A power battery state of charge estimation method based on improved particle filtering
By improving the particle filtering method, using hyperparallel member filtering and maximum posterior probability density function criteria, the feasible particle domain is screened, which solves the problem of accurate estimation of the nonlinear system of the power battery state of charge, and improves the estimation accuracy and adaptability , reducing particle degradation.
Patent Information
- Application Number
- CN202410311075.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-19
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-03-19
AI Technical Summary
The prior art is difficult to accurately estimate the state of charge of a power battery, especially when the system of the battery shows nonlinear characteristics after aging, resulting in inaccurate state of charge estimation results.
Using a method based on improved particle filtering, a nonlinear power battery state-of-charge estimation model is established. Through the superparallel group filtering idea and the maximum posterior probability density function criterion, the feasible particle domain is selected, new effective particles are generated, and the Sampling for more accurate state of charge estimation.
It improves the accuracy and adaptability of state-of-charge estimation, reduces particle degradation, maintains particle diversity, and achieves a good balance between calculation complexity and conservatism, which is suitable for state estimation of nonlinear systems.
Smart Images

Figure CN118191610B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a method for estimating the state of charge of a power battery based on improved particle filtering, and belongs to the technical field of power battery state estimation. Background Art
[0002] In the context of the world's increasing attention to environmental protection and sustainable development, new energy vehicles have rapidly emerged and become an important alternative to traditional energy vehicles. As the heart of new energy vehicles, the performance of power batteries directly determines the vehicle's endurance and the practical life of portable electronic devices. Power batteries are a complex electrochemical system, and accurate estimation of their state is the key to ensuring their normal operation. The state of charge (SOC) of a battery, that is, the ratio of remaining capacity to rated capacity, is the core indicator for evaluating the health of a battery. However, as an intrinsic property of a battery, the battery state of charge cannot be measured directly and must be indirectly inferred by external indicators such as voltage and current. As the use time increases, the power battery will gradually age, which will cause changes in the internal characteristics of the battery, making the battery state of charge system present nonlinear characteristics. Therefore, the estimation results of the battery state of charge obtained by the existing estimation methods for linear systems are inaccurate. Therefore, in order to better adapt to the performance changes of batteries in continuous use and provide solid technical support for the intelligent management of power batteries, it is particularly important and urgent to develop a battery state of charge estimation technology that is both accurate and adaptable.
[0003] In order to solve the state estimation problem of nonlinear non-Gaussian systems such as the battery state of charge system, scholars have proposed a particle filter based on the Monte Carlo method (for a detailed introduction, please refer to the introduction in "Askari I, Haile MA, Tu X, et al. Implicit Particle Filtering via a Bank of Nonlinear Kalman Filters [J]. 2022. DOI: 10.48550 / arXiv.2205.04521."). For the state estimation of nonlinear systems, the posterior probability density function in this method is characterized by a set of weighted random particles, and then the sample mean is used instead of the integral operation to obtain the minimum variance distribution of the state, thereby achieving relatively accurate state estimation for nonlinear non-Gaussian systems such as the battery state of charge system. This method has also been widely studied and applied in the fields of signal processing, image processing, and target tracking in complex environments. However, in this method, during the iteration process of the particle filter, due to its large dependence on the sample particle size, the resampling method used to improve the particle shortage phenomenon discards particles with small weights and selects significant particles for replication, which will lead to particle degradation. In order to solve this problem, many scholars have proposed various methods. Most of the improved methods involve selecting a high likelihood region for the particles and combining other algorithms to optimize the resampling process, such as the particle filter algorithm based on the hummingbird optimization algorithm combination, the simplified particle filter algorithm based on the genetic algorithm, etc. Experimental results show that these methods can reduce particle degradation and improve state estimation performance to a certain extent. However, most of these solutions are highly complex and rarely consider the quality of the resampled particles and their impact on state estimation. Summary of the invention
[0004] In order to solve the existing problems, the present invention provides a method for estimating the state of charge of a power battery based on an improved particle filter, comprising:
[0005] Step 1: Establish a nonlinear power battery state of charge estimation model with unknown but bounded noise;
[0006] Step 2: Generate initial particles and set the system at the current state point Perform a first-order linear expansion at , and bound the linearization error term according to the DC planning criterion;
[0007] Step 3: Prediction step: Introduce the hyperparallel body to wrap the linearized part, the linearized error term, and the noise term respectively, and reorganize the Minkowski sum of the three into different hyperparallel spaces, from which the minimum volume of the outer hyperparallel body state space containing feasible prior particles is selected, that is, the prediction step hyperparallel body
[0008] Step 4: Update step: Linearize the system output equation in the same way as step 2, and output the data y at time k k Construct a linear measurement set and normalize it to a measurement band S(c k,i ,p k,i ), the predicted superparallel body and the measurement band are respectively tightened and the intersection is taken in turn, and the recursive update step is performed to obtain the minimum outer boundary superparallel body As the feasible domain of the a posteriori particle:
[0009] Step 5: Particle filtering: According to the feasible domain of the posterior particles, the prior particles are retained or discarded, new valid particles are generated according to the MAP criterion, the posterior particle weights are normalized and resampled, and the posterior state estimation result at time k is obtained. As the estimated result of the power battery state of charge at time k.
