High-speed lithium battery energy state estimation method

Abstract

The invention discloses a method for estimating the energy state of a high-speed lithium battery, which belongs to the field of electric automobiles and comprises the following steps: performing multi-time scale modeling and hierarchical feature extraction on current, voltage and power data of the lithium battery through a deep echo state network, independently adjusting hyper-parameters of each layer by using a parameter differentiation strategy, and calculating the energy state of the lithium battery according to the hyper-parameters; therefore, different dynamic characteristic inputs can be adapted. Furthermore, sudden change output signals in the prediction process are removed through a modal decomposition noise reduction module, and the prediction precision and stability are improved. Experimental verification shows that the method shows high-speed prediction performance, high precision and high usability under various real automobile working conditions, not only has the capabilities of rapid training and efficient prediction, but also can effectively avoid the problem of gradient disappearance in a traditional recurrent neural network, and meanwhile, keeps higher precision and robustness.

Description

Technical Field

[0001] The present invention belongs to the field of electric vehicles, and particularly relates to a method for estimating the energy state of a high-speed lithium battery. Background Art

[0002] In an electric vehicle battery management system, the state of energy (SOE) of the battery is an important parameter for measuring the remaining available energy of the battery, which directly affects the endurance ability and energy management strategy of the electric vehicle.

[0003] However, due to the significant non-linearity and time-varying characteristics of the key measurable external data of lithium batteries, how to efficiently and accurately estimate the SOE of the battery has become the main difficulty and research focus in the current technological development. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides a method for estimating the energy state of a high-speed lithium battery, including:

[0005] Obtain the input data of the lithium battery, where the input data includes current, voltage, and power;

[0006] Obtain the original data set based on the general driving condition experiment;

[0007] Construct a deep echo state network, and train the deep echo state network based on the original data set to obtain a lithium battery SOE estimation model;

[0008] Input the input data into the lithium battery SOE estimation model for calculation to obtain an SOE prediction value with output noise;

[0009] Construct a modal decomposition noise reduction module, and remove the sudden output signal in the SOE prediction value with output noise based on the modal decomposition noise reduction module to obtain a smooth lithium battery energy state prediction value.

[0010] Preferably, the deep echo state network includes multiple stacked reservoir layers, and the input data is modeled on multiple time scales and hierarchically feature-extracted through the reservoir layers;

[0011] The deep echo state network adopts a parameter differentiation strategy to independently adjust the hyperparameters of different reservoir layers, and each adjusted reservoir layer processes input data with different dynamic characteristics.

[0012] Preferably, the steps for updating the reservoir state of the deep echo state network include:

[0013] Update the state of the first-layer reservoir according to the input data and the initialized weight matrix, combine the external input with the bias, and obtain the first-layer state through non-linear transformation;

[0014] Using the output of the first-layer reservoir as input, update the states of the second-layer and above reservoirs, and transmit information layer by layer to form a deep dynamics.

[0015] Preferably, the output weight training steps of the deep echo state network include:

[0016] Generate a joint state matrix according to the updated reservoir state;

[0017] Remove the transient data in the joint state matrix and retain the steady-state data for training;

[0018] Use the ridge regression method to calculate the output weights and obtain the SOE prediction value with output noise.

[0019] Preferably, the process of removing the mutant output signal in the SOE prediction value with output noise based on the noise reduction module includes:

[0020] Input the SOE prediction value with output noise into the modal decomposition noise reduction module;

[0021] Use the successive variational mode decomposition technique to decompose the prediction value into multiple intrinsic mode functions and a residual signal.

[0022] Preferably, the noise reduction steps of the modal decomposition noise reduction module include:

[0023] Perform Pearson correlation analysis on each decomposed intrinsic mode function and calculate its correlation coefficient with the original prediction value;

[0024] Screen out the intrinsic mode functions with high correlation according to a preset correlation threshold and remove the noise signals with low correlation.

[0025] Preferably, the reconstruction steps of the modal decomposition noise reduction module include: performing phase addition and reconstruction on the screened intrinsic mode functions with high correlation to obtain the final smooth and accurate lithium battery energy state prediction value.

