Lithium ion battery SOC estimation method based on LSTM

Abstract

The invention discloses a lithium ion battery SOC estimation method based on LSTM, and the method comprises the steps: building an equivalent circuit model of a lithium ion battery through employing a second-order Thevenin model, and obtaining a system state equation and an observation equation; on the basis of a system state equation and an observation equation, a whale optimization algorithm is adopted to identify model parameters online; on the basis of the equivalent circuit model subjected to parameter identification, according to mapping related to internal characteristics of the battery, obtaining a moving average voltage; and adding the moving average voltage into an input end, and calculating to obtain an SOC estimation result of the lithium ion battery according to an SOC function by adopting an evaluation method based on LSTM (Long Short Term Memory). According to the method, the equivalent circuit model of the lithium ion battery is established, the whale optimization algorithm is adopted to identify the model parameters online, and the SOC estimation effect based on the LSTM is verified according to experiments, so that the problem that the model precision and complexity are difficult to consider at the same time is solved, the accuracy of SOC estimation can be effectively improved, and the service life and the utilization efficiency of the battery can be improved.

Description

Technical Field

[0001] The present invention belongs to the field of lithium-ion battery management, and in particular relates to a lithium-ion battery SOC estimation method based on LSTM. Background Art

[0002] Lithium-ion batteries are widely used in power storage and new energy vehicles due to their high energy density and long cycle life. Lithium-ion batteries are a type of secondary battery that mainly relies on the movement of lithium ions between the positive and negative electrodes to work. Due to their high energy density, fast charging speed, long service life, small memory effect, low self-discharge rate, and low pollution, they have received widespread attention in the field of electric vehicles and are also key components related to the continuous, reliable, and safe operation of electric vehicles.

[0003] The battery state of charge (SOC) is one of the key states that need to be monitored in battery applications. It plays a vital role in the safe and rational use of the battery. Accurately estimating the battery SOC is crucial to improving the battery life and utilization efficiency. The battery SOC is defined as the percentage of the remaining available capacity of the battery relative to the nominal capacity. It is an estimated parameter rather than an observable parameter, so it cannot be directly obtained through sensor measurement, but needs to be indirectly estimated by the battery management system (BMS) using relevant algorithms. At present, battery SOC estimation often uses empirical and model-based methods.

[0004] Empirical estimation algorithms include: Luo Yong et al. proposed an ampere-hour integration method with capacity correction, which estimates the battery SOC by estimating the battery charge and discharge capacity in real time. In addition, Xing et al. also estimated the SOC of lithium batteries by measuring the open circuit voltage and SOC of lithium batteries and fitting the functional relationship between the two. Although this open-loop estimation algorithm is simple in principle, easy to implement and has low computational complexity, it usually needs to meet many prerequisites, and its robustness is poor due to the lack of feedback links.

[0005] Model-based estimation algorithms include: Liu et al. used an electrochemical single-particle model to design a terminal voltage feedback injection nonlinear observer to monitor the SOC of lithium-ion batteries. Zhang Xiaoyang et al. proposed a SOC estimation method based on a fractional-order model, using an adaptive extended Kalman particle filter to improve the estimation accuracy. Fu et al. proposed using a linear Kalman filter for online parameter identification to achieve SOC estimation.

[0006] However, the equivalent circuit model has a clear structure and easy-to-identify parameters, but it is difficult to balance the accuracy and complexity of the model. Therefore, a new technical solution is needed to solve these problems. Summary of the invention

[0007] Purpose of the invention: In order to overcome the shortcomings of the prior art, a SOC estimation method for lithium-ion batteries based on LSTM is provided. The equivalent circuit model of the lithium-ion battery is established using the second-order Thevenin model, and the model parameters are identified online using the whale optimization algorithm. The SOC estimation effect based on LSTM is verified by experiments, which solves the problem of balancing the accuracy and complexity of the model, can effectively improve the accuracy of SOC estimation, and is beneficial to improving the service life and utilization efficiency of the battery.

[0008] Technical solution: To achieve the above purpose, the present invention provides a lithium-ion battery SOC estimation method based on LSTM, comprising the following steps:

[0009] S1: Use the second-order Thevenin model to establish the equivalent circuit model of the lithium-ion battery and obtain the system state equation and observation equation;

[0010] S2: Based on the system state equation and observation equation, the whale optimization algorithm is used to identify the model parameters online;

[0011] S3: Based on the equivalent circuit model after parameter identification, the sliding average voltage is obtained according to the mapping related to the internal characteristics of the battery;

[0012] S4: Add the sliding average voltage to the input end, use the LSTM-based evaluation method, and obtain the lithium-ion battery SOC estimation result according to the SOC function calculation.

