Parameter identification method for lithium iron phosphate battery equivalent model based on improved scso

Abstract

The application discloses a lithium iron phosphate battery equivalent model parameter identification method based on an improved SCSO, which comprises the following steps: step 1, constructing a second-order equivalent circuit model to determine to-be-identified parameters; step 2, collecting current, voltage and time data of the lithium battery under a dynamic working condition, and constructing a fitness function based on the deviation of a model predicted voltage and a measured voltage; step 3, generating an initial population of the improved sand cat colony optimization algorithm by adopting a chaotic initialization strategy, and setting a search boundary in combination with parameter physical constraint settings; step 4, introducing a dynamic weight disturbance mechanism, an adaptive mutation strategy and a triangular walking strategy to iteratively update the sand cat population position, and reserving high-quality solutions by greedy selection; and step 5, repeatedly executing steps 3 and 4 until the fitness function reaches a preset precision or the number of iterations reaches a maximum value, and outputting optimal identification parameters. By dynamically adjusting the search weight to balance the global exploration and local development capability, in combination with the reinforced boundary behavior to avoid parameter out-of-bound to guarantee the physical rationality, and relying on the triangular walking strategy to improve the population diversity, the application effectively solves the inherent defects of the conventional SCSO algorithm in complex parameter optimization.

Description

Technical Field

[0001] This invention belongs to the field of battery management technology, specifically relating to a method for identifying equivalent model parameters of lithium iron phosphate batteries based on improved SCSO. Background Technology

[0002] In fields such as new energy vehicles and energy storage systems, lithium batteries serve as core energy storage components, and the accurate prediction of their operating state directly determines the system's safety, reliability, and lifespan. The accuracy of lithium battery state prediction highly depends on the accuracy of the equivalent circuit model; whether the model parameters can accurately match the actual electrochemical characteristics of the battery is a core prerequisite for state prediction. This process of inverting model parameters through experimental data is known as lithium battery parameter identification. From an engineering perspective, accurate parameter identification not only provides reliable model support for battery management systems (BMS), reducing safety risks such as overcharging and over-discharging, but also optimizes charging and discharging strategies to extend battery cycle life and provides data for battery fault diagnosis. Therefore, conducting research on high-precision, high-robust lithium battery parameter identification has significant theoretical value and practical application implications.

[0003] Currently, lithium battery parameter identification methods are mainly divided into two categories: traditional identification methods and intelligent optimization algorithms. Traditional identification methods are represented by least squares methods, such as recursive least squares (RLS) and weighted least squares (WLS). Their advantages lie in their clear principles, low computational complexity, and ability to quickly solve for parameters of linear or weakly nonlinear models. They were widely used in the identification of low-complexity battery models, such as first-order RC models. However, this method is highly dependent on the linearity of the model. When faced with second-order or higher complex RC models that need to simultaneously characterize ohmic polarization, double-layer polarization, and concentration polarization, the identification accuracy is easily reduced due to the nonlinear characteristics of the model. Furthermore, it is sensitive to noise in the operating data and lacks robustness. The other category is intelligent optimization algorithms, including particle swarm optimization (PSO), genetic algorithms (GA), and whale optimization algorithms (WOA). These algorithms achieve global optimization by simulating the behavior of natural biological groups, without relying on the linear assumptions of the model. They are more adaptable to complex nonlinear models and have better noise resistance than traditional methods. However, existing intelligent optimization algorithms still have obvious shortcomings. For example, the PSO algorithm is prone to getting stuck in local optima in the later stages of iteration, the GA algorithm has a slow convergence speed and high computational cost, and the WOA algorithm has a scattered search direction in high-dimensional parameter optimization, making it difficult to balance identification accuracy and efficiency, and cannot fully meet the parameter identification requirements of lithium batteries in the full SOC range and complex dynamic operating conditions.

[0004] Among emerging intelligent optimization algorithms, the Sand Cat Swarm Optimization (SCSO) algorithm, proposed by Amir Seyyedabbasi and Farzad Kiani in 2022, has shown great potential in function optimization and engineering design due to its unique biological behavior simulation mechanism. Inspired by the survival behavior of sand cats in desert ecosystems, the algorithm abstracts the process of sand cats using their keen hearing to detect low-frequency noise to locate prey and digging burrows with their front paws to capture prey underground into a two-stage optimization strategy of "searching for prey - attacking prey": In the search phase, the algorithm controls the search step size by using random angles and sound wave detection range to achieve extensive exploration of the solution space, corresponding to the algorithm's global exploration. In the attack phase, it achieves a local fine-grained search by clustering towards the optimal individual, attempting to balance the contradiction between global exploration and local exploitation, corresponding to the algorithm's local exploitation. Compared to traditional algorithms such as PSO and GA, its initial optimization framework exhibits stronger global search capabilities and convergence stability in multi-modal function optimization.