[0026] Preferably, the optimization process of the successive variational mode decomposition technique includes:

[0027] Introduce the augmented Lagrangian function to transform the modal decomposition problem into a constrained optimization problem;

[0028] Use the alternating direction method of multipliers for iterative optimization to update the mode function, central frequency, and Lagrange multiplier.

[0029] On the other hand, the present invention also provides an electronic device, including a memory, a processor, and a computing program stored in the memory and executable on the processor. When the processor executes the computing program, the method is implemented.

[0030] On the other hand, the present invention also provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the method is implemented.

[0031] Compared with the prior art, the present invention has the following advantages and technical effects:

[0032] (1) First, a multi-layer stacked reservoir structure is designed to enhance the non-linear expression ability, and a parameter differentiation strategy is adopted to explicitly control the dynamic characteristics of each layer. A deep echo state network for estimating the state of charge (SOC) of a lithium-ion battery is constructed to achieve rapid training and testing of the model.

[0033] (2) A modal decomposition noise reduction module is introduced into the model. The successive variational mode decomposition (SVMD) technology is adopted to decompose the network output into multiple intrinsic mode functions (IMFs). After performing Pearson correlation analysis on the IMFs, threshold screening is used to remove the noise signals, thereby weakening the influence of the output noise on the SOC estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The drawings forming a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application. In the drawings:

[0035] Figure 1 It is a schematic flow chart of the method according to the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0036] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The following will refer to the drawings and combine the embodiments to detail this application.

[0037] It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0038] Embodiment 1

[0039] As Figure 1 shown, a high-speed method for estimating the state of energy of a lithium battery is provided in this embodiment, including:

[0040] Obtain the input data of the lithium battery, where the input data includes current, voltage, and power;

[0041] Obtain the original data set based on the general driving condition experiment;

[0042] Construct a deep echo state network, train the deep echo state network based on the original dataset, and obtain a lithium battery state of energy (SOE) estimation model;

[0043] Input the input data into the lithium battery SOE estimation model for calculation to obtain an SOE prediction value with output noise;

[0044] Construct a modal decomposition noise reduction module, and based on the modal decomposition noise reduction module, remove the mutation output signal in the SOE prediction value with output noise to obtain a smooth and accurate lithium battery state of energy prediction value.

[0045] The concept of SOE was proposed by Keil Mamadou at the 216th ECS meeting in 2009. SOE is a complementary concept to the state of charge, representing the actual energy stored in the battery, that is, the energy percentage of the battery relative to its rated energy. SOE is usually used to more accurately represent the energy storage state of the battery, taking into account the rated energy of the battery. SOE changes with the charge and discharge process of the battery, and its range usually also ranges from 0% (fully discharged) to 100% (fully charged), and its expression is shown in formula (1).

[0046]

[0047] Where is the rated energy of the battery, is the remaining energy of the battery.

[0048] Furthermore, the deep echo state network includes multiple stacked reservoir layers, and the input data is subjected to multi-time scale modeling and hierarchical feature extraction through the reservoir layers;

[0049] The deep echo state network adopts a parameter differentiation strategy to independently adjust the hyperparameters of different reservoir layers, and each adjusted reservoir processes input data with different dynamic characteristics.

[0050] The flowchart of the method for estimating the SOE of a lithium battery proposed in this embodiment is as Figure 1 shown. The core of this method is the deep echo state network (DeepESN) and the output noise reduction strategy. The core of DeepESN is multiple stacked reservoirs. First, the lower-layer reservoir maps the original data to convert signals such as current and voltage into a series of local dynamic features. Subsequently, the upper-layer reservoir combines the information from the feature input reservoir to further capture the dynamic changes in signals such as current and voltage and complete the state update. The random fixed weight and linear output layer training mechanism of the reservoir in DeepESN enable DeepESN to not only have the capabilities of fast training and efficient prediction when performing SOE prediction, but also effectively avoid the gradient vanishing problem in traditional recurrent neural networks, while maintaining high accuracy and robustness.

[0051] In the construction of DeepESN, first, the input weight matrix and the inter-layer weight matrix are initialized as shown in Equation 2. The role of the input weight matrix is to initialize the weights from the input to the first layer of the reservoir, controlling the influence intensity of the input signal. The role of the inter-layer weight matrix is to initialize the weights for the layer to receive the output of the previous layer, controlling the intensity of information transfer between layers.

