Battery residual capacity estimation method and device, equipment and storage medium
By employing a hybrid resampling and weighted calculation method, the problem of insufficient particle diversity in particle filtering is solved, thereby improving the accuracy and reliability of battery capacity estimation and achieving more accurate estimation of remaining battery capacity.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XIAN ACTIONPOWER ELECTRIC
- Filing Date
- 2025-12-02
- Publication Date
- 2026-04-17
AI Technical Summary
Traditional particle filtering suffers from a lack of particle diversity in battery capacity estimation, leading to reduced estimation accuracy and failing to accurately reflect the actual battery capacity.
By employing a hybrid resampling and weighted calculation method, particles are first extracted from the prior distribution based on the battery capacity decay characteristics to form an initial particle set. The particle weights are then evaluated and divided into high-weight and low-weight sets. A hybrid resampling operation is then performed to generate a new particle set, which is then validated by combining high-weight particles. Finally, a weighted calculation is used to obtain an estimate of the remaining battery capacity.
It improves the accuracy and reliability of battery remaining capacity estimation, solves the problem of insufficient particle diversity, and ensures the quality of new particle sets and the accuracy of state vector estimation.
Smart Images
Figure CN121878477A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of battery technology, and more specifically, to a method, apparatus, device, and storage medium for estimating the remaining capacity of a battery. Background Technology
[0002] As a core component of new energy vehicles, the battery's capacity directly determines the vehicle's driving range. With each charge-discharge cycle, the battery capacity gradually decreases, leading to a discrepancy between the actual stored capacity and the nominal capacity. This makes estimating the remaining capacity and driving range quite difficult, thus requiring accurate estimation of the battery's actual capacity. Particle filtering is generally used, leveraging its strong estimation capabilities for nonlinear and non-Gaussian systems to estimate battery capacity.
[0003] In related technologies, traditional particle filtering suffers from a lack of particle diversity during the resampling process. Specifically, some high-weight particles are oversampled, while low-weight particles are eliminated, causing the particle set to gradually converge, which in turn reduces the accuracy of capacity estimation and results in a large error between the estimated remaining battery capacity and the actual capacity. Summary of the Invention
[0004] The problem addressed by this invention is how to improve the accuracy of battery capacity estimation.
[0005] To address the above problems, the present invention provides a method, apparatus, device, and storage medium for estimating the remaining capacity of a battery.
[0006] In a first aspect, the present invention provides a method for estimating the remaining capacity of a battery, comprising: Based on the degradation model corresponding to the battery capacity decay characteristics, a preset number of particles are extracted from the prior distribution to obtain an initial particle set. The degradation model contains a state vector that reflects the key characteristics of capacity decay. The weight of each particle in the initial particle set is evaluated to obtain a set of particles with high weight and a set of particles with low weight. Perform a hybrid resampling operation on the low-weight particle set to generate new particles, and combine the new particles with the high-weight particle set to obtain a new particle set; The estimated value of the state vector of the degenerate model is determined by weighted calculation based on the weights of the new particles in the new particle set. Substituting the estimated value of the state vector into the degradation model, we obtain the estimated value of the remaining battery capacity at the preset time.
[0007] Optionally, a predetermined number of particles are extracted from the prior distribution based on the degradation model corresponding to the battery capacity decay characteristics to obtain an initial particle set, including: By combining the battery's factory-specified parameters with the capacity data obtained from previous charge-discharge tests, a reasonable range of values for the state vector is determined. The prior distribution of the parameter distribution characteristics of the matching state vector is constructed based on a reasonable value range; A predetermined number of particles are extracted from the prior distribution, and an initial particle set is formed based on all the particles; wherein the dimension of the particles is the same as the dimension of the state vector.
[0008] Optionally, the weight of each particle in the initial particle set is evaluated to obtain a high-weight particle set and a low-weight particle set, including: Based on the observation equations of the degradation model, determine the battery capacity measurement at a preset time. Substitute each particle in the initial particle set into the degradation model to calculate the theoretical capacity estimate of each particle at a preset time. Based on the deviation between the theoretical capacity estimate and the capacity measurement, and combined with the noise characteristics of the degradation model, the weight of each particle is determined. All particles are weighted according to their weights, resulting in a set of high-weight particles and a set of low-weight particles.
[0009] Optionally, all particles are weighted according to their weights to obtain a set of high-weight particles and a set of low-weight particles, including: The particles are divided according to the relationship between their weights and the weight division threshold. Among them, particles with weights greater than or equal to the weight division threshold are assigned to the high-weight particle set; Particles with weights less than the weight partitioning threshold are assigned to a low-weight particle set.
[0010] Optionally, the hybrid resampling operation includes local optimization sampling and global optimization sampling. The hybrid resampling operation is performed on the low-weight particle set to generate new particles, and the new particles are combined with the high-weight particle set to obtain a new particle set, including: Based on random numbers within a preset numerical range, select to trigger either local optimization sampling or global optimization sampling; In response to triggering local optimization sampling, reference particles are extracted from the high-weight particle set. Through Gaussian mutation, the particles in the low-weight particle set are optimized based on the parameter information of the reference particles to generate the first type of new particles to be verified. In response to triggering global optimization sampling, the maximum weight particle in the high-weight particle set and the minimum weight particle in the low-weight particle set are obtained; Based on the principle of seeking benefits and avoiding harm, the particles in the low-weight particle set are adjusted according to the parameter differences between the maximum and minimum weight particles to generate a second type of new particles to be verified. The average weight of the low-weight particle set is used as the verification threshold. The verification threshold is used to determine whether the weight of the first type or the second type of new particle to be verified meets the verification threshold requirement. New particles that meet the verification threshold requirements are merged with the set of high-weight particles to generate a new particle set.