[0010] Optionally, the step 1 includes:
[0011] The nonlinear discrete general analysis model of the power battery state of charge estimation system is established as:
[0012]
[0013] in, Represents the state vector of the system at time k, including the battery state of charge (SOC) and charging current i k , represents the system output at time k, including the terminal voltage of the battery at time k; f(x k ) and h(x k ) are about x k The second-order nonlinear continuous differentiable function, ω k and v k are the unknown but bounded system noise and measurement noise in the system, satisfying where |ω k,j |≤r ω (j=1,…,n),|v k,p |≤r v (p=1,…,z), r ω 、r v is the noise boundary.
[0014] Optionally, the step 2 includes:
[0015] Using (0,1) uniform distribution, n initial hyperparallel search spaces are generated.
[0016] The initial particle is denoted by
[0017] Generate prior particles according to the system state function:
[0018]
[0019] in
[0020] For the nonlinear system in (1), at the current state point A first-order linear expansion is performed at , and the linearization error term is bounded, including:
[0021]
[0022] Among them A k For function f=[f1,f2,…,f n ] T The Jacobian matrix, e k is the linearization error of the state function.
[0023] According to DC programming theory, construct convex functions g1(x) and g2(x), and use the difference between the two to represent the component f of function f i (x), that is:
[0024] f i (x)=g1(x)-g2(x), (4)
[0025] g1(x)=f i (x)+αx T , (5)
[0026] g2(x)=αx T x. (6)
[0027] At the current state Linearizing the convex functions g1(x) and g2(x) yields:
[0028]
[0029]
[0030] Among them, G1 and G2 are the sub-gradients of g1 and g2 respectively, which can be calculated as:
[0031]
[0032] Therefore, the linearization error e in equation (3) can be k According to the DC planning criteria, it is defined as:
[0033]
[0034]
[0035] in is the system state quantity to be established including k-1 time points The search space of superparallel bodies.
[0036] Linearization error e k It has a superparallel boundary of the following form:
[0037]
[0038] in represents the center of the superparallel body, represents the radius of the interval where the superparallel body is located, Represents the shape matrix of the superparallel body. α represents any vector in the n-dimensional unit box; R n×n Represents n×n dimensional real number space.
[0039] The state equation of the system at time k can be linearized as:
[0040] x k =f L (x k-1 )+e k +ω k (13)
[0041] Optionally, step 3 includes:
[0042] In the state equation of the linearized system described in formula (13), for the linearized part f L (x k-1 ), linearization error term e k and the noise term u k , respectively corresponding to the introduction of superparallel The Minkowski sum of the three can be written as a fully symmetric polyhedron
[0043]
[0044] in
[0045] The fully symmetric polytope Split into Half-spaces:
[0046]
[0047] in
[0048]
[0049]
[0050]
[0051]
[0052] in, represents a matrix consisting of m-n+1 vectors randomly taken from the matrix T, Represents C + The lth row vector of l Represents the l-th row element of d.
[0053] For each half space There is a half space parallel to it. Two parallel half-spaces can form a strip space
[0054]
[0055] in, A superparallel state space is obtained by the intersection of each two non-parallel band spaces, from which the minimum volume of the outer superparallel state space containing feasible prior particles can be selected. Assuming that when l = vmin, the volume of the superparallel state space is the smallest, then this superparallel state space It can be expressed as:
[0056]
[0057] in
[0058]
[0059]
[0060] in, Right now The nth component of the vminth row vector of Represents C + The i-th component of the vmin-th row vector of Represents the i-th component of the vmin-th row vector of Δd.