[0052]

[0053] where represents the number of neurons in each layer of the reservoir, represents the input dimension, represents the input scaling factor, inter-layer connection scaling factor, represents the norm operation, which is used to normalize the norm of the random matrix to a unit value, ensuring good numerical stability and dynamic behavior in the initialization of the weight matrix.

[0054] Subsequently, the initialization of the recurrent weight matrix is carried out as shown in Equations (3) and (4). The recurrent weight matrix balances the historical state and the current input by introducing a leakage rate to ensure the stability of the recurrent dynamics.

[0055]

[0056]

[0057] where is the original random matrix of the layer recurrent weight matrix, is the preset spectral radius, is the matrix spectral radius calculation, is the scaling matrix after adjusting the spectral radius, is the leakage rate, is the identity matrix.

[0058] Furthermore, the steps for updating the state of the reservoir of the deep echo state network include:

[0059] According to the input data and the initialized weight matrix, update the state of the first layer of the reservoir, combine the external input with the bias, and obtain the first layer state through a non-linear transformation;

[0060] Using the output of the first - layer reservoir as input, update the states of the second - layer and above reservoirs, transmit information layer by layer, and form deep dynamics.

[0061] The update of the reservoir state is the core step of reservoir computing. Its significance lies in using the dynamic characteristics and high - dimensional feature space of the reservoir to capture the temporal dependence, non - linear characteristics, and dynamic patterns of the input signal. Since this embodiment uses a parameter differentiation strategy to control the dynamic characteristics of each layer, the state update formulas of each layer are slightly different. The state update formula of the first layer combines the external input with the bias and updates the state of the first layer through non - linear transformation, as shown in formula (5).

[0062]

[0063] Where represents the state vector at time step \(t = 1\), which is used to characterize the dynamic memory of the network at time step \(t = 1\). is the external input. is the input bias. is the activation function. The activation function transforms the linear input into a non - linear mapping, enhancing the representation ability of the reservoir for complex input patterns. is the recurrent weight matrix. represents the state vector of the reservoir at the previous time step \(t-1\).

[0064] The state update of the \(l\) - th layer forms deep dynamics by using the output of the previous layer as input and transmitting information layer by layer, as shown in formula (6).

[0065]

[0066] Where represents the state vector at time step \(t\). represents the output state of the \(l\) - th layer at time step \(t\), which is also the input signal of the \((l + 1)\) - th layer. represents the state vector of the \(l\) - th layer at time step \(t\). represents the \((l-1)\) - th layer at time step \(t\).

[0067] Furthermore, the output weight training step of the deep echo state network includes:

[0068] Generate a joint state matrix according to the updated reservoir state;

[0069] Remove the transient data in the joint state matrix and retain the steady-state data for training;

[0070] Use the ridge regression method to calculate the output weights and obtain the SOE prediction value with output noise.

[0071] As shown in formula (7), a joint state matrix will be generated after the state update , which is used for linear regression of the output layer, where is the number of layers of the deep echo state.

[0072]

[0073] represents the state value of the deepest layer (the Nth layer) in the reservoir network at the last moment T.

[0074] After generating the joint state matrix, transients need to be removed. As shown in formula (8), removing transients is a key step in the DeepESN training process, aiming to discard the unstable states during the initial operation of the network and ensure that only effective steady-state data is used for subsequent training.

[0075]

[0076] where is the joint state matrix, is the target output matrix, is the transient duration, is the total number of time steps, is the training input matrix after removing transients, is the training target matrix after removing transients.

[0077] Finally, the output weight training of the reservoir and joint prediction are performed. As shown in formula (9), the deep echo state network uses ridge regression to calculate the output weights , where is the regularization coefficient, which prevents overfitting by minimizing the output error. Subsequently, the concatenated multi-layer states are combined with the bias, and the predicted output is generated through linear transformation , as shown in formula (10).

[0078]

[0079]

[0080] Furthermore, the process of removing the mutant output signal from the SOE prediction value with output noise based on the noise reduction module includes:

[0081] Input the SOE prediction value with output noise into the modal decomposition noise reduction module;

[0082] Use the successive variational mode decomposition technique to decompose the prediction value into multiple intrinsic mode functions and a residual signal.