[0011] Optionally, a weighted calculation is performed based on the weights of the new particles in the new particle set to determine the estimated value of the state vector of the degenerate model, including: The weights of all particles in the new particle set are normalized to obtain the normalized weights of the particles, where the sum of the normalized weights of all particles is 1. By performing a weighted summation operation on all particles based on their weights, an estimate of the state vector of the degenerate model is obtained.
[0012] Optionally, the estimated value of the state vector is substituted into the degradation model to obtain an estimated value of the remaining battery capacity at a preset time, including: Substitute the estimated value of the state vector into the degradation model to determine the parameter value of each parameter of the state vector; By using a double-exponential empirical formula, based on the parameter values and the time parameter at a preset time, the estimated value of the remaining battery capacity at a preset time is determined.
[0013] In a second aspect, the battery remaining capacity estimation device of the present invention includes: The data processing unit is used to extract a preset number of particles from the prior distribution according to the degradation model corresponding to the battery capacity decay characteristics to obtain an initial particle set. The degradation model contains a state vector that reflects the key characteristics of capacity decay. The evaluation unit is used to evaluate the weight of each particle in the initial particle set to obtain a high-weight particle set and a low-weight particle set. The hybrid resampling unit is used to perform a hybrid resampling operation on the low-weight particle set to generate new particles, and then combine the new particles with the high-weight particle set to obtain a new particle set. The computational unit is used to perform weighted calculations based on the weights of the new particles in the new particle set to determine the estimated value of the state vector of the degenerate model; The capacity estimation unit is used to substitute the estimated value of the state vector into the degradation model to obtain the estimated value of the remaining battery capacity at a preset time.
[0014] Thirdly, the electronic device of the present invention includes a memory and a processor; Memory, used to store computer programs; A processor, used to implement the battery remaining capacity estimation method described above when executing a computer program.
[0015] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the battery remaining capacity estimation method described above.
[0016] The battery remaining capacity estimation method, apparatus, device, and storage medium of the present invention first extracts a preset number of particles from a prior distribution based on the degradation model corresponding to the battery capacity decay characteristics to obtain an initial particle set. This ensures that the particle set can cover multiple possible states of battery capacity decay. By reasonably selecting the prior distribution and degradation model, the initial particle set can better reflect the key characteristics of battery capacity decay. The weight of each particle is determined by calculating the degree of matching between each particle and the actual observed data. High-weight particles are closer to the actual capacity decay state of the battery, while low-weight particles are relatively deviated from it. Through this classification, it can be clearly identified which particles are more representative in the current estimation. Furthermore, a hybrid resampling operation is performed on the low-weight particle set to generate new particles. The new particles are then combined with the high-weight particle set to obtain a new particle set. This effectively solves the problem of insufficient particle diversity in traditional particle filtering. The hybrid resampling operation increases the diversity of the particle set by reintroducing low-weight particles and avoids particle convergence. Simultaneously, verification is performed using a high-weight particle set to ensure the quality of the new particle set. This retains the accuracy of high-weight particles while introducing more state space coverage through resampling of low-weight particles. This invention, through the combination of hybrid resampling and verification, enables the new particle set to more comprehensively reflect the complex characteristics of battery capacity decay. Then, using the weight of each particle in the new particle set, a weighted average is used to calculate the estimated value of the state vector. Because the new particle set has undergone hybrid resampling and verification, its weight distribution is more reasonable, more accurately reflecting the actual state of battery capacity decay. This weighted calculation method not only improves estimation accuracy but also reduces error accumulation caused by unreasonable particle weight distribution, making the estimated value of the state vector closer to the actual value. Finally, the estimated value of the state vector is transformed into a specific capacity estimate through a degradation model. Hybrid resampling and weighted calculation ensure the accuracy and reliability of the state vector estimate, and the final capacity estimate more accurately reflects the actual capacity of the battery at a preset time.
[0017] This invention achieves end-to-end optimization, from increasing particle diversity to improving estimation accuracy. The hybrid resampling operation not only solves the problem of insufficient particle diversity but also ensures the quality of the new particle set through particle ensemble analysis. Weighted computation further improves the accuracy of state vector estimation, and the final capacity estimate obtained through the degradation model more reliably reflects the actual battery capacity, significantly improving the accuracy of remaining battery capacity estimation. Attached Figure Description
[0018] Figure 1 This is one of the flowcharts illustrating the battery remaining capacity estimation method according to an embodiment of the present invention; Figure 2 This is a second schematic flowchart of the battery remaining capacity estimation method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the actual capacity decay curve of the lithium battery according to an embodiment of the present invention; Figure 4 This is a comparative diagram showing the estimation results of lithium battery capacity in an embodiment of the present invention; Figure 5 This is a schematic diagram of the battery remaining capacity estimation device according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Detailed Implementation
[0019] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Although some embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the accompanying drawings and embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.
[0020] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.
[0021] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to"; the term "based on" means "at least partially based on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments"; and the term "optionally" means "optional embodiments". Definitions of other terms will be given in the following description. It should be noted that the concepts of "first," "second," etc., mentioned in this invention are used only to distinguish different devices, modules, or units, and are not intended to limit the order of functions performed by these devices, modules, or units or their interdependencies.