[0061] Optionally, step 4 includes:
[0062] The output equation in system (1) is linearized and DC-planned in a manner similar to equation (3). After linearization, the measurement equation can be expressed as:
[0063]
[0064] Among them B k is the Jacobian matrix of function h, o kis the linearization error of the measurement function. Similar to equations (4) to (11), the linearization error can be calculated according to the DC planning criterion. k Determining the boundaries of a hyperparallel body Calculate k and the measurement noise v k The Minkowski sum, denoted as auxiliary variable μ, is:
[0065]
[0066] According to the output data y of the nonlinear discrete model of the system at time k k Get the set of linear measurements:
[0067]
[0068] Normalize this measurement set into the form of an n-dimensional measurement band, that is:
[0069]
[0070] The prediction step will be super parallel With measuring tape S(c k,i ,p k,i ) Take the intersection in turn and perform the recursive update step. Calculate the first recursion As an example, show the recursive process.
[0071] remember p0=p k,1 、c0=c k,1 .
[0072] calculate like Then replace until Established.
[0073] Tighten the measuring tape S(c k,1 ,p k,1 ) to make it parallel to the predicted step Intersect to obtain a tight measurement band in:
[0074]
[0075]
[0076]
[0077] By scaling the vector Reduce the prediction step superparallel The overlapping volume with the measuring tape. When , the superparallel body does not need to be compacted; when By setting the parameter The compact superparallel expression is obtained. Combining the two cases, the compact superparallel expression is obtained. expression:
[0078]
[0079] in
[0080]
[0081] Compacting superparallel With tightening measuring tape Intersect and get the updated superparallel body
[0082]
[0083] in
[0084]
[0085]
[0086]
[0087] After repeating the above algorithm z times, the minimum outer boundary superparallel body is obtained in That is the feasible region of the a posteriori particle.
[0088] Optionally, step 5 includes:
[0089] According to the feasible domain of the posterior particles, the retention and abandonment of the prior particles are completed. The set of n prior particles generated within k time is The feasible region of the particle is The weight of each particle is calculated as:
[0090]
[0091] The retained particles are denoted as The discarded particles are recorded as The maximum a posteriori probability density function criterion (MAP criterion) is used to map particles outside the feasible domain to the feasible domain, ensuring that the total number of particles n remains unchanged and generating new valid particles. To ensure the correlation between particles and models. According to the MAP criterion, the state quantity The estimation process is:
[0092]
[0093] Visible particle domain Priori particle The constraint region, the optimization equation is:
[0094]
[0095] Thus, the posterior particle can be expressed as:
[0096]
[0097] Normalize the posterior particle weights to get the normalized weight value
[0098]
[0099] The particles are resampled using the resampling method, and the posterior state estimation result at time k is obtained as follows:
[0100]
[0101] This result is used as the power battery state of charge estimation result at time k, realizing the power battery state of charge estimation.
[0102] We can further use the optimal estimate of the battery state of charge at time k Determine the state of charge estimation interval at time k:
[0103]
[0104] in, and They represent the upper and lower bounds of the battery state of charge estimation interval, and represent the upper and lower bounds of the unknown but bounded state estimation error interval, respectively.
[0105] The present application also provides the application of the above method in state estimation of nonlinear non-Gaussian systems and power battery production.
[0106] The beneficial effects of the present invention are:
[0107] Different from the existing particle filter state estimation method, the method of the present application obtains a nonlinear discrete general analysis model of the power battery state of charge estimation system containing unknown but bounded noise, and based on the particle filter idea improved by hyperparallel body set membership filtering, designs an update strategy for the intersection of the hyperparallel body space and the band space reorganized by half space, and screens the hyperparallel body space by the minimum volume standard, which provides a particle feasible domain for the particle filter, thereby realizing particle screening within the particle feasible domain. That is, the method of the present application innovatively proposes a method of splitting the fully symmetric polyhedron into In the way of half-spaces, two parallel half-spaces form a strip space, and then a super-parallel state space is obtained by the intersection of each two non-parallel strip spaces. The minimum volume outsourcing super-parallel state space containing feasible a priori particles is selected from all super-parallel state spaces, so as to realize particle screening within the feasible domain of particles; this application also maps particles outside the feasible domain to the feasible domain based on the maximum a posteriori probability density function criterion (MAP criterion), so that while maintaining particle diversity, the particle distribution is closer to the real state. This method achieves a good balance between computational complexity and conservatism, and can be effectively applied to nonlinear system state estimation engineering scenarios such as power battery charge state estimation. BRIEF DESCRIPTION OF THE DRAWINGS
[0108] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0109] Figure 1 The present invention discloses a flow chart of a method for estimating the state of charge of a power battery based on an improved particle filter in one embodiment of the present invention.