[0013] Furthermore, the step S1 is specifically as follows:

[0014] According to Kirchhoff's law, we can get:

[0015]

[0016] Calculate the battery state of charge SOC:

[0017]

[0018] Among them, SOC0 represents the initial state of charge of the battery, η represents the charge and discharge efficiency, C n Indicates the battery capacity;

[0019] Discretize equations (1) and (2), and select SOC, U1, and U2 as the state variables of the system to obtain the system state equation and observation equation.

[0020] The system state equation and observation equation are expressed as follows:

[0021]

[0022] Uout (k) = U ocv (SOC(k))-R0I(k)-U1(k)-U2(k)+v(k) (4) Wherein, Δt represents the sampling time interval, which is set to 1s in the present invention, and k represents discrete time.

[0023] Furthermore, the method of online identifying model parameters using the whale optimization algorithm in step S2 includes:

[0024] A1: According to equations (3) and (4), the equivalent circuit model Laplace equation can be obtained

[0025]

[0026] Among them, α1, α2, α3, α4, and α5 represent the parameters to be identified;

[0027] A2: Get the objective function of the whale optimization algorithm

[0028]

[0029] Among them, U out (k) represents the model terminal voltage, Indicates the measured terminal voltage;

[0030] A3: Solve the objective function of the whale optimization algorithm and complete parameter identification.

[0031] Furthermore, WOA is a meta-heuristic swarm intelligence optimization algorithm, which mainly simulates the "spiral bubble net" predation strategy of humpback whales, and has the characteristics of few optimization parameters, fast convergence speed and easy acquisition of global optimal solutions. The solution steps in step A3 include encircling prey, bubble net hunting and searching prey, which are as follows:

[0032] A3-1: Surrounding the Prey

[0033] At this stage, the algorithm aims to move the solution (humpback whale) to the best solution in the search space, updating the mathematical expression for the whale's position

[0034]

[0035] Where l represents the number of cycles, It represents the position vector of the best solution obtained from the beginning of the algorithm to the lth iteration. represents the position vector of the current solution, Iteration vector representing the current solution position approaching the current optimal solution position;

[0036] and Represents parameter vector

[0037]

[0038] in, represents the parameter vector, whose value decreases linearly from 2 to 0 during the algorithm iteration process, r represents a random number, r∈[0,1]; A3-2: Bubble Net Hunting

[0039] The WOA algorithm reduces the parameter vector The value of , simulates the shrinkage and encirclement mechanism of humpback whale hunting; According to formula (13), the parameter vector The value of decreases with the decrease of, in other words, The value of is a random number between [-a, a]; when When it decreases to 1, The value range of becomes [-1,1]. At this time, the next new position of the solution (humpback whale) is defined at any position between its current position and the current best solution position.

[0040] The WOA algorithm creates a spiral equation between the position of the humpback whale and its prey to simulate the humpback whale's spiral hunting behavior.

[0041]

[0042] in, represents the distance between the whale and the current best solution position, b is a constant used to define the spiral shape, q represents a random number, q∈[-1,1];

[0043] Humpback whales swim along spiral paths, and also swim in shrinking circles; suppose during the iteration of the algorithm, the probability of humpback whales choosing to shrink or spiral is 50%

[0044]

[0045] A3-3: Searching for prey

[0046] The WOA algorithm designs a random search mechanism to enhance the global search capability. Contrary to the shrinking and encircling stage, when When the modulus is greater than 1, the whale is searching for prey.

[0047]

[0048] in, represents a random position vector.

[0049] Furthermore, the addition of the sliding average voltage in step S3 makes the network input information have a slow time-varying characteristic, which is conducive to learning the mapping related to the internal characteristics of the battery when the battery voltage and current change drastically, thereby improving the SOC estimation accuracy. The expression of the sliding average voltage is as follows:

[0050]

[0051] Among them, U AV (t) represents the sliding average voltage at time t, and m represents the sliding window time.