[0022] It should be noted that the terms "one" and "more" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".
[0023] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, data stored, data displayed, etc.) and signals involved in this application are all authorized by the user or fully authorized by all parties. The collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.
[0024] Combination Figure 1 As shown in the figure, an embodiment of the present invention provides a method for estimating the remaining capacity of a battery, comprising: Based on the degradation model corresponding to the battery capacity decay characteristics, a preset number of particles are extracted from the prior distribution to obtain an initial particle set. The degradation model contains a state vector that reflects the key characteristics of capacity decay.
[0025] Specifically, firstly, a degradation model reflecting the battery capacity decay characteristics is established. This model describes the key characteristics of battery capacity decay through a state vector, which includes parameters related to capacity decay, such as changes in battery internal resistance and capacity retention rate. Simultaneously, to estimate the remaining battery capacity, a certain number of particles are drawn from a prior distribution. This prior distribution is a probability distribution set based on a preliminary understanding and assumptions about the battery capacity decay characteristics. For example, the number of particles drawn can be set according to actual needs and computational resources; for instance, 100 or more particles can be drawn, each representing a possible state of battery capacity decay. These particles combine to form an initial particle set.
[0026] The weight of each particle in the initial particle set is evaluated to obtain a high-weight particle set and a low-weight particle set.
[0027] After obtaining the initial particle set, the weight of each particle is evaluated, reflecting the degree of matching between the particle and the actual observed data. In this optional embodiment, the weight is calculated by comparing the state represented by the particle with the actual observed battery capacity data. Particles with higher weights indicate that their state is closer to the actual observed data, while particles with lower weights indicate that their state is less closely matched. Based on the magnitude of the weights, the particles in the initial particle set are divided into a high-weight particle set and a low-weight particle set. The high-weight particle set contains particles with higher weights, which better reflect the actual state of the battery; the low-weight particle set contains particles with lower weights, which require further adjustment to better match the actual state.
[0028] Perform a hybrid resampling operation on the low-weight particle set to generate new particles, and combine the new particles with the high-weight particle set to obtain a new particle set.
[0029] Specifically, firstly, each particle is resampled from the set of low-weight particles, i.e., a random number between [0, 1] is generated. If random numbers Less than or equal to the threshold Then, either local optimization sampling or global optimization strategies are used to generate new particles. Local optimization sampling generates new particles using a Gaussian distribution, while global optimization adjusts the distribution based on the difference between high-weight particles and low-weight particles. If Greater than If the original particles are retained unchanged, then the newly generated particles through resampling are merged with the set of high-weight particles to form a new particle set. This ensures that the new particle set contains both optimized low-weight particles and retained high-weight particles, thereby improving the overall quality of the particle set and the accuracy of the estimation. Finally, based on the new particle set and the current observation data, the weight of each particle is recalculated. This step ensures that the particle weights reflect the degree of matching between the particles and the latest observation data, providing more accurate information for subsequent estimation processes. Through this hybrid resampling process, the particle filtering algorithm can effectively improve the diversity and representativeness of the particle set, thereby improving the accuracy and reliability of estimating the remaining capacity of lithium batteries.
[0030] The mean and variance of the Gaussian noise in this embodiment can be set according to actual needs, for example, the mean is 0 and the variance is 0.1. Then, according to the principle of seeking the best and avoiding the worst, a particle with the largest weight is selected from the set of high-weight particles as the reference particle, and then each particle in the set of low-weight particles is adjusted so that its state vector is closer to the state vector of the reference particle.
[0031] In a preferred embodiment of the present invention, to avoid excessive particle concentration, a random perturbation term can be introduced. The new particles generated after the mixing and resampling operation are merged with the set of high-weight particles to form a new particle set. During the merging process, the weights of the new particles are re-evaluated to ensure the quality of the new particle set.
[0032] The estimated state vector of the degradation model is determined by weighting the new particles in the new particle set. Substituting this estimated state vector into the degradation model yields an estimated remaining battery capacity at a preset time.
[0033] Specifically, the state vectors of the particles are weighted and averaged using the weights of each particle in the new particle set. Particles with higher weights contribute more to the result, and the final state vector estimate more accurately reflects the current state of the battery. The state vector estimate is then substituted into the degradation model, and the remaining battery capacity estimate at a preset time is calculated according to the model formula. This remaining capacity estimate is used to assess the battery, such as the health status and remaining lifespan of a lithium battery.
[0034] The battery remaining capacity estimation method in this embodiment first extracts a preset number of particles from the prior distribution based on the degradation model corresponding to the battery capacity decay characteristics, obtaining an initial particle set. This ensures that the particle set can cover multiple possible states of battery capacity decay. By reasonably selecting the prior distribution and degradation model, the initial particle set can better reflect the key characteristics of battery capacity decay. The weight of each particle is determined by calculating the degree of matching between each particle and the actual observed data. High-weight particles are closer to the actual capacity decay state of the battery, while low-weight particles are relatively deviated from it. Through this classification, it can be clearly identified which particles are more representative in the current estimation. Furthermore, a hybrid resampling operation is performed on the low-weight particle set to generate new particles. The new particles are then combined with the high-weight particle set to obtain a new particle set, thus effectively solving the problem of insufficient particle diversity in traditional particle filtering. The hybrid resampling operation increases the diversity of the particle set by reintroducing low-weight particles, avoiding particle convergence. At the same time, verification is performed using the high-weight particle set to ensure the quality of the new particle set, preserving the accuracy of high-weight particles while introducing more state space coverage through the resampling of low-weight particles. This invention combines hybrid resampling and verification to enable a new particle ensemble that more comprehensively reflects the complex characteristics of battery capacity decay. Then, using the weights of each particle in the new ensemble, a weighted average is used to calculate the estimated state vector. Because the new particle ensemble has undergone hybrid resampling and verification, its weight distribution is more reasonable, more accurately reflecting the actual state of battery capacity decay. This weighted calculation method not only improves estimation accuracy but also reduces error accumulation caused by unreasonable particle weight distribution, making the estimated state vector closer to the actual value. Finally, a degradation model is used to transform the estimated state vector into a specific capacity estimate. Hybrid resampling and weighted calculation ensure the accuracy and reliability of the state vector estimate, and the final capacity estimate more accurately reflects the actual capacity of the battery at a preset time.