[0110] Figure 2 Schematic diagram of the feasible region of the true value of the battery state of charge and the particles when the method described in the present application is used to estimate the state of charge of the power battery CM model when the sampling interval is Δk=10.
[0111] Figure 3 This is a graph showing the state of charge estimation results of the power battery CM model at the same time using the method provided by the present application and the existing particle filtering method.
[0112] Figure 4 The absolute value curve of the state of charge estimation error of the power battery CM model using the method provided by the application and the existing particle filtering method at the same time. DETAILED DESCRIPTION
[0113] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.
[0114] Embodiment 1:
[0115] This embodiment provides a method for estimating the state of charge of a power battery based on an improved particle filter. Figure 1 , the method comprising:
[0116] Step 1: Establish a nonlinear discrete CM analysis model for the power battery state of charge estimation system:
[0117]
[0118] The specific expression is as follows:
[0119]
[0120] Among them, x k =[x 1,k x 2,k ] T =[SOC k i k ] T represents the state vector of the system at time k, the Coulomb efficiency of the battery is η = 0.8, Δt = 5 is the sampling interval, C is the nominal capacity of the battery, SOC k represents the battery state of charge at time k, i k represents the charging current at time k, ω k is the unknown but bounded system noise in the system and |ω k,j |≤0.002(j=1,2); represents the system output at time k, which refers to the terminal voltage of the battery at time k, E0 is the initial electromotive force of the battery, R is the internal resistance of the battery, K1, K2, K3, K4 are the model matching coefficients of a single battery (for a detailed introduction, please refer to "Wu Lanhua, Yang Xiuzhi, Zheng Mingkui, et al. A lithium battery SOC estimation algorithm based on optimized particle filtering [J]. Journal of Fuzhou University (Natural Science Edition), 2018, 46(02): 186-191"), as shown in Table 1, v k is the unknown but bounded system noise in the system and |v k |≤0.001.
[0121] Table 1: Power battery parameter values
[0122]
[0123] Step 2: Generate n initial particles in the initial hyperparallel search space using the (0,1) uniform distribution, denoted as According to the state function of the power battery state of charge estimation system, a priori particles are generated (the above subscripts distinguish a priori particles from a posteriori particles, where negative represents a priori particles and positive represents a posteriori particles), that is, the a priori particles at time k-1 are used to obtain the a priori particles at time k:
[0124]
[0125] in
[0126] For the nonlinear system in (1), at the current state point A first-order linear expansion is performed at , and the linearization error term is bounded, including:
[0127]
[0128] Among them A k For function f = [f1,f2,…,f n ] T The Jacobian matrix, e k is the linearization error of the state function.
[0129] According to DC programming theory, construct convex functions g1(x) and g2(x), and use the difference between the two to represent the component f of function f i (x), that is:
[0130] f i (x)=g1(x)-g2(x), (4)
[0131] g1(x)=f i (x)+αx T , (5)
[0132] g2(x)=αx T x. (6)
[0133] At the current state Linearizing the convex functions g1(x) and g2(x) yields:
[0134]
[0135]
[0136] Among them, G1 and G2 are the sub-gradients of g1 and g2 respectively, which can be calculated as:
[0137]
[0138] Therefore, the linearization error e in equation (3) can be k According to the DC planning criteria, it is defined as:
[0139]
[0140]
[0141] in is the system state quantity to be established including k-1 time points The search space of superparallel bodies.
[0142] Linearization error e kIt has a superparallel boundary of the following form:
[0143]
[0144] in
[0145] The state equation of the system at time k can be linearized as:
[0146] x k =f L (x k-1 )+e k +ω k (13)
[0147] Step 3: Introduce the hyperparallel body to wrap the linearized part, the linearized error term, and the noise term respectively, and reorganize the Minkowski sum of the three into different hyperparallel spaces, from which the minimum volume of the outer hyperparallel body state space containing all the prior particles is selected, that is, the prediction step hyperparallel body
[0148] In the state equation of the linearized system described in formula (13), for the linearized part f L (x k-1 ), linearization error term e k and the noise term u k , respectively corresponding to the introduction of superparallel The Minkowski sum of the three can be written as a fully symmetric polyhedron
[0149]
[0150] in
[0151] The fully symmetric polytope Split into Half-spaces:
[0152]
[0153] in
[0154]
[0155]
[0156]
[0157]
[0158] in, represents a matrix consisting of m-n+1 vectors randomly taken from the matrix T, Represents C + The lth row vector of l Represents the l-th row element of d.