[0005] Applying the SCSO algorithm to lithium battery parameter identification offers the following significant advantages: First, lithium battery parameter identification is essentially a multi-parameter optimization problem based on minimizing the deviation between model-predicted voltage and measured voltage. The global optimization characteristic of the SCSO algorithm effectively avoids the dependence of traditional least squares methods on model linearity, adapting to the nonlinear identification needs of second-order and higher-order complex RC models. Second, considering the large differences in the range of lithium battery parameter values ​​and the coupling between parameters, the SCSO algorithm can flexibly set search boundaries to accurately match the physical constraints of different parameters, avoiding parameter out-of-bounds errors. Model distortion; third, compared with the high computational cost of GA and the local optima problem of PSO, the two-stage optimization mechanism of the SCSO algorithm can improve the convergence speed by dynamically adjusting parameters while ensuring global exploration capability, making it more suitable for the identification efficiency requirements in engineering practice; fourth, the SCSO algorithm has better fault tolerance for the initial population than some intelligent algorithms. Even if there is a local deviation in the distribution of the initial population, the missing information can be made up by extensive search in the later stage, improving the identification robustness. It is suitable for parameter identification under complex working conditions such as Dynamic Stress Test (DST) and Federal Urban Driving Program (FUDS).

[0006] However, conventional SCSO algorithms still exhibit numerous limitations in practical lithium battery parameter identification applications. Regarding convergence performance, as iterations progress, their position update mechanism relies solely on a single strategy of "clustering towards the optimal individual," which easily leads to the search agent prematurely falling into local convergence. This makes it difficult to break through local optima when optimizing lithium battery parameters in 6-dimensional or higher dimensions, failing to accurately match the parameter variation patterns across the entire SOC range of the battery, resulting in insufficient convergence accuracy. Furthermore, in the early search phase, the algorithm lacks dynamic adjustment mechanisms for parameters such as the acoustic detection range and search step size. Excessive step size can cause search oscillations, while insufficient step size leads to slow convergence, significantly increasing computational costs and time overhead. When facing high-dimensional optimization problems, the "curse of dimensionality" becomes prominent: in the high-dimensional space of lithium battery parameter identification, the search space expands rapidly. The search direction control mechanism of conventional SCSO algorithms lacks specificity, leading to reduced information interaction efficiency. The algorithm struggles to effectively utilize searched information to guide subsequent optimization, and optimization performance degrades severely with increasing dimensionality. Meanwhile, the conventional SCSO algorithm still relies heavily on the initial population: if the initial population is unevenly distributed and concentrated in a local area of ​​the search space, the algorithm lacks effective information on key parameters in the initial stage, making it difficult to fully explore the solution space, which in turn affects the final optimization result. It is not robust enough and cannot meet the high-precision parameter identification requirements of lithium batteries throughout their entire life cycle and under complex working conditions. Summary of the Invention

[0007] This invention aims to overcome two major defects in the existing technology: First, the traditional second-order RC equivalent circuit model of lithium battery does not consider the dynamic changes in the contact state of the electrode-electrolyte interface with the state of charge, resulting in significant deviations in voltage prediction in the low SOC region, making it difficult to meet the identification accuracy requirements of the entire SOC range; Second, the conventional sandcat swarm optimization algorithm is prone to getting trapped in local optima in high-dimensional parameter optimization, has a slow convergence speed, and poor adaptability to complex working conditions, thus affecting the robustness and efficiency of parameter identification.

[0008] To address the aforementioned issues, this invention proposes an improved Sand Cat Swarm Optimization (SCSO) algorithm that integrates adaptive dynamic weights, enhanced boundary constraint behavior, and a triangle walking strategy. This algorithm is applied to parameter identification of a second-order RC equivalent circuit model of a lithium battery containing dynamic interface contact resistance. This improved algorithm balances global exploration and local exploitation capabilities by dynamically adjusting search weights, combines enhanced boundary behavior to prevent parameter overshooting to ensure physical rationality, and leverages the triangle walking strategy to enhance population diversity. This effectively solves the inherent defects of conventional SCSO algorithms in complex parameter optimization. Furthermore, by introducing dynamic interface contact resistance into the equivalent circuit model, the dynamic changes in the battery interface state are accurately characterized, significantly improving the accuracy and robustness of parameter identification for the lithium battery equivalent circuit model. This enables accurate prediction of battery state in practical engineering applications.

