CN106055775A - Prediction method for life of secondary battery based on particle filter and mechanism model - Google Patents

Abstract

A secondary battery life prediction method combining particle filter and mechanism model, the invention is to solve the defect that the traditional secondary battery life prediction based on particle filter is completely based on data drive, ignoring the mechanism characteristics of the predicted object, which leads to the The problem of poor accuracy of life prediction results. In the training stage, the particle filter method is used to track the real value of the internal state variables of the battery to obtain the regression equation of the state variables changing with the number of charge-discharge cycles as a new state equation; in the prediction stage, the new state equation is used to calculate the estimated value of the state variable during the unknown charge-discharge cycle. Generate multiple particles and substitute them into the observation equation to obtain the estimated values of multiple capacity observations, and use the median of the estimated values of multiple capacity observations to predict the battery capacity in a certain future charge-discharge cycle. The lower limit of battery capacity, the difference between the number of cycles corresponding to the capacity prediction value and the number of cycles used in the training phase is the remaining number of cycles available for the battery.

Description

A Life Prediction Method for Secondary Batteries Combining Particle Filter and Mechanism Model

technical field

The invention relates to a new method for predicting battery life by combining secondary battery (including lithium ion battery and lead-acid battery, hereinafter referred to as battery) mechanism model simulation technology and particle filter algorithm. It belongs to the field of equipment reliability.

Background technique

In recent years, secondary rechargeable batteries such as lead-acid batteries and lithium-ion batteries have been widely used in fields such as electric vehicles and smart grids. From the perspective of use, battery life has become a bottleneck restricting the development of electric vehicles and smart grids.

Accurately predicting battery life is a basic requirement for state-based battery system maintenance, and is crucial to improving battery system reliability and cost savings. The life prediction methods of batteries can be divided into three categories: "Based on aging mechanism", it is necessary to know and model the aging mechanisms that lead to battery aging, such as reduction of catalyst effective area, reduction of available conductive ion concentration, growth of electrode passivation film, etc. , the modeling of a single aging mechanism is very complicated, and the various aging modes are coupled with each other, so the life prediction method based on the aging mechanism is difficult to realize; "data-driven", based on the historical data trend of battery capacity, combined with nonlinear Algorithms such as regression, Kalman filter, and particle filter are used to predict battery performance. This method ignores the physical meaning of data and battery objects, and it is difficult to achieve good prediction accuracy; Features predict battery life. Usually, such features are difficult to select, and the relationship between feature quantities and battery capacity is difficult to quantify.

The idea of particle filtering is based on the Monte Carlo method, which uses particle sets to represent probability and can be used in any form of state space model. Its core idea is to express the distribution of state variables through random state particles extracted from the posterior probability, which is a sequential importance sampling method. In simple terms, the particle filter method refers to the process of approximating the probability density function by finding a group of random samples propagated in the state space, and replacing the integral operation with the sample mean value, so as to obtain the minimum variance distribution of the state. The samples here refer to particles, which can approach any form of probability density distribution when the number of samples approaches infinity. Particle filter has the characteristics of non-parameterization, which gets rid of the constraint that the random quantity must satisfy the Gaussian distribution when solving the nonlinear filtering problem, and can express a wider distribution than the Gaussian model, and also has a stronger modeling of the nonlinear characteristics of variable parameters ability. Therefore, the particle filter can more accurately express the posterior probability distribution based on observations and control quantities, and can obtain more accurate system state estimation results.

Contents of the invention

The present invention aims to solve the problem that the traditional particle filter-based secondary battery life prediction is completely based on data drive and ignores the mechanism characteristics of the predicted object, resulting in poor accuracy of battery life prediction results. A secondary battery life prediction method combining particle filter and mechanism model is now provided.

A secondary battery life prediction method combining particle filter and mechanism model, which includes the following contents:

Step 1, constructing a mechanism model of the secondary battery, the mechanism model of the secondary battery can simulate the curve of the charging and discharging voltage of the battery as a function of time under any current condition;

Step 2. Training phase: Aging the secondary battery in step 1 under normal operating conditions for a period of time, and measuring the charging and discharging curves of the secondary battery during the aging process off-line with a fixed number of charge and discharge cycles at each interval using dynamic conditions. The voltage obtained at this time is the actual secondary battery output voltage U,

Input the same dynamic working condition current to the mechanism model simulation of the secondary battery, and use the model to simulate the output To simulate the actual output U of the secondary battery at different aging stages, use the genetic algorithm or the least square method to realize the identification of the secondary battery model parameter set P according to the objective function, and combine the identified multiple aging stages of the secondary battery The parameter set P is used as training data,