[0083] Furthermore, the noise reduction steps of the modal decomposition noise reduction module include:

[0084] Perform Pearson correlation analysis on each decomposed intrinsic mode function and calculate its correlation coefficient with the original prediction value;

[0085] Filter out the intrinsic mode functions with high correlation according to a preset correlation threshold and remove the noise signals with low correlation.

[0086] Furthermore, the reconstruction step of the modal decomposition noise reduction module includes: performing phase addition and reconstruction on the selected intrinsic mode functions with high correlation to obtain the final smooth lithium battery energy state prediction value.

[0087] Furthermore, the optimization process of the successive variational mode decomposition technique includes:

[0088] Introduce the augmented Lagrangian function to transform the modal decomposition problem into a constrained optimization problem;

[0089] Use the alternating direction multiplier method for iterative optimization to update the mode function, central frequency, and Lagrange multiplier.

[0090] After obtaining the SOE prediction value with output noise generated by the deep recurrent state network, in order to obtain a smooth and stable prediction result, a noise reduction module is designed to remove the output noise. The core of the noise reduction module is the successive variational mode decomposition (SVMD). As shown in formulas (11) and (12), SVMD can decompose the original signal into the current mode and the residual signal , where the residual signal contains the set of extracted modes and the unprocessed residual .

[0091]

[0092]

[0093] The goal of SVMD is to minimize the combination of three criteria while satisfying the signal constraint conditions, as shown in formulas (13) to (15).

[0094]

[0095]

[0096]

[0097] where is the spectral compactness, is the separation between the residual signal and the modal spectrum, is the inter-modal spectral separation. In Equation (13), is the time derivative operator, is the unit impulse function, is the time-domain signal, is the central frequency of the current mode. In Equation (14), is the filter for the residual signal, which realizes spectral filtering through the time-domain equivalent operation of frequency-domain multiplication. In Equation (15), is the inter-modal separation filter, which forces the spectrum of the current mode to be far from the central frequencies of all the extracted modes through multiple groups of filters to avoid modal aliasing.

[0098] To solve this optimization problem, the augmented Lagrangian function is introduced. The augmented Lagrangian function transforms the original optimization problem into a constrained optimization problem, and at the same time ensures the exact signal reconstruction through the penalty term and the Lagrange multiplier, as shown in Equation (16).

[0099]

[0100] where is the regularization parameter, is the Lagrange multiplier, which ensures the integrity of signal decomposition.

[0101] SVMD is optimized using the alternating direction method of multipliers (ADMM), and the following variables are iteratively updated. First, each mode is updated in the frequency domain. Each mode is updated in the form of a frequency-domain filter, and the influence of other modes is excluded during the calculation, as shown in Equation (17).

[0102]

[0103] where is the frequency-domain updated value of the current modal component, is the frequency-domain representation of the original signal, is the current central frequency, is the current iterative mode, is the frequency-domain representation of the Lagrange multiplier.

[0104] Secondly, update the central frequency of each mode as shown in Equation (18), where is the updated central frequency.

[0105]

[0106] Then, use the dual ascent method to update the Lagrange multipliers to ensure that the signal reconstruction constraints are gradually satisfied during the optimization process, as shown in Equation (19).

[0107]

[0108] where is the dual ascent step size used to control the convergence rate.

[0109] Finally, check whether convergence is achieved by examining the mode convergence and residual energy conditions, as shown in Equations (20) and (21). If convergence is not achieved, continue the iterative update using the alternating direction method of multipliers; otherwise, output the decomposition result.

[0110]

[0111]

[0112] where is the mode convergence threshold used to control the relative error of mode updates. is the residual convergence threshold that controls the deviation of the residual energy from a preset threshold, represents the estimated value of the \(L\)-th mode after the \((n + 1)\)-th iteration, represents the estimated value of the \(L\)-th mode after the \(n\)-th iteration, is the preset residual energy threshold.

[0113] When SVMD converges, output multiple intrinsic mode functions (IMFs), and then calculate the Pearson correlation coefficient according to Equation (22) to measure the linear correlation degree between the prediction results of the DeepESN network and the IMFs.

