A fast response method for lithium battery state of charge estimation

Abstract

The present application relates to the technical field of machine learning, and especially relates to a fast response method for lithium battery state of charge estimation, which utilizes a joint series-parallel second-order RC model to estimate real-time state of charge, establishes a Volterra model based on big data modeling theory, generates a model relationship between current voltage and state of charge, establishes a corresponding model of current voltage and state of charge, and thus estimates the state of charge; wherein the memory characteristics of the Volterra model can greatly improve the accuracy of model prediction, and the multi-direction algorithm has a faster convergence speed. The present application solves the problem that traditional methods require high-precision measuring instruments and require a long discharge time for testing in real applications, thus not having practical application value; the present application has a wide range of applications, and theory and simulation verification have proved the accuracy and rapidity of the method.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and in particular to a fast response method for estimating the state of charge of lithium batteries. Background Technology

[0002] Due to the energy crisis, the market share of electric vehicles has been increasing, and many traditional automakers have also entered the electric vehicle market. However, the most serious challenge facing electric vehicles is the issue of energy storage. Lithium-ion batteries, due to their high energy density and environmental friendliness, have become an important power source for new energy vehicles. To reduce lithium battery consumption and increase battery life, estimating the state of charge (SOC) of lithium batteries has become particularly important.

[0003] Traditional methods such as Kalman filtering and equivalent circuit methods have the following problems:

[0004] (1) The battery needs to be kept open for at least one hour, which is difficult to achieve in a real industrial environment.

[0005] (2) The accuracy requirements of the measuring instruments are particularly high. The accumulation of errors in the measuring instruments can easily lead to a larger error in the state of charge of the battery. Summary of the Invention

[0006] To address the shortcomings of existing algorithms, this invention addresses the nonlinear relationship between the State of Charge (SOC) and the battery's input and output in a second-order RC circuit model. Accurately describing this relationship is crucial for estimating the SOC. The Volterra model, with its memory factor and various orders, can approximate any nonlinear system. Theoretically, by adjusting the order and memory factor based on the system's input and output, the Volterra model can approximate arbitrarily complex nonlinear systems, provided the order and memory factor are sufficiently high. However, higher memory factors and orders lead to a dramatic increase in the dimensionality of the Volterra model's parameters, causing the curse of dimensionality. Higher system dimensionality makes it more difficult to guarantee both accuracy and speed in identification. This invention proposes a fast-response method for estimating the SOC of lithium-ion batteries using the high-dimensional Volterra model, enabling rapid and accurate estimation.

[0007] The technical solution adopted in this invention is as follows: a fast response method for estimating the state of charge of a lithium battery includes the following steps: using a second-order RC model and estimating the state of charge of the second-order RC model battery with the help of a Volterra model; wherein, the input of the Volterra model is current and voltage, and the output is the current state of charge of the battery, and the state of charge is accurately fitted by adjusting the length of the memory factor.

[0008] Further, detailed steps include:

[0009] S1. Collect accurate values ​​of current, voltage, and SOC of the battery circuit at every moment.

[0010] S2. Construct a Volterra model using current and voltage as inputs and SOC (State of Charge) as output.

[0011] Furthermore, the Volterra model is as follows:

[0012] SOC(t)=a1I(t)+a2I(t-1)+…a i I(t-i+1)…a n I(t-n+1)+b1U(t)+…b p U(t-p+1)…+b m U(t-m+1)+a 1,1 I 2 (t)+a 1,2 I(t)I(t-1)+…a i,j I(t-i+1)I(t-j+1)…a n,n I 2 (t-n+1)+b 1,1 U 2 (t)+…b p,q U(t-p+1)U(t-q+1)…+b m,m U 2 (t-m+1)+v(t)

[0013] Where SOC is the state of charge of the battery circuit at a certain moment, I is the current of the battery circuit at a certain moment, and U is the current of the battery circuit at a certain moment.

[0014] S3. Construct L sets of input and output data, design a loss function, and minimize the loss function value by selecting different parameters θ.