Select L mechanism model parameters related to the aging process from the multiple parameter sets P used for training as the state vector X, where L is a positive integer, and the battery capacity Q under the actual load current is taken as the observation quantity, and the same load current Simulation of battery mechanism model and estimated value of converted capacity under the condition of The process of the aging process is used as an observation equation, and the particle filter algorithm is used to make the estimated value of the state vector of each stage in the aging process close to the real value X;

Step 3. Prediction process: use the state vector estimated value sequence in the training process of the particle filter algorithm in step 2, and use the method of polynomial regression to obtain the regression equation of the state vector X with respect to the number of cycles k, and use this as a new state equation. When k is the number of charge and discharge cycles in the future, the estimated value of the state vector is obtained through the new state equation Substitute into the equation:

In the formula, k is the number of cycles, X _P,i,j (k) represents the i-th component of the j-th particle, and the mean value is The variance is σ _{w, the normal distribution of i} , 1≤i≤L, 1≤j≤M, w _i is the system process noise of the state variable Xi,

Obtain multiple particles that satisfy the Gaussian distribution, and substitute multiple particles into the observation equation in step 2 to obtain the median of the estimated values of multiple observations as the predicted value of the future battery capacity. When the predicted capacity reaches the pre- When the lower limit of the battery capacity is set, the difference between the corresponding number of cycles and the number of cycles used in the training phase in step 2 is the remaining number of cycles available for the battery, so as to realize the prediction of the remaining life of the secondary battery.

The beneficial effects of the present invention are: in the training stage, the particle filter method is used to track the real value of the internal state vector of the battery, and the regression equation obtained by changing the state vector with the number of charging and discharging cycles is used as a new state equation. In the prediction stage, use the new state equation to calculate the estimated value of the state variable during the unknown charge-discharge cycle, generate multiple particles on this basis, and substitute them into the observation equation to obtain the estimated value of multiple capacity observations. The median of the estimated value of the observed value is used as the prediction of the battery capacity for a certain future charge-discharge cycle. The difference in the number of cycles used is the remaining number of cycles available for the battery. It is used to predict the lifetime of electrochemical power sources.

For the first time, the mechanism electrochemical model is combined with the particle filter algorithm, and it is applied to the life prediction of the secondary battery. The prediction result of the life of the secondary battery obtained by this method is compared with the life prediction result of the secondary battery obtained by the existing method. The prediction error Reduced within 10%. This method breaks through the traditional life prediction method based entirely on data-driven particle filter. In this method, the mechanism model simulation program is used as the observer, and the mechanism model parameters that change regularly with battery aging are used as the state variables to improve the traditional particle filter prediction method. Compared with the traditional particle filter method, this method has the characteristics of high accuracy of observation equations and clear physical meaning of state variables, and can accurately predict the remaining life of the battery. It can be used for life prediction of secondary batteries of different principles.

Description of drawings

Fig. 1 is a flow chart of a secondary battery life prediction method combining particle filter and mechanism model described in Embodiment 1;

Figure 2 is a current curve diagram of a lead-acid battery under DST conditions;

Figure 3 is a voltage curve diagram of a lead-acid battery under DST conditions.

detailed description

Specific Embodiment 1: This embodiment will be described in detail with reference to FIGS. 1 to 3. A secondary battery life prediction method combining particle filter and mechanism model described in this embodiment includes the following contents:

Step 1, constructing a mechanism model of the secondary battery, the mechanism model of the secondary battery can simulate the curve of the charging and discharging voltage of the battery as a function of time under any current condition;

Step 2. Training phase: Aging the secondary battery in step 1 under normal operating conditions for a period of time, and measuring the charging and discharging curves of the secondary battery during the aging process off-line with a fixed number of charge and discharge cycles at each interval using dynamic conditions. The voltage obtained at this time is the actual secondary battery output voltage U,

Input the same dynamic working condition current to the mechanism model simulation of the secondary battery, and use the model to simulate the output To simulate the actual output U of the secondary battery at different aging stages, use the genetic algorithm or the least square method to realize the identification of the secondary battery model parameter set P according to the objective function, and combine the identified multiple aging stages of the secondary battery The parameter set P is used as training data,

Select L mechanism model parameters related to the aging process from the multiple parameter sets P used for training as the state vector X, where L is a positive integer, and the battery capacity Q under the actual load current is taken as the observation quantity, and the same load current Simulation of battery mechanism model and estimated value of converted capacity under the condition of The process of the aging process is used as an observation equation, and the particle filter algorithm is used to make the estimated value of the state vector of each stage in the aging process close to the real value X;

