time battery indexes, and the real-time heating information, and outputting the SOC prediction value.

8. The battery state of charge prediction system as described in claim 7, wherein the prediction model comprises three hidden layers.

9. The battery state of charge prediction system as described in claim 8, wherein the three hidden layers contain 64, 128 and 64 neurons, respectively.

10. The battery state of charge prediction system as described in claim 9, wherein each of the three hidden layers uses a ReLU activation function.

Images