[0035] This embodiment achieves end-to-end optimization, from increasing particle diversity to improving estimation accuracy. The hybrid resampling operation not only solves the problem of insufficient particle diversity but also ensures the quality of the new particle set through particle ensemble analysis. Weighted computation further improves the accuracy of state vector estimation, and the final capacity estimate obtained through the degradation model more reliably reflects the actual battery capacity, significantly improving the accuracy of remaining battery capacity estimation.
[0036] Optionally, a predetermined number of particles are extracted from the prior distribution based on the degradation model corresponding to the battery capacity decay characteristics to obtain an initial particle set, including: By combining the battery's factory-specified parameters with the capacity data obtained from previous charge-discharge tests, a reasonable range of values for the state vector is determined. The prior distribution of the parameter distribution characteristics of the matching state vector is constructed based on a reasonable value range; A predetermined number of particles are extracted from the prior distribution, and an initial particle set is formed based on all the particles; wherein the dimension of the particles is the same as the dimension of the state vector.
[0037] Specifically, based on the battery's factory-specified parameters, such as initial capacity and internal resistance, and the capacity decay data obtained from previous charge-discharge tests, the range and variation patterns of these data are analyzed to determine the reasonable value ranges for each parameter in the state vector. For example, if the nominal initial capacity is 2Ah, and test data shows that its fluctuation range within a certain number of cycles is 1.9Ah to 2.1Ah, then the reasonable value range for the initial capacity can be determined to be 1.9Ah to 2.1Ah. Similarly, the value ranges for other state parameters are determined. Based on the determined reasonable value ranges, a prior distribution is constructed for each parameter in the state vector. The prior distribution should reflect the possible distribution characteristics of the parameter. For example, for the initial capacity, a normal distribution can be used to describe its probability distribution within the reasonable range; for the decay rate, if its change is relatively uniform, a uniform distribution can be used. In this way, a multidimensional prior distribution matching the characteristics of the state vector parameters is constructed. The number of particles to be extracted is set according to actual needs, such as 100 or more. A corresponding number of particles are randomly extracted from the constructed prior distribution, and the dimension of each particle is consistent with the dimension of the state vector. For example, if the state vector contains three parameters: initial capacity, decay rate, and internal resistance, then each particle should also contain values for these three parameters, and these values are all drawn from the prior distribution.
[0038] All the extracted particles are gathered together to form an initial particle set. Each particle in this set represents a possible state of battery capacity decay, providing the initial input for the subsequent particle filtering process.
[0039] In this optional embodiment, by combining the battery's factory-specified parameters with prior charge-discharge test data, a reasonable range of values for the state vector is accurately determined. Based on this, a matching prior distribution is constructed, enabling the extraction of an initial particle set highly correlated with the actual battery state from the prior distribution. This not only ensures the diversity and representativeness of the initial particles but also provides an accurate starting point for subsequent particle filtering estimation, effectively solving the problem of large estimation errors caused by unreasonable initial particle selection in traditional methods.
[0040] Optionally, the weight of each particle in the initial particle set is evaluated to obtain a high-weight particle set and a low-weight particle set, including: Based on the observation equations of the degradation model, determine the battery capacity measurement at a preset time. Substitute each particle in the initial particle set into the degradation model to calculate the theoretical capacity estimate of each particle at a preset time. Based on the deviation between the theoretical capacity estimate and the capacity measurement, and combined with the noise characteristics of the degradation model, the weight of each particle is determined. All particles are weighted according to their weights, resulting in a set of high-weight particles and a set of low-weight particles.
[0041] Specifically, using the observation equations of the degradation model and combining them with actual measurement data, the battery capacity measurement value at a preset time is determined. The observation equations relate the actual measured battery capacity to the model output, converting the actual measured value into the format required by the model and providing a benchmark for subsequent weight evaluation. Each particle in the initial particle set is substituted into the degradation model, and the theoretical capacity estimate of each particle at the preset time is calculated based on the model's dynamic equations and the observation equations. This embodiment simulates the capacity performance of the state represented by each particle at the preset time, providing a theoretical basis for weight evaluation. The deviation between the theoretical capacity estimate and the actual capacity measurement value of each particle is calculated, and combined with the noise characteristics of the degradation model, such as the distribution of observation noise, the weight of each particle is calculated using a probability density function. The weight reflects the degree of matching between the state represented by the particle and the actual observation data; the higher the weight, the closer the particle's state is to the true state. Based on the calculated weights, the particles in the initial particle set are divided into a high-weight particle set and a low-weight particle set. The high-weight particle set contains particles with higher weights, which better reflect the actual state of the battery; the low-weight particle set contains particles with lower weights, which may require further adjustment or optimization.