[0159] For each half space There is a half space parallel to it. Two parallel half-spaces can form a strip space
[0160]
[0161] in, A superparallel state space is obtained by the intersection of each two non-parallel band spaces, from which the minimum volume of the outer superparallel state space containing feasible prior particles can be selected. Assuming that when l = vmin, the volume of the superparallel state space is the smallest, then this superparallel state space It can be expressed as:
[0162]
[0163] in
[0164]
[0165]
[0166] Step 4: Linearize and DC plan the output equation in system (1) in a similar way to equation (3). After linearization, the measurement equation can be expressed as:
[0167]
[0168] Among them B k is the Jacobian matrix of function h, o k is the linearization error of the measurement function. Similar to equations (4) to (11), the linearization error can be calculated according to the DC planning criterion. k Determining the boundaries of a hyperparallel body Calculate k and the measurement noise v k The Minkowski sum, denoted as auxiliary variable μ, is:
[0169]
[0170] According to the output data y of the nonlinear discrete model of the system at time k k Get the set of linear measurements:
[0171]
[0172] Normalize this measurement set into the form of an n-dimensional measurement band, that is:
[0173]
[0174] The prediction step will be super parallel With measuring tape S(c k,i ,p k,i ) Take the intersection in turn and perform the recursive update step. Calculate the first recursion As an example, show the recursive process.
[0175] remember p0=p k,1 、c0=c k,1 .calculate like Then replace until Established.
[0176] Tighten the measuring tape S(c k,1 ,p k,1 ) to make it parallel to the predicted step Intersect to obtain a tight measurement band in:
[0177]
[0178]
[0179]
[0180] By scaling the vector Reduce the prediction step superparallel The overlapping volume with the measuring tape. When , the superparallel body does not need to be compacted; when By setting the parameter The compact superparallel expression is obtained. Combining the two cases, the compact superparallel expression is obtained. expression:
[0181]
[0182] in
[0183]
[0184] Compacting superparallel With tightening measuring tape Intersect and get the updated superparallel body
[0185]
[0186] in
[0187]
[0188]
[0189]
[0190] After repeating the above algorithm z times, the minimum outer boundary superparallel body is obtained in That is, the a posteriori particle feasible region.
[0191] Step 5: Based on the feasible domain of the posterior particles, the retention and discard of the prior particles are completed.
[0192] The set of n prior particles generated within k time is The feasible region of the particle is The weight of each particle is calculated as:
[0193]
[0194] The retained particles are denoted as The discarded particles are recorded as The MAP criterion is used to map particles outside the feasible domain to the feasible domain, ensuring that the total number of particles n remains unchanged and generating new valid particles. To ensure the correlation between particles and models. According to the MAP criterion, the state quantity The estimation process is:
[0195]
[0196] Visible particle domain Priori particle The constraint region, the optimization equation is:
[0197]
[0198] Thus, the posterior particle can be expressed as:
[0199]
[0200] Normalize the posterior particle weights to get the normalized weight value
[0201]
[0202] The particles are resampled using the resampling method, and the posterior state estimation result at time k is obtained as follows:
[0203]
[0204] This result is used as the power battery state of charge estimation result at time k, realizing the power battery state of charge estimation.
[0205] We can further use the optimal estimate of the battery state of charge at time k Determine the state of charge estimation interval at time k:
[0206]
[0207] in, and They represent the upper and lower bounds of the battery state of charge estimation interval, and represent the upper and lower bounds of the unknown but bounded state estimation error interval, respectively.
[0208] Figure 2 It represents the schematic diagram of the state true value of the battery state of charge and the feasible domain of the particles when the method described in this application is used to estimate the state of charge of the power battery CM model when the sampling interval is Δk=10. It can be observed from the figure that the hyperparallel feasible domain calculated by the method described in this application can effectively and tightly wrap the state true value of the current battery state of charge, providing a more accurate trust domain for particle filtering. With the continuous evolution of the feasible domain, the defined particle feasible domain continues to shrink and approach the true value, and the algorithm accuracy gradually improves.