[0009] To achieve the above-mentioned objectives, the present invention adopts the following technical solution: An equivalent model parameter for lithium iron phosphate batteries based on an improved Sand Cat Group Optimization (SCSO) algorithm is the method, comprising the following steps:

[0010] Step 1: Construct a second-order equivalent circuit model and determine the parameters to be identified;

[0011] Step 2: Collect current, voltage, and time data of lithium battery under dynamic operating conditions, and construct a fitness function based on the deviation between the model-predicted voltage and the measured voltage.

[0012] Step 3: Use a chaotic initialization strategy to generate the initial population of the improved sand cat swarm optimization algorithm, and set the search boundary in combination with parameter physical constraints;

[0013] Step 4: Introduce a dynamic weight perturbation mechanism, an adaptive mutation strategy, and a triangle walking strategy to iteratively update the position of the sand cat population, and retain high-quality solutions through greedy selection;

[0014] Step 5: Repeat steps 3 and 4 until the fitness function reaches the preset accuracy or the number of iterations reaches the maximum value, and output the optimal identification parameters.

[0015] In step 1, the current is defined as positive during battery charging and negative during battery discharging. Simultaneously, to correct the voltage prediction bias caused by the deterioration of the electrode-electrolyte interface contact state in traditional models under low charge conditions, a model incorporating dynamic interface contact resistance is constructed. The second-order equivalent circuit model, The dynamic impedance characterizes the interfacial contact state between the electrode active material and the electrolyte during battery charging and discharging as a function of state of charge (SOC) and current density. At low SOC, the utilization rate of the interfacial active material decreases. As the resistance increases, the voltage drop can compensate for the prediction bias of the model in this region, and its state equation is:

[0016] (1)

[0017] in: Open circuit voltage, For ohmic internal resistance, , Polarization resistor, For dynamic interface contact resistance, , Polarized capacitor, Terminal voltage, The current in the circuit, , The voltage across the parallel RC network.

[0018] In step 2, the goal of identifying the optimal parameters of the second-order equivalent circuit model is to obtain the optimal circuit parameters based on the dynamic operating condition experimental data, so as to minimize the error between the calculated terminal voltage value output by the model and the experimental measurement value.

[0019] In step 2, the lithium battery terminal voltage data measured under DST conditions is used as the true value y*, the predicted voltage output by the second-order equivalent circuit model is used as the estimated value y, and the mean absolute error (MAE) between the true value and the estimated value is used as the fitness function. Defined as: (2)

[0020] In the formula, The length of the data sequence. For time step, For the first Real-time measured terminal voltage For the first The time-based model predicts the terminal voltage, which is calculated by combining the second-order equivalent circuit model with the current parameters.

[0021] fitness function The smaller the value, the smaller the deviation between the model-predicted voltage and the measured voltage, and the better the parameter identification result. By improving the sand cat swarm optimization algorithm for iterative optimization, the... Reaching the minimum value yields the optimal combination of circuit parameters.

[0022] In step 3, a chaotic initialization strategy is used to generate a sand cat population, and the initial position is determined based on the physical constraints of the parameters to be identified. The specific process is as follows:

[0023] First, the parameters to be identified are second-order RC model parameters, including ohmic internal resistance. Polarization resistance , Polarized capacitors , First, determine the search boundaries for these parameters, then use chaotic mapping to generate the initial sand cat individual positions, as shown in the formula:

[0024] (3)

[0025] in, Indicates the first The number of individual sand cats in the first Initial position on the dimension parameter, The first Lower and upper bounds of the dimensional parameter, To use Logistic chaotic mapping The generated chaotic sequence. By utilizing the ergodicity and randomness of the chaotic sequence, it is ensured that the initial population is uniformly distributed in the solution space, avoiding search limitations caused by the concentration of initial positions, and laying the foundation for subsequent global exploration.

[0026] Then, for each sand cat individual generated by chaotic initialization, the fitness function defined in step 2 is substituted to calculate its corresponding fitness value. The initial population is sorted according to the fitness value, and the position of the individual with the best current fitness is marked as the initial "optimal solution candidate". This provides a reference benchmark for subsequent dynamic weight adjustment and adaptive mutation strategy, enabling the algorithm to focus on high-potential search areas in the initial stage.