[0052] Furthermore, the calculation expression of the lithium-ion battery SOC estimation result in step S4 is as follows:

[0053]

[0054] Among them, f(t), i(t), o(t) are the forget gate, input gate and output gate respectively; is the candidate state, c(t) is the cell state; h(t) is the hidden layer state; x(t) is the input vector, including battery voltage, current, temperature, and average voltage; SOC(t) is the estimated SOC at time t; g(·) represents the activation function of the output layer, and the present invention adopts the Sigmoid function; U, W, V are weight matrices, and b is the bias vector.

[0055] Beneficial effects: Compared with the prior art, the present invention uses the second-order Thevenin model to establish an equivalent circuit model of a lithium-ion battery, adopts the whale optimization algorithm to identify model parameters online, and verifies the SOC estimation effect based on LSTM through experiments, thereby solving the problem of balancing model accuracy and complexity. It can effectively improve the accuracy of SOC estimation, which is beneficial to improving the service life and utilization efficiency of the battery. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 is a flow chart of the method of the present invention;

[0057] Figure 2 This is the second-order RC circuit model diagram of a lithium battery;

[0058] Figure 3 This is a flow chart of lithium battery model parameter identification based on WOA;

[0059] Figure 4 It is the minimum unit diagram of LSTM;

[0060] Figure 5 This is the battery parameter identification and verification result diagram based on WOA;

[0061] Figure 6 This is the SOC estimation result diagram. DETAILED DESCRIPTION

[0062] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, various equivalent forms of modifications to the present invention by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0063] like Figure 1 As shown, the present invention provides a lithium-ion battery SOC estimation method based on LSTM, comprising the following steps:

[0064] S1: Reference Figure 2 , using the second-order Thevenin model, the equivalent circuit model of the lithium-ion battery is established to obtain the system state equation and observation equation;

[0065] According to Kirchhoff's law, we can get:

[0066]

[0067] Calculate the battery state of charge SOC:

[0068]

[0069] Among them, SOC0 represents the initial state of charge of the battery, η represents the charge and discharge efficiency, C n Indicates the battery capacity;

[0070] Discretize equations (1) and (2), and select SOC, U1, and U2 as the state variables of the system to obtain the system state equation and observation equation.

[0071] The system state equation and observation equation are expressed as follows:

[0072]

[0073] U out (k) = U ocv (SOC(k))-R0I(k)-U1(k)-U2(k)+v(k) (4)

[0074] Wherein, Δt represents the sampling time interval, which is set to 1s in the present invention, and k represents discrete time.

[0075] S2: Based on the system state equation and observation equation, the whale optimization algorithm is used to identify the model parameters online;

[0076] Methods for online identification of model parameters using the whale optimization algorithm include:

[0077] A1: According to equations (3) and (4), the equivalent circuit model Laplace equation can be obtained

[0078]

[0079]

[0080] Among them, α1, α2, α3, α4, and α5 represent the parameters to be identified;

[0081] A2: Get the objective function of the whale optimization algorithm

[0082]

[0083] Among them, U out (k) represents the model terminal voltage, Indicates the measured terminal voltage;

[0084] A3: Solve the objective function of the whale optimization algorithm and complete parameter identification.

[0085] WOA is a meta-heuristic swarm intelligence optimization algorithm, which mainly simulates the humpback whale's "spiral bubble net" predation strategy. It has the characteristics of few optimization parameters, fast convergence speed and easy acquisition of global optimal solution. The solution steps in step A3 include encircling prey, bubble net hunting and searching prey. Figure 3 , the specific process is as follows:

[0086] A3-1: Surrounding the Prey

[0087] At this stage, the algorithm aims to move the solution (humpback whale) to the best solution in the search space, updating the mathematical expression for the whale's position

[0088]

[0089] Where l represents the number of cycles, It represents the position vector of the best solution obtained from the beginning of the algorithm to the lth iteration. represents the position vector of the current solution, Iteration vector representing the current solution position approaching the current optimal solution position;

[0090] and Represents parameter vector

[0091]

[0092] in, represents the parameter vector, whose value decreases linearly from 2 to 0 during the algorithm iteration process, r represents a random number, r∈[0,1];

[0093] A3-2: Bubble Net Hunting

[0094] The WOA algorithm reduces the parameter vector The value of , simulates the shrinkage and encirclement mechanism of humpback whale hunting; According to formula (13), the parameter vector The value of decreases with the decrease of, in other words, The value of is a random number between [-a, a]; when When it decreases to 1, The value range of becomes [-1,1]. At this time, the next new position of the solution (humpback whale) is defined at any position between its current position and the current best solution position.