[0042] In this optional embodiment, the weight of each particle is accurately evaluated based on the actual measurement data of the battery and the theoretical estimates of the degradation model, and the particles are classified into high-weight and low-weight particle sets. This not only improves the quality of the particle sets and enhances the accuracy of the particle filtering algorithm in estimating the battery state, but also provides a basis for subsequent resampling operations.
[0043] Optionally, all particles are weighted according to their weights to obtain a set of high-weight particles and a set of low-weight particles, including: The particles are divided according to the relationship between their weights and the weight division threshold. Among them, particles with weights greater than or equal to the weight division threshold are assigned to the high-weight particle set; Particles with weights less than the weight partitioning threshold are assigned to a low-weight particle set.
[0044] Specifically, a weighting threshold is first determined to distinguish between high-weight particles and low-weight particles. In optional embodiments, the weighting threshold can be set based on the overall weight distribution of the particle set, such as the median or average of all weights in the particle set. Then, the weight of each particle is compared with this threshold to determine the particle classification. For each particle, if its weight is greater than or equal to the weighting threshold, it is classified into the high-weight particle set. Particles in the high-weight particle set are considered to better reflect the actual state of the battery; higher weights indicate a better match between the state and the actual observation data. For each particle, if its weight is less than the weighting threshold, it is classified into the low-weight particle set. Particles in the low-weight particle set are considered to have a lower match with the actual state of the battery and may require adjustment or optimization through subsequent resampling operations.
[0045] In this optional embodiment, by setting a weight division threshold and dividing the particles accordingly, the particle set can be quickly and effectively divided into high-weight and low-weight parts. This process not only simplifies the management of the particle set, but also provides a clear basis for subsequent resampling operations, enabling resampling to more effectively optimize the diversity and representativeness of the particle set. The high-weight particle set can be directly used in the subsequent estimation process, while the low-weight particle set is further improved through resampling operations, thereby improving the efficiency and accuracy of the entire particle filtering algorithm.
[0046] Optionally, the hybrid resampling operation includes local optimization sampling and global optimization sampling. The hybrid resampling operation is performed on the low-weight particle set to generate new particles, and the new particles are combined with the high-weight particle set to obtain a new particle set, including: Based on random numbers within a preset numerical range, select to trigger either local optimization sampling or global optimization sampling; In response to triggering local optimization sampling, reference particles are extracted from the high-weight particle set. Through Gaussian mutation, the particles in the low-weight particle set are optimized based on the parameter information of the reference particles to generate the first type of new particles to be verified. In response to triggering global optimization sampling, the maximum weight particle in the high-weight particle set and the minimum weight particle in the low-weight particle set are obtained; Based on the principle of seeking benefits and avoiding harm, the particles in the low-weight particle set are adjusted according to the parameter differences between the maximum and minimum weight particles to generate a second type of new particles to be verified. The average weight of the low-weight particle set is used as the verification threshold. The verification threshold is used to determine whether the weight of the first type or the second type of new particle to be verified meets the verification threshold requirement. New particles that meet the verification threshold requirements are merged with the set of high-weight particles to generate a new particle set.
[0047] Specifically, in the hybrid resampling operation, a preset numerical range is first generated, typically a random number within the range [0, 1]. Based on the value of this random number, it is determined whether to trigger local or global optimization sampling. For example, a threshold (such as 0.5) can be set; if the random number is less than this threshold, local optimization sampling is triggered; otherwise, global optimization sampling is triggered.
[0048] When local optimization sampling is triggered, a reference particle is randomly selected from the high-weight particle set. In one embodiment, the selection of the reference particle can be based on its weight, for example, selecting the particle with the largest weight. Then, the parameter information of the reference particle is used to optimize each particle in the low-weight particle set. A Gaussian mutation method is used to add a standard normal distribution random perturbation with a preset variance to each particle in the low-weight particle set, thereby achieving particle optimization and generating a first type of new particles to be verified. Specifically, firstly, the particle with the largest weight is selected from the high-weight particle set as the reference particle, and its state vector parameters will be used as the optimization target. For each particle in the low-weight particle set, an adjustment coefficient is calculated. This coefficient is dynamically determined based on the weight ratio of the reference particle and the low-weight particle, and is used to control the intensity of optimization. Subsequently, the parameters of the low-weight particles are linearly adjusted using this coefficient, gradually bringing them closer to the parameters of the reference particle, thus generating new particles.
[0049] When global optimization sampling is triggered, the particle with the largest weight is found from the high-weight particle set, and the particle with the smallest weight is found from the low-weight particle set. The parameter difference between these two particles is used for subsequent adjustment. Based on the principle of seeking advantages and avoiding disadvantages, the optimization direction is determined by the parameter difference between the particle with the largest and smallest weight. Two random numbers between [0,1] are introduced to adjust the optimization amplitude. The parameters of each particle in the low-weight particle set are adjusted to generate a second type of new particles to be verified and their weights are calculated. Specifically, firstly, the particle with the largest weight is determined from the high-weight particle set, and the particle with the smallest weight is found from the low-weight particle set. The parameter difference between these two particles is used as the adjustment benchmark. For each particle in the low-weight particle set, its parameter difference with that of the particle with the largest weight is calculated, and an adjustment coefficient is used to control the adjustment amplitude. The parameters of the low-weight particles gradually move closer to the parameters of the particle with the largest weight, thereby generating a new particle set.