[0027] Finally, based on the above chaotic initialization of the population, the key parameters of the algorithm are set as follows: the population size is a preset value, and the maximum number of iterations is a preset value; at the same time, the initial weight coefficients in the dynamic weight perturbation mechanism are initialized (e.g., ...). =0.9), the initial mutation probability in the adaptive mutation strategy, and the trigger interval of the triangle walking strategy are defined to ensure that each improved strategy can work together according to the preset logic during the iteration process, thereby enhancing the algorithm's ability to balance global search and local development.

[0028] In step 4, the location of the sand cat population is iteratively updated by combining a dynamic weight perturbation mechanism, an adaptive mutation strategy, and a triangle walking strategy. A greedy selection process is used to retain high-quality solutions. The specific process is as follows: In each iteration, based on the current iteration number... Dynamically adjust weighting factors And a random perturbation term and Levy flight step size are introduced to update the individual sand cat position, the formula is: (4)

[0029] in, For the first The generation The position vector of each individual , These are the lower and upper bound vectors of the parameters, respectively. For random disturbance terms; The inertia weights are linearly decreasing. This represents the current iteration number. (Maximum number of iterations) This is the scaling factor; Let Levy's flight random step size vector be ( ), This represents the dot product operation.

[0030] Levy flight enhances global exploration capabilities by generating random step sizes with heavy-tailed characteristics, which follow a Levy distribution:

[0031] (5)

[0032] in , ,and This distribution characteristic enhances global exploration capabilities while effectively preventing premature convergence.

[0033] Then, the mutation probability is dynamically adjusted based on the individual fitness value, and high-probability mutations are applied to individuals with poor fitness. The formula is as follows:

[0034] (6)

[0035] in This represents the individual fitness value. , These are the best and worst fitness values ​​for the current population, respectively. To prevent extremely small constants with a denominator of zero, The baseline mutation probability. When random numbers At that time, perform Gaussian-Cauchy mixed mutation:

[0036] (7)

[0037] in To determine the current optimal individual, a hybrid mutation is used to balance local fine-grained search and global jump capability.

[0038] After that, every interval In the next iteration, the triangle traversal update strategy is executed:

[0039] (8)

[0040] in , For two randomly selected different individuals , By constructing a triangular search path, we can break through the local optimum trap by utilizing multi-directional information interaction.

[0041] Next, the fitness value of the updated individual is calculated. If the fitness of the new position is better than that of the original position, the update is accepted; otherwise, the original position is retained. The formula is:

[0042] (9)

[0043] At the same time, update the globally optimal individual. This ensures that high-quality solutions are not lost.

[0044] Finally, for all updated individuals, check whether they satisfy the parameter boundary constraints. If they exceed the boundaries, correct them using a reflection strategy.

[0045] (10)

[0046] This strategy reflects out-of-bounds parameters back into the feasible region, avoiding information loss caused by simple truncation. Through this improved strategy, the algorithm can maintain population diversity during iteration, gradually converge to the global optimum, and ensure the physical feasibility of all parameters.

[0047] In step 5, steps 3 and 4 are repeated. Specifically, starting with the chaotic initial population generated in step 3, the individual sand cat positions are updated generation by generation according to the dynamic weight perturbation mechanism, adaptive mutation strategy, triangle walking strategy, and greedy selection rule defined in step 4. In each iteration, the current optimal position is selected by calculating the individual fitness value, and the search direction for the next iteration is adjusted based on this, so that the population gradually gathers towards the potential optimal solution region.

[0048] During the iteration process, two termination conditions are monitored in real time: if the current population's optimal fitness value is... If the model's prediction accuracy is less than the preset expected value, meaning the model's prediction accuracy meets the engineering requirements, the algorithm terminates early. If the number of iterations reaches the maximum number of iterations set in step 3, the algorithm terminates regardless of whether the current fitness value meets the standard to avoid overcomputation.

[0049] When any of the above termination conditions are met, the iteration stops, and the parameter combination corresponding to the individual with the smallest fitness value in the current population is output. This parameter combination is the optimal identification result of the second-order equivalent circuit model of the lithium battery, which can be directly used for battery state prediction and performance analysis.