[0095] The WOA algorithm creates a spiral equation between the position of the humpback whale and its prey to simulate the humpback whale's spiral hunting behavior.

[0096]

[0097] in, represents the distance between the whale and the current best solution position, b is a constant used to define the spiral shape, q represents a random number, q∈[-1,1];

[0098] Humpback whales swim along spiral paths, and also swim in shrinking circles; suppose during the iteration of the algorithm, the probability of humpback whales choosing to shrink or spiral is 50%

[0099]

[0100] A3-3: Searching for prey

[0101] The WOA algorithm designs a random search mechanism to enhance the global search capability. Contrary to the shrinking and encircling stage, when When the modulus is greater than 1, the whale is searching for prey.

[0102]

[0103] in, represents a random position vector.

[0104] S3: Based on the equivalent circuit model after parameter identification, the sliding average voltage is obtained according to the mapping related to the internal characteristics of the battery;

[0105] The expression of sliding average voltage is as follows:

[0106]

[0107] Among them, U AV (t) represents the sliding average voltage at time t, and m represents the sliding window time.

[0108] S4: Add the sliding average voltage to the input terminal, use the LSTM-based evaluation method, and obtain the SOC estimation result of the lithium-ion battery according to the SOC function calculation;

[0109] The addition of sliding average voltage makes the network input information have slow time-varying characteristics, which is conducive to learning the mapping related to the internal characteristics of the battery when the battery voltage and current change drastically, thereby improving the accuracy of SOC estimation;

[0110] Reference Figure 4 , the calculation expression of the lithium-ion battery SOC estimation result is as follows:

[0111]

[0112] Among them, f(t), i(t), o(t) are the forget gate, input gate and output gate respectively; is the candidate state, c(t) is the cell state; h(t) is the hidden layer state; x(t) is the input vector, including battery voltage, current, temperature, and average voltage; SOC(t) is the estimated SOC at time t; g(·) represents the activation function of the output layer, and the present invention adopts the Sigmoid function; U, W, V are weight matrices, and b is the bias vector.

[0113] Based on the above scheme, in order to verify the effectiveness and performance of the method of the present invention, this embodiment is experimentally verified, as follows:

[0114] This example analyzes the equivalent circuit model of lithium-ion batteries. In order to balance accuracy and complexity, a second-order RC circuit model of a 2.8Ah lithium-ion battery was built in Matlab / simulink, and SOC, U1, and U2 were selected as the state variables of the system, and discretized to establish the system state equation and observation equation.

[0115] This embodiment uses the whale optimization algorithm to identify model parameters online to improve modeling accuracy. In order to verify the effectiveness of the proposed WOA-based parameter identification method, a data set released by the National Aeronautics and Space Administration (NASA) is used. The FUDS working condition under a 25°C environment is selected, and the WOA solution steps: surrounding prey, bubble net hunting, and searching prey are used to identify model parameters and verify them. The verification results are as follows: Figure 5 shown.

[0116] The root mean square error (RMSE) between the estimated value and the actual value of the battery terminal voltage is usually used to show the performance of different parameter identification methods. The smaller the RMSE, the more accurate the algorithm. Figure 5The RMSE of the WOA-based parameter identification method proposed in the present invention is only 0.0678, which shows that the present invention can effectively identify the model parameters and the battery model has a high accuracy.

[0117] This embodiment uses an LSTM-based evaluation method to estimate the SOC of lithium-ion batteries. Since the dynamic response of the battery is highly nonlinear, an additional slow time-varying information is introduced, namely the battery sliding average voltage. After adding the sliding average voltage to the input end, the LSTM SOC is calculated using the FUDS operating condition experiment under a 25°C environment, and the SOC estimation effect is verified.

[0118] The estimated curve is Figure 6 As shown, the initial SOC value is set to 90%. Figure 6 It can be seen that the algorithm proposed in the present invention can quickly converge to the true value at the beginning of discharge and track it closely. The RMSE index is only 0.0269, which verifies the effectiveness of the proposed invention.

[0119] It can be seen from the results that the method of the present invention can effectively calculate the SOC of the lithium-ion battery and improve the estimation accuracy of the SOC of the lithium-ion battery.