[0050] In a preferred embodiment of the present invention, a certain random perturbation can be introduced during the adjustment process to maintain the diversity of particles, thereby generating a second type of new particles to be verified.
[0051] The average weights of all particles in the low-weight particle set are calculated and used as the verification threshold. Then, the weights of the first and second categories of new particles to be verified are evaluated to determine if they meet the verification threshold. Particles that meet the threshold are considered to have high confidence and can be retained. The first and second categories of new particles that meet the verification threshold are merged with the high-weight particle set to form a new particle set. This new particle set will be used in subsequent estimation processes to improve the accuracy and efficiency of particle filtering.
[0052] In this optional embodiment, a hybrid resampling operation, combining local and global optimization sampling, effectively improves the quality and diversity of low-weight particles. Local optimization sampling, by referencing the parameter information of high-weight particles, allows low-weight particles to more quickly approach the true state; global optimization sampling, by adjusting parameter differences, further optimizes the overall distribution of the particle set. By using a verification threshold, the quality and reliability of the new particle set are ensured, thereby significantly improving the accuracy and reliability of the particle filtering algorithm in battery remaining capacity estimation.
[0053] Optionally, a weighted calculation is performed based on the weights of the new particles in the new particle set to determine the estimated value of the state vector of the degenerate model, including: The weights of all particles in the new particle set are normalized to obtain the normalized weights of the particles, where the sum of the normalized weights of all particles is 1. By performing a weighted summation operation on all particles based on their weights, an estimate of the state vector of the degenerate model is obtained.
[0054] Specifically, the sum of the weights of all particles in the new particle set is calculated. Then, the weight of each particle is divided by this sum to obtain the normalized weights. The normalized weights represent the relative importance of each particle in the set, while ensuring that the sum of the normalized weights of all particles is 1. The weighted parameters of all particles are summed separately to obtain an estimate of the degradation model state vector. This estimate is a weighted average of the state vectors of all particles, which can more accurately reflect the actual state of the battery.
[0055] In this optional embodiment, normalized weights are used to perform weighted calculations on each particle in the new particle set, ultimately obtaining the state vector estimate of the degradation model. This process not only ensures that the contribution of each particle to the estimate is proportional to its weight, but also avoids the influence of excessively large or small weights on the calculation results through normalization. The final state vector estimate can more accurately reflect the actual state of the battery, thereby significantly improving the accuracy and reliability of the remaining battery capacity estimation.
[0056] Optionally, the estimated value of the state vector is substituted into the degradation model to obtain an estimated value of the remaining battery capacity at a preset time, including: Substitute the estimated value of the state vector into the degradation model to determine the parameter value of each parameter of the state vector; By using a double-exponential empirical formula, based on the parameter values and the time parameter at a preset time, the estimated value of the remaining battery capacity at a preset time is determined.
[0057] In this embodiment, step 1: The battery capacity is described using a double-exponential empirical degradation model based on capacity decay, and a state vector is defined. , , , , Degeneration model The four parameter estimates at time t are all real numbers, and T denotes the transpose of the vector. The process equation of the degradation model is: (1) In the formula, for Always obey Process noise, The standard deviation of process noise.
[0058] The observation equation for the degradation model is: (2) In the formula, for The battery capacity measurement value at any given time. To obey Observation noise, The standard deviation of the observation noise is represented by exp, which is an abbreviation for the exponential function and represents e raised to the power of e, where e is the natural logarithm, approximately equal to 2.71828.
[0059] Step 2: Set when The initial value of time, ,from Extracting the initial particle set .
[0060] Step 3, from Extracting particle sets The particle weights are calculated according to the following formula: (3) (4) in, Represents the state at the given previous time step. Under the given conditions, the current state The probability density function, Let f(x) represent a normal distribution, where f(x) is a normal distribution. mean Indicates the first i Each particle at time k The weights, Indicates the first i Each particle at time k A weight of -1, Indicates the state given Under the conditions, observation The probability of occurrence; Represents the state at the given previous time step. Under the given conditions, the current state The probability of; Represents the state at the given previous time step. and all observations Under the condition, the current state The probability of; This represents the particle weights after normalization. This represents the sum of the weights of all particles.
[0061] Step 4: Extract the particle set According to the following formula, it is divided into a high-weight particle set. and low-weight particle set ; (5) (6) (7) In the formula, Indicates rounding up; and : Represent the set of low-weight particles and the set of high-weight particles, respectively; This represents the threshold for dividing particles. This indicates the number of effective particles.
[0062] Step 5: For the low-weight particle set Each particle in the sample is resampled, thus generating a... Random numbers between ,if If the first proposed distribution is selected, then step 6 is executed in response to triggering local optimization sampling; otherwise, proceed to step 7 in response to triggering global optimization sampling. (8) in, This represents the resampling factor, and N represents the total number of particles.
[0063] Step 6 (in response to triggering local optimization sampling): Generate new particles using Gaussian mutation according to the following formula. And calculate the weights. Proceed to step 9.
[0064] (9) In the formula, These are particles resulting from Gaussian mutation, among which... , This represents the number of particles in the low-weight particle set. To randomly select a particle from the set of high-weight particles, , The number of particles in the high-weight particle set. The variance is a pre-set real number; Step 7 (i.e., in response to triggering global optimization sampling), generate new particles according to the principle of seeking advantages and avoiding disadvantages based on the following formula. And calculate the weights. Proceed to step 9.