[0015] Furthermore, the Volterra model for the L sets of data:

[0016]

[0017] Where Y(L) is the output vector constructed from L outputs, and Φ(L) is the output vector constructed from L information vectors. The constructed information matrix, V(L), is a noise vector constructed from L noise v(t);

[0018] Furthermore, the loss function is:

[0019]

[0020] S4. For the Volterra model, multiple directions are constructed using a multi-directional algorithm;

[0021] Furthermore, in detail: Let the parameter estimate of the (k-1)th iteration be θ. k-1 First, construct the first direction of the k-th iteration, which is the source direction of the k-th iteration:

[0022]

[0023] Construct the remaining l-1 directions:

[0024]

[0025] Where matrix N is a non-unit singular matrix, generally to reduce computational cost, N is assumed to be a diagonal matrix, and normalization is performed for each direction, i.e.:

[0026]

[0027] Then the l directions are:

[0028] D k =[d k,1 ,…,d k,l (7)

[0029] Any two of these directions are orthogonal.

[0030] S5. Based on the direction and loss function, construct the step size corresponding to each direction;

[0031] Furthermore, the step size R corresponding to each direction k The formula is as follows:

[0032]

[0033] Based on the fact that the derivative of the above function is zero, we can then derive:

[0034]

[0035] S6. Update system parameters using a multi-directional algorithm, the formula is:

[0036] θ k =θ k-1 +D k R k (10)

[0037] S7. Set the number of training iterations K, and repeat steps S4-S6.

[0038] The beneficial effects of this invention are:

[0039] 1. It can measure the current state of charge in real time;

[0040] 2. It does not require high precision in measuring instruments, but it provides more accurate and faster estimation of battery state of charge;

[0041] 3. Applicable to SOC estimation in any situation, with low requirements for external environment and equipment, and easy to implement in practical engineering practice. Attached Figure Description

[0042] Figure 1 This is a circuit diagram illustrating the current, voltage, and SOC of the present invention;

[0043] Figure 2 This is a schematic diagram of the lithium battery testing platform of the present invention;

[0044] Figure 3 This is a schematic diagram of the multi-directional algorithm flow of the present invention;

[0045] Figure 4 These are different rate graphs corresponding to different numbers of directions in the multi-directional algorithm of this invention;

[0046] Figure 5 This is a diagram of a lithium battery testing platform, used to collect data on lithium battery charging and discharging. Detailed Implementation

[0047] The present invention will be further described below with reference to the accompanying drawings and embodiments. The drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0048] like Figure 1 The second-order RC model represents the current and voltage. The second-order RC equivalent circuit model adds a parallel RC circuit in series with the Thevenin model. The two parallel RC circuits simulate concentration polarization and electrochemical polarization respectively. While describing the nonlinear characteristics of lithium-ion batteries, it more accurately describes the polarization effect during the charging and discharging process of lithium batteries.

[0049] In lithium battery SOC estimation, a second-order RC model is constructed to treat battery voltage U and current I as inputs and SOC as output, thereby obtaining an accurate state of charge based on the current battery voltage and current.

[0050] like Figure 2 This is a schematic diagram of the lithium battery testing platform of the present invention, including a lithium battery pack, testing instruments and a computer. The testing instruments collect the voltage and current of the lithium battery pack, and the computer contains a Volterra model.

[0051] A fast response method for estimating the state of charge of lithium batteries includes the following steps:

[0052] First, the model relationship between the state of charge and voltage and current is approximated using the Volterra model;

[0053] The construction of the Volterra model includes:

[0054] S1. Data Acquisition: Acquire the accurate values ​​of current, voltage and SOC of the battery circuit at each moment; where current I = [I(1), ..., I(L)], voltage U = [U(1), ..., U(L)], and the state of charge of SOC is: SOC = [SOC(1), ..., SOC(L)];

[0055] S2. Constructing the Volterra model: Using current and voltage as inputs and SOC (State of Charge) as output, construct the following Volterra model:

[0056]

[0057] Where SOC is the state of charge of the battery circuit at a certain moment, I is the current of the battery circuit at a certain moment, and U is the current of the battery circuit at a certain moment.

[0058] The unknown parameter of the model is a i ,a i,j ,b p ,b p,q There are a total of There are several parameters, where n and m are the memory factors for current and voltage, respectively. As the memory factor increases, the dimension of the corresponding parameter increases quadratically, which leads to the curse of dimensionality.