[0050] This invention generates a sand cat population through a chaotic initialization strategy. The ergodicity and randomness of the chaotic sequence enhance the uniformity of the initial population distribution in the solution space, avoiding search bias caused by initial position limitations and laying the foundation for global optimization. During iteration, a dynamic weight perturbation mechanism adjusts the search step size in real time, balancing global exploration and local development capabilities to accelerate convergence. An adaptive mutation strategy dynamically adjusts the mutation intensity based on individual fitness, enhancing local search accuracy while effectively preventing the population from getting trapped in local optima. A triangle-walking strategy expands the search path through multi-directional information interaction, further improving the algorithm's coverage of complex solution spaces. Finally, a greedy selection and constraint handling mechanism ensures that the optimization process always focuses on high-potential regions within the feasible domain.

[0051] Applying this improved SCSO algorithm to lithium battery parameter identification can accurately optimize key parameters such as ohmic internal resistance, polarization resistance, polarization capacitance, and voltage bias in the second-order equivalent circuit model, significantly reducing the mean square error between the model's predicted voltage and the measured voltage, especially performing better in regions with significant nonlinearity such as low SOC. Compared with conventional algorithms, this invention improves the accuracy and robustness of parameter identification, providing reliable parameter support for lithium battery state estimation, which helps extend battery life and ensure the safe and stable operation of new energy vehicles and energy storage systems, and has significant engineering application value. Attached Figure Description

[0052] Figure 1 This is a flowchart of the parameter identification method for equivalent model of lithium iron phosphate batteries based on the improved SCSO.

[0053] Figure 2 This is the iterative update graph of the SCSO method.

[0054] Figure 3 This is the equivalent circuit diagram of a battery;

[0055] Figure 4 This is a graph showing the fitting of the SOC and OCV functions;

[0056] Figure 5 This is a current-voltage diagram of a lithium battery under DST conditions.

[0057] Figure 6 Parameter identification results diagram;

[0058] Figure 7 These are the mean absolute error (MAE) and the SCSO voltage error. Detailed Implementation

[0059] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0060] like Figure 1 As shown, a method for identifying equivalent model parameters of lithium iron phosphate batteries based on improved SCSO includes the following steps:

[0061] Step 1: Construct a second-order equivalent circuit model and determine the parameters to be identified;

[0062] The current is specified to be positive during battery charging and negative during battery discharging. To correct voltage prediction errors in traditional models caused by deterioration of the electrode-electrolyte interface contact under low charge conditions, a model incorporating dynamic interface contact resistance is constructed. The second-order equivalent circuit model, such as Figure 2 As shown. The dynamic impedance characterizes the interfacial contact state between the electrode active material and the electrolyte during battery charging and discharging as a function of state of charge (SOC) and current density. At low SOC, the utilization rate of the interfacial active material decreases. As the resistance increases, the voltage drop can compensate for the prediction bias of the model in this region, and its state equation is:

[0063] (1)

[0064] in: Open circuit voltage, For ohmic internal resistance, , Polarization resistor, For dynamic interface contact resistance, , Polarized capacitor, Terminal voltage, The current in the circuit, , The voltage across the parallel RC network.

[0065] Step 2: Collect current, voltage, and time data of lithium battery under dynamic operating conditions, and construct a fitness function based on the deviation between the model-predicted voltage and the measured voltage.

[0066] Using the lithium battery terminal voltage data measured under DST conditions as the true value y*, and the predicted voltage output by the second-order equivalent circuit model as the estimated value y, the average absolute error between the true value and the estimated value is used as the fitness function. Defined as:

[0067] (2)

[0068] In the formula, The length of the data sequence. For time step, For the first Real-time measured terminal voltage For the first The time-based model predicts the terminal voltage, which is calculated by combining the second-order equivalent circuit model with the current parameters.

[0069] fitness function The smaller the value, the smaller the deviation between the model-predicted voltage and the measured voltage, and the better the parameter identification result. By improving the sand cat swarm optimization algorithm for iterative optimization, the... Reaching the minimum value means obtaining the optimal combination of circuit parameters.

[0070] Step 3: Use a chaotic initialization strategy to generate the initial population of the improved sand cat swarm optimization algorithm, and set the search boundary in combination with parameter physical constraints;

[0071] First, the parameters to be identified are the parameters of the second-order equivalent circuit model, including the ohmic internal resistance. Polarization resistance , Polarized capacitors , First, determine the search boundaries for these parameters, then use chaotic mapping to generate the initial sand cat individual positions, as shown in the formula:

[0072] (3)

[0073] in, Indicates the first The number of individual sand cats in the first Initial position on the dimension parameter, , The first Lower and upper bounds of the dimensional parameter, To use Logistic chaotic mapping The generated chaotic sequence;

[0074] Then, for each sand cat individual generated by the chaotic initialization, substitute it into the fitness function defined in step 2 to calculate its corresponding fitness value; sort the initial population according to the fitness value, and mark the position of the individual with the best current fitness as the candidate for the initial optimal solution;

[0075] Finally, based on the above chaotic initialization population, the key parameters of the algorithm are set: the population size is a preset value, and the maximum number of iterations is a preset value; at the same time, the initial weight coefficients in the dynamic weight perturbation mechanism and the initial mutation probability in the adaptive mutation strategy are initialized, and the trigger interval of the triangle walking strategy is defined to ensure that each improvement strategy can work together according to the preset logic during the iteration process, thereby enhancing the algorithm's ability to balance global search and local development.