[0065] (10) In the formula, For the newly generated particles, For the particles to be sampled, where, , This represents the number of particles in the low-weight particle set. It is the particle with the largest weight in the high-weight particle set. It is the particle with the smallest weight in the low-weight particle set. The number of particles in the high-weight particle set. and Each is a random number between 0 and 1; Step 8: Calculate according to formula (3) weights ,like ,but ,otherwise ,Will Add to collection In the middle, and Merging yields a new set of particles ,in, , .
[0066] Step 9: Normalize the weights of the new particles according to equation (4), and calculate the final estimated value according to the following formula; (11) Step 10, In Substitute into the following formula to estimate Battery capacity value at any given time: (12) Step 11 Proceed to step 2.
[0067] In general, combining Figure 2 As shown, in this embodiment of the invention, relevant parameters are first initialized, including the number of particles, the initial state vector, and weights. A set of particles is drawn from the prior distribution to form an initial particle set, and the initial weight of each particle is calculated. The iterative process is then initiated and repeated until a termination condition is met, such as reaching the maximum number of iterations. L In each iteration, a particle set is extracted and its weights are calculated. The weights reflect the contribution of each particle to the battery state estimation. The particle set is divided into a high-weight particle set and a low-weight particle set based on the particle's weight. A random number r is generated and compared with a threshold λ. If r ≦ λ, a Gaussian distribution is used to generate new particles (local optimization); otherwise, a principle of seeking benefits and avoiding harm is used to generate new particles (global optimization). Based on this decision, two types of new particles to be verified are generated. These two newly generated particle sets are merged and combined with the high-weight particle set to form a new particle set. Then, the weight of each particle in the new particle set is calculated. The battery capacity estimate is calculated using the new particle set. This includes calculating the estimated state vector (e.g., remaining capacity). The iteration counter is then updated. k = k +1, then check if the termination condition is met. k>λ. If the condition is not met, return to step 3 and continue iterating; if the condition is met, end the iteration. By iteratively optimizing the particle set, the representativeness and diversity of the particles are enhanced, thus more accurately tracking the battery's capacity decay process.
[0068] In this embodiment of the invention, firstly, prior distribution is utilized. Extracting particle sets And calculate particle weights ,in N For the number of particles, This is a time series. Next, the particle set... Based on particle weight, they are divided into high-weight particle sets. and low-weight particle set ,right Retain. Again, according to and The particle number ratio is selected using either Gaussian mutation or a resampling method that prioritizes the best-case scenario. Each particle in the sample is resampled to obtain a new set of particles. Finally, and The particles are merged to obtain a new set of particles. The particles in the new set are then weighted and summed to obtain the parameter estimates of the battery degradation model, and finally the battery capacity estimate.
[0069] In a preferred embodiment of the present invention, a lithium cobalt oxide battery of model Li-ion 18650, rated capacity 2Ah, and nominal voltage 3.7V was used for verification. The lithium battery capacity degradation data is as follows: Figure 3 As shown. The initial values of the model parameters are set to... , , , Set the number of simulation steps. The number of particles in the particle set is set to 168. The data on battery capacity degradation of B0007 is processed and compared with that of a traditional particle filter (PF) to obtain the system's filtering result, such as... Figure 4 As shown.
[0070] Furthermore, to more accurately analyze the effectiveness of the embodiments of the present invention, RMSE (Root Mean Square Error) was selected as a system indicator for comparative analysis. Table 1 shows that after improvement, the RMSE of the complete filtering process in this embodiment is reduced by 65.3% compared to the PF method. This demonstrates that the present embodiment has smaller deviations, better filtering effects, and higher accuracy and precision.
[0071] Table 1. Comparison of RMSE for Battery Capacity Estimation
[0072] like Figure 5 As shown, the battery remaining capacity estimation device of the present invention includes: The data processing unit is used to extract a preset number of particles from the prior distribution according to the degradation model corresponding to the battery capacity decay characteristics to obtain an initial particle set. The degradation model contains a state vector that reflects the key characteristics of capacity decay. The evaluation unit is used to evaluate the weight of each particle in the initial particle set to obtain a high-weight particle set and a low-weight particle set. The hybrid resampling unit is used to perform a hybrid resampling operation on the low-weight particle set to generate new particles, and then combine the new particles with the high-weight particle set to obtain a new particle set. The computational unit is used to perform weighted calculations based on the weights of the new particles in the new particle set to determine the estimated value of the state vector of the degenerate model; The capacity estimation unit is used to substitute the estimated value of the state vector into the degradation model to obtain the estimated value of the remaining battery capacity at a preset time.
[0073] The battery remaining capacity estimation device of the present invention has the same advantages over the prior art as the battery remaining capacity estimation method described above, and will not be repeated here.
[0074] like Figure 6 As shown, the electronic device of the present invention includes a memory and a processor; Memory, used to store computer programs; A processor, used to implement the battery remaining capacity estimation method described above when executing a computer program.
[0075] The electronic device of the present invention has the same advantages over the prior art as the aforementioned battery remaining capacity estimation method, and will not be repeated here.
[0076] The present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the battery remaining capacity estimation method described above.
[0077] The computer-readable storage medium of the present invention has the same advantages over the prior art as the aforementioned battery remaining capacity estimation method over the prior art, and will not be repeated here.
[0078] While the present invention has been disclosed above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and all such changes and modifications will fall within the scope of protection of the present invention.