[0209] Figure 3 In order to use the method provided by this application and the existing particle filtering method (the existing method is the traditional particle filtering based on the Monte Carlo method, please refer to "Askari I, Haile MA, Tu X, et al. Implicit Particle Filtering via a Bank of Nonlinear Kalman Filters [J]. 2022. DOI: 10.48550 / arXiv.2205.04521." Implicit particle filtering via a bank of nonlinear Kalman filters) at the same time, the state of charge estimation result diagram of the power battery CM model, the sampling interval Δt = 5. From Figure 3It can be observed that, compared with the existing methods, the battery state of charge estimate calculated by the method proposed in this application is closer to the true value because the method further accurately optimizes the sub-feasible domain of the particle; and due to the addition of the ultra-parallel ensemble member filtering algorithm, the method proposed in this application can further calculate the estimated upper and lower bounds of the battery state of charge, which can be subsequently used in the field of power battery fault diagnosis and isolation.
[0210] Figure 4 To estimate the absolute value curve of the state of charge error of the power battery CM model at the same time using the method provided by the present application and the existing particle filtering method, the sampling interval Δt=5. Figure 4 It can be observed that the absolute value of the estimation error obtained by the method provided by the present application is significantly smaller than that of the existing method, which further demonstrates that the method provided by the present application has better conservatism and higher accuracy.
[0211] Some steps in the embodiments of the present invention may be implemented using software, and the corresponding software program may be stored in a readable storage medium, such as a CD or a hard disk.
[0212] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for estimating the state of charge of a power battery based on improved particle filtering, characterized in that: The method comprises: Step 1: Based on the nonlinear characteristics of the power battery state of charge system, a nonlinear power battery state of charge estimation model with unknown but bounded noise is established; Step 2: Based on the particle filter idea, a group of initial particles are randomly generated, and then the state function of the state of charge system is combined to generate a priori particles, and the state of charge system is placed at the current state point A first-order linear expansion is performed at the position, and the linearization error term is bounded according to the DC planning criterion; the state point of the state of charge system is based on the battery state of charge SOC k and the charging current i k Indicates; the linearization error refers to the error generated by the state function when performing a first-order linear expansion; Step 3: Introduce the hyperparallel body to wrap the linearized part, the linearized error term, and the noise term respectively, and reorganize the Minkowski sum of the three into different hyperparallel spaces, from which the minimum volume of the outer hyperparallel body state space containing all the prior particles is selected, that is, the prediction step hyperparallel body in, represents the predicted value of the system state at time k, which is the center of the prediction step superparallel body of the prediction step superparallel body. The shape matrix representing the predicted superflat shape; Step 4: Linearize the output equation of the state of charge system and output data y at time k k Construct a linear measurement set and normalize it to a measurement band S(c k,i ,p k,i ), the predicted step superparallel body and the measurement band are respectively compressed and the intersection is taken in turn, and recursively updated to obtain the minimum outer boundary superparallel body As the feasible domain of the a posteriori particles; Step 5: According to the feasible domain of the posterior particles, the prior particles are retained or discarded, and new valid particles, namely the posterior particles, are generated according to the MAP criterion. The posterior particle weights are normalized and resampled to obtain the posterior state estimation result at time k. As the estimated result of the power battery state of charge at time k; The step 1 comprises: Establish a nonlinear discrete general analysis model for the power battery state of charge estimation system: in, Represents the state vector of the system at time k, including the battery state of charge SOC k and the charging current i k , represents the system output at time k, including the terminal voltage of the battery at time k; f(x k ) and h(x k ) are all about the state vector x k The second-order nonlinear continuous differentiable function, ω k and v k are the unknown but bounded system noise and measurement noise in the system, satisfying where |ω k,j |≤r ω (j=1,…,n),|v k,p |≤r v (p=1,…,z), r ω 、r v is the noise boundary; The step 2 comprises: Generate n initial particles in the initial hyperparallel search space using (0,1) uniform distribution, and generate prior particles according to the system state function; Set the state of charge system to the current state point A first-order linear expansion is performed at , and the linearization error term is bounded, including: Among them A k For function f=[f1,f2,…,f n ] T The Jacobian matrix, e k is the linearization error term of the state function; According to DC programming theory, convex functions g1(x) and g2(x) are constructed, and