[0076] Step 4: Introduce a dynamic weight perturbation mechanism, an adaptive mutation strategy, and a triangle walking strategy to iteratively update the position of the sand cat population, and retain high-quality solutions through greedy selection;

[0077] like Figure 3 As shown, the final and key parameters controlling the transition between the exploration and development phases are: ,when When the sensitivity is greater than 1, the sand cat searches for prey. The search relies on the release of low-frequency noise, and each sand cat updates its position based on the best candidate location, its current position, and its sensitivity range. Therefore, the sand cat is able to find other possible optimal prey locations. When the sensitivity level is ≤1, the sand cat attacks its prey. First, a random position is generated using the optimal position and the current position. Assuming the sand cat's sensitivity range is a circle, an angle is randomly selected for each sand cat using a roulette wheel method. Then it attacks its prey. The random position ensures the sand cat gets close to the prey, and the random angle prevents the algorithm from getting stuck in local optima.

[0078] In each iteration, based on the current iteration number Dynamically adjust weighting factors And a random perturbation term and Levy flight step size are introduced to update the individual sand cat position, the formula is:

[0079] (4)

[0080] in, For the first The generation The position vector of each individual , These are the lower and upper bound vectors of the parameters, respectively. For random disturbance terms; The inertia weights are linearly decreasing. This represents the current iteration number. The maximum number of iterations, This is the scaling factor; Let Levy's flight random step size vector be . , This represents the dot product operation;

[0081] Levy flight enhances global exploration capabilities by generating random step sizes with heavy-tailed characteristics, which follow a Levy distribution:

[0082] (5)

[0083] in , ,and ;

[0084] Then, the mutation probability is dynamically adjusted based on the individual fitness value, and high-probability mutations are applied to individuals with poor fitness. The formula is as follows:

[0085] (6)

[0086] in This represents the individual fitness value. , These are the best and worst fitness values ​​for the current population, respectively. To prevent extremely small constants with a denominator of zero, As the baseline mutation probability, when random numbers At that time, perform Gaussian-Cauchy mixed mutation:

[0087] (7)

[0088] in To determine the current optimal individual, a hybrid mutation is used to balance local fine-grained search and global jump capability;

[0089] After that, every interval In the next iteration, the triangle traversal update strategy is executed:

[0090] (8)

[0091] in , For two randomly selected different individuals , By constructing a triangular search path, multi-directional information interaction is utilized to break through the local optimum trap;

[0092] The fitness value of the updated individual is then calculated. If the fitness of the new position is better than that of the original position, the update is accepted; otherwise, the original position is retained. The formula is as follows:

[0093] (9)

[0094] At the same time, update the globally optimal individual. To ensure that high-quality solutions are not lost;

[0095] Finally, for all updated individuals, check whether they satisfy the parameter boundary constraints. If they exceed the boundaries, correct them using a reflection strategy.

[0096] (10).

[0097] Step 5: Repeat steps 3 and 4 until the fitness function reaches the preset accuracy or the number of iterations reaches the maximum value, and output the optimal identification parameters.

[0098] Starting with the chaotic initial population generated in step 3, the individual positions of sand cats are updated generation by generation according to the dynamic weight perturbation mechanism, adaptive mutation strategy, triangle walking strategy and greedy selection rule defined in step 4. In each iteration, the current optimal position is selected by calculating the individual fitness value, and the search direction of the next iteration is adjusted based on this, so that the population gradually gathers towards the potential optimal solution region.

[0099] During the iteration process, two termination conditions are monitored in real time: if the current population's optimal fitness value is... If the model's prediction accuracy is less than the preset expected value, meaning the model's prediction accuracy meets the engineering requirements, the algorithm terminates early. If the number of iterations reaches the maximum number of iterations set in step 3, the algorithm terminates regardless of whether the current fitness value meets the target to avoid overcomputation.