Claims
1. A method for estimating the remaining capacity of a battery, characterized in that, include: Based on the degradation model corresponding to the battery capacity decay characteristics, a preset number of particles are extracted from the prior distribution to obtain an initial particle set. The degradation model contains a state vector that reflects the key characteristics of capacity decay. The weight of each particle in the initial particle set is evaluated to obtain a high-weight particle set and a low-weight particle set; A hybrid resampling operation is performed on the low-weight particle set to generate new particles, and a new particle set is obtained by combining the new particles with the high-weight particle set. The estimated value of the state vector of the degenerate model is determined by weighted calculation based on the weights of the new particles in the new particle set. Substituting the estimated value of the state vector into the degradation model, an estimated value of the remaining battery capacity at a preset time is obtained.
2. The battery remaining capacity estimation method according to claim 1, characterized in that, The process involves extracting a predetermined number of particles from the prior distribution based on the degradation model corresponding to the battery capacity decay characteristics to obtain an initial particle set, including: By combining the battery's factory-specified parameters with the capacity data obtained from previous charge-discharge tests, a reasonable range of values for the state vector is determined. Based on the reasonable value range, a prior distribution matching the parameter distribution characteristics of the state vector is constructed; The preset number of particles are extracted from the prior distribution, and the initial particle set is formed based on all the particles; wherein the dimension of the particles is the same as the dimension of the state vector.
3. The battery remaining capacity estimation method according to claim 1, characterized in that, The step of evaluating the weight of each particle in the initial particle set to obtain a high-weight particle set and a low-weight particle set includes: Based on the observation equation of the degradation model, determine the capacity measurement value of the battery at the preset time. Substitute each particle in the initial particle set into the degradation model to calculate the theoretical capacity estimate of each particle at the preset time. Based on the deviation between the theoretical capacity estimate and the capacity measurement, and in conjunction with the noise characteristics of the degradation model, the weight of each particle is determined. All particles are classified according to their weights to obtain a set of high-weight particles and a set of low-weight particles.
4. The battery remaining capacity estimation method according to claim 3, characterized in that, The step of classifying all the particles according to their weights to obtain a set of high-weight particles and a set of low-weight particles includes: The particles are divided according to the relationship between their weights and the weight division threshold. Specifically, particles with weights greater than or equal to the weight division threshold are assigned to the high-weight particle set; Particles with weights less than the weight division threshold are assigned to the low-weight particle set.
5. The battery remaining capacity estimation method according to claim 1, characterized in that, The hybrid resampling operation includes local optimization sampling and global optimization sampling. The process of performing the hybrid resampling operation on the low-weight particle set to generate new particles, and then combining these new particles with the high-weight particle set to obtain a new particle set, includes: Based on a random number within a preset numerical range, either the local optimization sampling or the global optimization sampling is triggered. In response to triggering local optimization sampling, reference particles are extracted from the high-weight particle set, and the particles in the low-weight particle set are optimized based on the parameter information of the reference particles through Gaussian mutation to generate a first type of new particles to be verified. In response to triggering global optimization sampling, the maximum weight particle in the high-weight particle set and the minimum weight particle in the low-weight particle set are obtained; Based on the principle of seeking benefits and avoiding harm, the particles in the low-weight particle set are adjusted according to the parameter differences between the maximum-weight particle and the minimum-weight particle to generate a second type of new particles to be verified. The average weight of the low-weight particle set is used as the verification threshold. The verification threshold is used to determine whether the weight of the first type of new particle to be verified or the second type of new particle to be verified meets the verification threshold requirement. The new particles that meet the verification threshold requirements are merged with the set of high-weight particles to generate the new particle set.
6. The battery remaining capacity estimation method according to claim 1, characterized in that, The step of determining the estimated value of the state vector of the degenerate model by weighted calculation based on the weights of the new particles in the new particle set includes: The weights of all particles in the new particle set are normalized to obtain the normalized weights of the particles, wherein the sum of the normalized weights of all particles is 1. The estimated value of the state vector of the degradation model is obtained by performing a weighted summation operation on all the particles based on their weights.
7. The battery remaining capacity estimation method according to claim 1, characterized in that, The step of substituting the estimated value of the state vector into the degradation model to obtain the estimated value of the remaining battery capacity at a preset time includes: Substitute the estimated value of the state vector into the degradation model to determine the parameter value of each parameter of the state vector; Using a double-exponential empirical formula, based on the parameter values and the time parameter at the preset time, an estimated value of the remaining battery capacity at the preset time is determined.
8. A battery remaining capacity estimation device, characterized in that, include: The data processing unit is used to extract a preset number of particles from the prior distribution according to the degradation model corresponding to the battery capacity decay characteristics to obtain an initial particle set. The degradation model includes a state vector that reflects the key characteristics of capacity decay. An evaluation unit is used to evaluate the weight of each particle in the initial particle set to obtain a high-weight particle set and a low-weight particle set. A hybrid resampling unit is used to perform a hybrid resampling operation on the low-weight particle set to generate new particles, and to obtain a new particle set by combining the new particles with the high-weight particle set. A computing unit is used to perform weighted calculations based on the weights of the new particles in the new particle set to determine the estimated value of the state vector of the degenerate model; The capacity estimation unit is used to substitute the estimated value of the state vector into the degradation model to obtain an estimated value of the remaining battery capacity at a preset time.
9. An electronic device, characterized in that, Including memory and processor; The memory is used to store computer programs; The processor is configured to implement the battery remaining capacity estimation method as described in any one of claims 1-7 when executing the computer program.
10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program that, when executed by a processor, implements the battery remaining capacity estimation method as described in any one of claims 1-7.