finally the linearization error e in equation (3) is k According to the DC planning criteria, it is defined as: in is the system state quantity to be established including k-1 time points The search space of superparallel bodies; and Respectively represent the convex functions g1(x) and g2(x) at the state point The linear expansion form at f i,L (x k-1 ) represents f in formula (3) L (x k-1 )’s i-th component; Linearization error e k The corresponding superparallel body boundary is: in α represents any vector in the n-dimensional unit box; The state equation of the system at time k is linearized as: x k =f L (x k-1 )+e k +ω k (13) Among them, f L (x k-1 ) represents the linearized part of the state equation, e k and ω k denote the linearization error term and the noise term respectively; The step 3 comprises: For the state equation after the system is linearized, for the linearized part f L (x k-1 ), linearization error term e k and the noise term ω k , respectively corresponding to the introduction of superparallel Package; the Minkowski sum of the three corresponds to the fully symmetric polytope for: in Fully symmetric polytope Split into Half-spaces: in in, represents a matrix consisting of m-n+1 vectors randomly taken from the matrix T, Represents C + The lth row vector of l Represents the l-th row element of d; For each half space There is a half space parallel to it. Two parallel half-spaces form a strip space in, A superparallel body state space is obtained by the intersection of every two non-parallel strip spaces, from which the minimum volume outer superparallel body state space containing feasible prior particles is selected; Assuming that when l = vmin, the volume of the superparallel state space is the smallest, then the superparallel state space It is expressed as: in in, Represents C + The i-th component of the vmin-th row vector of represents the i-th component of the vmin-th row vector of Δd; The step 4 comprises: The output equation of the state of charge system is linearized to obtain the linearized expression of the measurement equation: Among them B k is the Jacobian matrix of function h, o k is the linearization error of the measurement function; According to the DC planning principle, the linearization error is o k Determining the boundaries of a hyperparallel body Calculate k and the measurement noise v k The Minkowski sum, denoted as auxiliary variable μ, is: According to the output data y of the nonlinear discrete model of the system at time k k Get the set of linear measurements: Normalize this measurement set into the form of an n-dimensional measurement band, that is: The prediction step will be super parallel With measuring tape S(c k,i ,p k,i ) Take the intersection in turn and recursively update until the minimum outer boundary superparallel body is obtained That is, the feasible domain of the a posteriori particle; G k,k =T k , participate in the iteration at time k+1.
2. The method according to claim 1, characterized in that The step 5 comprises: The set of n prior particles generated within k time is The feasible region of the particle is The weight of each particle is calculated as: The retained particles are denoted as The discarded particles are recorded as The MAP criterion is used to map particles outside the feasible domain to the feasible domain, ensuring that the total number of particles n remains unchanged and generating new valid particles. To ensure the correlation between particles and the model: According to the MAP criterion, the state quantity The estimation process is: Visible particle domain For the prior particle The constraint region, the optimization equation is: Therefore, the posterior particle is expressed as: Normalize the posterior particle weights to get the normalized weight value The particles are resampled using the resampling method, and the posterior state estimation result at time k is obtained as follows: This result is used as the power battery state of charge estimation result at time k, realizing the power battery state of charge estimation.
3. The method according to claim 2, characterized in that The method further comprises step 6, based on the optimal estimated value of the battery state of charge at time k Determine the state of charge estimation interval at time k: in, and They represent the upper and lower bounds of the battery state of charge estimation interval, and represent the upper and lower bounds of the unknown but bounded state estimation error interval, respectively.
4. The method according to claim 3, characterized in that The specific expression of the nonlinear discrete universal analysis model of the power battery state of charge estimation system is: Among them, x k =[x 1,k x 2,k ] T =[SOC k i k ] T represents the state vector of the system at time k, η is the Coulomb efficiency of the battery, Δt is the sampling interval, C is the nominal capacity of the battery, SOC k represents the battery state of charge at time k, i k represents the charging current at time k, ω k is the unknown but bounded system noise in the system, and |ω k,j |≤0.002(j=1,2); It represents the system output at time k, including the terminal voltage of the battery at time k, E0 is the initial electromotive force of the battery, R is the internal resistance of the battery, and K1, K2, K3, K4 are the parameters of a single battery.
5. The method according to claim 4, characterized in that The method further includes acquiring output data of the battery terminal voltage within a predetermined time range as an output value of the system.
6. Application of the method according to any one of claims 1 to 5 in a power battery production process.