[0100] When any of the above termination conditions are met, the iteration stops, and the parameter combination corresponding to the individual with the smallest fitness value in the current population is output. This parameter combination is the optimal identification result of the second-order equivalent circuit model of the lithium battery, which can be directly used for battery state prediction and performance analysis. Example

[0101] This embodiment uses the A123 lithium iron phosphate battery, which has extremely high safety and ultra-long cycle life. Its excellent cycle life, superior thermal stability, and safety performance provide a highly reliable data foundation for algorithm verification, with very good consistency. The specifications of the A123 battery are shown in Table 1.

[0102]

[0103] Table 1 A123 Battery Specifications

[0104] The DST cycle, as a standard test procedure for evaluating the dynamic response characteristics of power batteries, simulates the complex current fluctuations during actual vehicle operation, comprehensively stimulating the battery's ohmic voltage drop, polarization effect, and recovery characteristics. The data sample selected in this invention covers complete DST cycle test data of batteries at an ambient temperature of 25°C. The visualization results are as follows... Figure 4 As shown.

[0105] Furthermore, this embodiment performs charge-discharge tests on the battery according to the following steps to obtain the SOC-OCV relationship, such as... Figure 5 As shown.

[0106] 1. Charge the battery to the cutoff voltage using a constant current at a 1C rate;

[0107] 2. Charge at a constant voltage until the current drops to 0.01C;

[0108] 3. Discharge at a constant rate of C / 20 until the voltage drops to the cutoff voltage;

[0109] 4. Charge at a constant rate of C / 20 until fully charged to the cutoff voltage.

[0110] After improving the SCSO algorithm, the following parameter identification results were obtained, such as... Figure 6 As shown.

[0111] As can be seen, although the voltage error of the improved SCSO algorithm increases slightly over time, the overall error is still very small, with a mean absolute error (MAE) of only 0.068786.

[0112] Since the essence of parameter identification lies in constructing a reasonable mathematical model and applying optimization algorithms based on the system's input and output data or some observable information, the results are within an acceptable range.

[0113] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for identifying equivalent model parameters of lithium iron phosphate batteries based on improved SCSO, characterized in that, Includes the following steps: Step 1: Construct a second-order equivalent circuit model and determine the parameters to be identified; Step 2: Collect current, voltage, and time data of lithium battery under dynamic operating conditions, and construct a fitness function based on the deviation between the model-predicted voltage and the measured voltage. Step 3: Use a chaotic initialization strategy to generate the initial population of the improved sand cat swarm optimization algorithm, and set the search boundary in combination with parameter physical constraints; Step 4: Introduce a dynamic weight perturbation mechanism, an adaptive mutation strategy, and a triangle walking strategy to iteratively update the position of the sand cat population, and retain high-quality solutions through greedy selection; The specific steps of step 4 are as follows: In each iteration, based on the current iteration number Dynamically adjust weighting factors And a random perturbation term and Levy flight step size are introduced to update the individual sand cat position, the formula is: (4) in, For the first The generation The position vector of each individual , These are the lower and upper bound vectors of the parameters, respectively. For random disturbance terms; The inertia weights are linearly decreasing. This represents the current iteration number. The maximum number of iterations, This is the scaling factor; Let Levy's random step size vector be the flight path. , This represents the dot product operation; Levy flight enhances global exploration capabilities by generating random step sizes with heavy-tailed characteristics, which follow a Levy distribution: (5) in , ,and ; Then, the mutation probability is dynamically adjusted based on the individual fitness value, and high-probability mutations are applied to individuals with poor fitness. The formula is as follows: (6) in This represents the individual fitness value. , These are the best and worst fitness values ​​for the current population, respectively. To prevent extremely small constants with a denominator of zero, As the baseline mutation probability, when random numbers At that time, perform Gaussian-Cauchy mixed mutation: (7) in To determine the current optimal individual, a hybrid mutation is used to balance local fine-grained search and global jump capability; After that, every interval In the next iteration, the triangle traversal update strategy is executed: (8) in , For two randomly selected different individuals , By constructing a triangular search path, the local optimum trap can be broken through by utilizing multi-directional information interaction; The fitness value of the updated individual is then calculated. If the fitness of the new position is better than that of the original position, the update is accepted; otherwise, the original position is retained. The formula is as follows: (9) At the same time, update the globally optimal individual. To ensure that high-quality solutions are not lost; Finally, for all updated individuals, check whether they satisfy the parameter boundary constraints. If they exceed the boundaries, correct them using a reflection strategy. (10) Step 5: Repeat steps 3 and 4 until the fitness function reaches the preset accuracy or the number of iterations reaches the maximum value, and output the optimal identification parameters.