[0114]

[0115] where, is the Pearson correlation coefficient, is the decomposed signal IMF, is the prediction result of the DeepESN neural network, is the number of samples, is the mean value of the decomposed signal IMF, is the mean of the prediction results.

[0116] After obtaining the correlation coefficients between each IMF and the prediction results, by setting a threshold judge the correlation between each modal component and the original signal to determine the content of noise. The purpose is to consider the components with high correlation as effective signals and the components with low correlation as noise. In this embodiment, the correlation threshold is set to 0.6.

[0117] After screening out the high-correlation intrinsic mode functions (HIMFs), the signal is phase-reconstructed, and finally a smooth and more accurate SOE prediction value is obtained, as shown in formula (23).

[0118]

[0119] where is the number of IMFs greater than the threshold, represents the th high-correlation intrinsic mode function.

[0120] On the other hand, this embodiment also provides an electronic device, including a memory, a processor, and a computing program stored in the memory and executable on the processor. When the processor executes the computing program, the method is implemented.

[0121] On the other hand, this embodiment also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method is implemented.

[0122] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for estimating the energy state of a high-speed lithium battery, characterized in that Including: Obtain input data of a lithium battery, where the input data includes current, voltage, and power; Obtain an original data set based on a general driving condition experiment; Construct a deep echo state network, and train the deep echo state network based on the original data set to obtain a lithium battery state of energy (SOE) estimation model; Input the input data into the lithium battery SOE estimation model for calculation to obtain an SOE prediction value with output noise; Construct a modal decomposition noise reduction module, and remove the sudden output signal in the SOE prediction value with output noise based on the modal decomposition noise reduction module to obtain a smooth lithium battery energy state prediction value.

2. The method according to claim 1, characterized in that The deep echo state network includes multiple stacked reservoir layers, and the input data is subjected to multi-time scale modeling and hierarchical feature extraction through the reservoir layers; The deep echo state network adopts a parameter differentiation strategy to independently adjust the hyperparameters of different reservoir layers, and each adjusted reservoir processes input data with different dynamic characteristics.

3. The method according to claim 2, characterized in that, The steps for updating the reservoir state of the deep echo state network include: According to the input data and the initialized weight matrix, update the state of the first-layer reservoir, combine the external input with the bias, and obtain the first-layer state through a non-linear transformation; Use the output of the first-layer reservoir as the input to update the states of the second-layer and above reservoirs, and transmit information layer by layer to form a deep dynamics.

4. The method according to claim 1, wherein The steps for training the output weights of the deep echo state network include: Generate a joint state matrix according to the updated reservoir state; Remove the transient data in the joint state matrix and retain the steady-state data for training; Use the ridge regression method to calculate the output weights to obtain an SOE prediction value with output noise.

5. The method according to claim 4, characterized in that, The process of removing the sudden output signal in the SOE prediction value with output noise based on the noise reduction module includes: Input the SOE prediction value with output noise into the modal decomposition noise reduction module; Use the successive variational mode decomposition technique to decompose the SOE prediction value into multiple intrinsic mode functions and a residual signal.

6. The method according to claim 1, wherein The noise reduction steps of the modal decomposition noise reduction module include: Perform Pearson correlation analysis on each decomposed intrinsic mode function to calculate its correlation coefficient with the original prediction value; Screen out the high-correlation intrinsic mode functions according to a preset correlation threshold and remove the low-correlation noise signals.

7. The method according to claim 1, wherein The reconstruction steps of the modal decomposition noise reduction module include: performing phase addition and reconstruction on the screened high-correlation intrinsic mode functions to obtain the final smooth and accurate lithium battery energy state prediction value.

8. The method according to claim 5, wherein The optimization process of the successive variational mode decomposition technique includes: Introduce an augmented Lagrangian function to transform the modal decomposition problem into a constrained optimization problem; Use the alternating direction method of multipliers for iterative optimization to update the mode function, central frequency, and Lagrange multiplier.

9. An electronic device, comprising a memory, a processor, and a computing program stored in the memory and executable on the processor, characterized in that, When the processor executes the computing program, it implements the method according to any one of claims 1-8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the method according to any one of claims 1-8.

Images