Claims

1. A lithium-ion battery SOC estimation method based on LSTM, characterized in that: The steps include: S1: Use the second-order Thevenin model to establish the equivalent circuit model of the lithium-ion battery and obtain the system state equation and observation equation; S2: Based on the system state equation and observation equation, the whale optimization algorithm is used to identify the model parameters online; S3: Based on the equivalent circuit model after parameter identification, the sliding average voltage is obtained according to the mapping related to the internal characteristics of the battery; S4: Add the sliding average voltage to the input end, use the LSTM-based evaluation method, and obtain the lithium-ion battery SOC estimation result according to the SOC function calculation.

2. The method for estimating SOC of a lithium-ion battery based on LSTM according to claim 1, characterized in that: The step S1 is specifically as follows: According to Kirchhoff's law, we can get: Calculate the battery state of charge SOC: Among them, SOC0 represents the initial state of charge of the battery, η represents the charge and discharge efficiency, C n Indicates battery capacity; Discretize equations (1) and (2), and select SOC, U1, and U2 as the state variables of the system to obtain the system state equation and observation equation.

3. A lithium-ion battery SOC estimation method based on LSTM according to claim 2, characterized in that: The system state equation and observation equation are expressed as follows: U out (k)=U ocv (SOC(k))-R0I(k)-U1(k)-U2(k)+v(k) (4) Where Δt represents the sampling time interval and k represents the discrete time.

4. The method for estimating SOC of a lithium-ion battery based on LSTM according to claim 3, characterized in that: The method for online identification of model parameters using the whale optimization algorithm in step S2 includes: A1: According to equations (3) and (4), the equivalent circuit model Laplace equation can be obtained Among them, α1, α2, α3, α4, and α5 represent the parameters to be identified; A2: Get the objective function of the whale optimization algorithm Among them, U out (k) represents the model terminal voltage, Indicates the measured terminal voltage; A3: Solve the objective function of the whale optimization algorithm and complete parameter identification.

5. The method for estimating SOC of a lithium-ion battery based on LSTM according to claim 4, characterized in that: The solution steps in step A3 include surrounding the prey, hunting with a bubble net and searching for the prey, which are specifically as follows: A3-1: Surrounding the Prey Move the solution to the best solution in the search space and update the mathematical expression of the whale's position Where l represents the number of cycles, It represents the position vector of the best solution obtained from the beginning of the algorithm to the lth iteration. represents the position vector of the current solution, Iteration vector representing the current solution position approaching the current optimal solution position; and Represents parameter vector in, represents the parameter vector, whose value decreases linearly from 2 to 0 during the algorithm iteration process, r represents a random number, r∈[0,1]; A3-2: Bubble Net Hunting The WOA algorithm reduces the parameter vector The value of , simulates the shrinkage and encirclement mechanism of humpback whale hunting; According to formula (13), the parameter vector The value of decreases with the decrease of, in other words, The value of is a random number between [-a, a]; when When it decreases to 1, The value range of becomes [-1,1], at this time, the next new position of the solution is defined at any position between its current position and the current best solution position; The WOA algorithm creates a spiral equation between the position of the humpback whale and its prey to simulate the humpback whale's spiral hunting behavior. in, represents the distance between the whale and the current best solution position, b is a constant used to define the spiral shape, q represents a random number, q∈[-1,1]; Humpback whales swim along spiral paths, and also swim in shrinking circles; suppose during the iteration of the algorithm, the probability of humpback whales choosing to shrink or spiral is 50% A3-3: Searching for prey The WOA algorithm designs a random search mechanism to enhance the global search capability. Contrary to the shrinking and surrounding stage, when When the modulus is greater than 1, the whale is searching for prey. in, represents a random position vector.

6. The method for estimating SOC of a lithium-ion battery based on LSTM according to claim 4, characterized in that: The expression of the sliding average voltage in step S3 is as follows: Among them, U AV (t) represents the sliding average voltage at time t, and m represents the sliding window time.

7. The method for estimating SOC of a lithium-ion battery based on LSTM according to claim 6, characterized in that: The calculation expression of the lithium-ion battery SOC estimation result in step S4 is as follows: Among them, f(t), i(t), o(t) are the forget gate, input gate and output gate respectively; is the candidate state, c(t) is the cell state; h(t) is the hidden layer state; x(t) is the input vector, including battery voltage, current, temperature, and average voltage; SOC(t) is the estimated SOC at time t; g(·) represents the activation function of the output layer; U, W, V are weight matrices, and b is the bias vector.

Images