[0059] S3. Construct a Volterra model for the L sets of data:

[0060]

[0061] Where Y(L) is the output vector constructed from L outputs, and Φ(L) is the output vector constructed from L information vectors. The constructed information matrix, V(L), is a noise vector constructed from L noise v(t).

[0062] Design the loss function:

[0063]

[0064] By selecting an appropriate parameter θ, the value of the above loss function can be minimized;

[0065] like Figure 3 This is a schematic diagram of the multi-directional algorithm flow. S3: Initialize Volterra model parameters;

[0066] S4. For the Volterra model, multiple directions are constructed using a multi-directional algorithm. Assume the parameter estimate for the (k-1)th iteration is θ. k-1 First, construct the first direction of the k-th iteration, which is the source direction of the k-th iteration:

[0067]

[0068] Construct the remaining l-1 directions:

[0069]

[0070] Where matrix N is a non-unit singular matrix, generally to reduce computational cost, N is assumed to be a diagonal matrix, and normalization is performed for each direction, i.e.:

[0071]

[0072] Then the l directions are:

[0073] D k =[d k,1 ,…,d k,l (7)

[0074] Any two of these directions are orthogonal.

[0075] S5. Based on the direction and loss function, construct the step size R for each direction. k The formula is as follows:

[0076]

[0077] Based on the fact that the derivative of the above function is zero, we can then derive:

[0078]

[0079] S6. Update system parameters using a multi-directional algorithm, the formula is:

[0080] θ k =θ k-1 +D k R k (10)

[0081] S7. Set the number of training iterations K, and repeat steps S4-S6.

[0082] like Figure 4 The graph shows the different rates corresponding to different numbers of directions in the multi-directional algorithm. The parameter estimation speed corresponds to different directions. The blue direction has 1 direction, the red direction has 4 directions, the green direction has 6 directions, and the black direction has the same number of directions as the system dimension, which is 10. It can be seen from the graph that the convergence speed is the fastest when the number of directions is 10.

[0083] The multi-directional algorithm is compared with the least squares method and the gradient algorithm. The results are shown in the table below. Figure 5 As shown:

[0084] algorithm Least squares algorithm gradient algorithm Multi-directional algorithm (3 directions) Algorithm execution time (seconds) 4.529 5.575 3.892 Number of jumps (only multiplication and division are counted) 1 074 060 1 166 040 994 890

[0085] The multi-directional algorithm uses multiple orthogonal directions to identify system parameters each time, and its computational cost is significantly reduced compared to the least squares algorithm. Because it involves multiple directions, its convergence speed is much higher than that of the gradient algorithm.

[0086] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A fast response method for estimating the state of charge of a lithium battery, characterized in that, Includes the following steps: The state of charge of the second-order RC model battery is estimated by using the Volterra model and adjusting the memory factor length to achieve an accurate fit to the state of charge. Specifically, it includes: S1. Collect accurate values ​​of current, voltage, and SOC of the battery circuit at every moment. S2. Construct a Volterra model using current and voltage as inputs and SOC (State of Charge) as output. The formula for the Volterra model is: in, SOC This represents the state of charge of the battery circuit at a certain moment. I Let be the current in the battery circuit at a certain moment. U The voltage of the battery circuit at a certain moment; These are the memory factors for current and voltage, respectively; S3, Construction L Given a set of data, design a loss function by selecting different parameters. To minimize the value of the loss function; Volterra model for group data: （2） in, yes L The output vector is constructed from the outputs. It is by L Information vectors The constructed information matrix yes L One noise, The constructed noise vector; The loss function is: （3） S4. For the Volterra model, multiple directions are constructed using a multi-directional algorithm; Multi-directional algorithms construct multiple directions, including: Let the first The parameter estimation of the second time is First, construct the first The first direction: （4） Build the remainder One direction: （5） Among them, matrix These are non-unit singular matrices, normalized for each direction: （6） but The directions are: （7） Any two of these directions are orthogonal; S5. Based on the direction and loss function, construct the step size corresponding to each direction; Step size for each direction The formula is as follows: （8） Based on the fact that the derivative of the function in equation (8) is zero, we can obtain: （9） S6. Update system parameters using a multi-directional algorithm, the formula is: （10） S7. Set the number of training sessions. K And repeat steps S4-S6.

Images