2. The method for identifying equivalent model parameters of lithium iron phosphate batteries based on improved SCSO according to claim 1, characterized in that, The method for constructing the second-order equivalent circuit model in step 1 is as follows: The current is specified to be positive during battery charging and negative during battery discharging. To correct voltage prediction errors in traditional models caused by deterioration of the electrode-electrolyte interface contact under low charge conditions, a model incorporating dynamic interface contact resistance is constructed. The second-order equivalent circuit model, Characterizing the contact state between the electrode active material and the electrolyte interface during battery charging and discharging. Dynamic impedance to changes in current density, low The utilization rate of surfactants decreases over time. As the resistance increases, the voltage drop can compensate for the prediction bias of the model in this region, and its state equation is: in: Open circuit voltage, For ohmic internal resistance, , Polarization resistor, For dynamic interface contact resistance, , Polarized capacitor, Terminal voltage, The current in the circuit, , The voltage across the parallel RC network.

3. The method for identifying equivalent model parameters of lithium iron phosphate batteries based on improved SCSO according to claim 2, characterized in that, The method for constructing the fitness function in step 2 is as follows: Using the lithium battery terminal voltage data measured under DST conditions as the true value y*, and the predicted voltage output by the second-order equivalent circuit model as the estimated value y, the average absolute error between the true value and the estimated value is used as the fitness function. Defined as: (2) In the formula, The length of the data sequence. For time step, For the first Real-time measured terminal voltage For the first The time-based model predicts the terminal voltage, which is calculated by combining the second-order equivalent circuit model with the current parameters. fitness function The smaller the value, the smaller the deviation between the model-predicted voltage and the measured voltage, and the better the parameter identification result. By improving the sand cat swarm optimization algorithm for iterative optimization, the... Reaching the minimum value means obtaining the optimal combination of circuit parameters.

4. The method for identifying equivalent model parameters of lithium iron phosphate batteries based on improved SCSO according to claim 2, characterized in that, The specific steps of step 3 are as follows: First, the parameters to be identified are the parameters of the second-order equivalent circuit model, including the ohmic internal resistance. Polarization resistance , Polarized capacitors , First, determine the search boundaries for these parameters, then use chaotic mapping to generate the initial sand cat individual positions, as shown in the formula: (3) in, Indicates the first The number of individual sand cats in the first Initial position on the dimension parameter, , The first Lower and upper bounds of the dimensional parameter, To use Logistic chaotic mapping The generated chaotic sequence; Then, for each sand cat individual generated by the chaotic initialization, substitute it into the fitness function defined in step 2 to calculate its corresponding fitness value; sort the initial population according to the fitness value, and mark the position of the individual with the best current fitness as the candidate for the initial optimal solution; Finally, based on the above chaotic initialization population, the key parameters of the algorithm are set: the population size is a preset value, and the maximum number of iterations is a preset value; at the same time, the initial weight coefficients in the dynamic weight perturbation mechanism and the initial mutation probability in the adaptive mutation strategy are initialized, and the trigger interval of the triangle walking strategy is defined to ensure that each improvement strategy can work together according to the preset logic during the iteration process, thereby enhancing the algorithm's ability to balance global search and local development.

5. The method for identifying equivalent model parameters of lithium iron phosphate batteries based on improved SCSO according to claim 3, characterized in that, The specific steps of step 5 are as follows: Starting with the chaotic initial population generated in step 3, the individual positions of sand cats are updated generation by generation according to the dynamic weight perturbation mechanism, adaptive mutation strategy, triangle walking strategy and greedy selection rule defined in step 4. In each iteration, the current optimal position is selected by calculating the individual fitness value, and the search direction of the next iteration is adjusted based on this, so that the population gradually gathers towards the potential optimal solution region. During the iteration process, two termination conditions are monitored in real time: if the current population's optimal fitness value is... If the model's prediction accuracy is less than the preset expected value, meaning the model's prediction accuracy meets the engineering requirements, the algorithm terminates early. If the number of iterations reaches the maximum number of iterations set in step 3, the algorithm terminates regardless of whether the current fitness value meets the target to avoid overcomputation. When any of the above termination conditions are met, the iteration stops, and the parameter combination corresponding to the individual with the smallest fitness value in the current population is output. This parameter combination is the optimal identification result of the second-order equivalent circuit model of the lithium battery, which can be directly used for battery state prediction and performance analysis.

Images