Step 3. Prediction process: use the state vector estimated value sequence in the training process of the particle filter algorithm in step 2, and use the method of polynomial regression to obtain the regression equation of the state vector X with respect to the number of cycles k, and use this as a new state equation. When k is the number of charge and discharge cycles in the future, the estimated value of the state vector is obtained through the new state equation Substitute into the equation:

In the formula, k is the number of cycles, X _P,i,j (k) represents the i-th component of the j-th particle, and the mean value is The variance is σ _{w, the normal distribution of i} , 1≤i≤L, 1≤j≤M, w _i is the system process noise of the state variable Xi,

Obtain multiple particles that satisfy the Gaussian distribution, and substitute multiple particles into the observation equation in step 2 to obtain the median of the estimated values of multiple observations as the predicted value of the future battery capacity. When the predicted capacity reaches the pre- When the lower limit of the battery capacity is set, the difference between the corresponding number of cycles and the number of cycles used in the training phase in step 2 is the remaining number of cycles available for the battery, so as to realize the prediction of the remaining life of the secondary battery.

In this embodiment, for the secondary battery, the aging time is, for example, 20 cycles. Under normal working conditions, the secondary battery is aged for a period of time at a fixed interval of charge and discharge cycles. According to the objective function, multiple sets of parameter sets under multiple charge and discharge cycles are measured, and the smallest sets of parameter sets are selected as training data. The particle filter algorithm is used to make the estimated value of the state vector in the training data After many cycles of charging and discharging, it is close to the real value X; using the state vector estimated value sequence in the training process of the particle filter algorithm, and using the polynomial regression method, the state equation X of the new L mechanism model parameters after multiple cycles is obtained. (k).

1. Electrochemical Mechanism Modeling

The electrochemical mechanism model here refers to its performance simulation model, that is, the input of the model is the charging or discharging current of the battery, and the output of the model is the curve of the corresponding terminal voltage changing with time.

Electrochemical models include mathematical descriptions of battery electrode thermodynamic reversible voltage (open circuit voltage), liquid phase diffusion and migration, solid phase diffusion, electrochemical reaction kinetics, etc., usually expressed as partial differential equations and their boundary conditions, initial value conditions can be solved iteratively by the finite difference method. The input and output relationship of the model can be expressed as:

U(t)=f[I(t),P(k)] (1)

Among them, the function map f( ) is the numerical simulation process of calculating the terminal voltage by the mechanism model given the current I(t) and the parameter set P(k), which describes the change of the voltage U with the charging and discharging time under specific working conditions; k is the number of charge and discharge cycles, and it is considered that the parameter set P changes with the increase of the number of charge and discharge cycles.

Because the input of the performance simulation model is the current battery state and operating conditions, and the output is the external measurable voltage, it can be used as the observation equation of the battery system. In the aging problem, the observed quantity of battery aging is usually the capacity. The discharge voltage curve of the battery under the actual load condition can be simulated by formula (1) under the condition of the battery is fully charged, and the discharge cut-off time can be determined according to the cut-off point of the discharge voltage. From the beginning of the discharge to the end of the discharge, the current is integrated with time (Ampere-hour integration method) to obtain the capacity of the battery. The calculation process of capacity can be described as

Q(k)=q[I(t),P(k)] (2)

Among them, Q(k) is the estimated value of the capacity, I(t) is the current used to measure the capacity, P(k) is the model parameter set, and q[ ] represents the process of calculating the discharge capacity according to the discharge curve and discharge time.

2. Prepare training data

In order to obtain the parameter sets of the battery at different aging stages, it is necessary to test the charge and discharge curves of the battery at different aging stages of the battery. The working condition of testing the charge-discharge curve can be selected as the dynamic stress test condition DST (DynamicStressTest), which includes the conditions reflecting various processes of the battery, the obtained voltage, current, and power data sets are rich in information, and the parameter identification results are robust. strong sex. The typical current curve of DST working condition and the corresponding voltage curve of a lead-acid battery are shown in Figure 2 and Figure 3.

The goal of parameter identification is to select a set of parameters so that the mechanism model simulates the voltage output when the same current is input. The error between the actual battery voltage output U is the smallest, and the objective function is shown in formula (3).

$\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; J</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{<span\; class="notranslate">\; \&lsqb;</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; u</span>}}<span\; class="notranslate">\; \&lsqb;</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; \&rsqb;</span><span\; class="notranslate">\; \&rsqb;</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}\\ \mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; S</span>}\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

Among them, I(t) is the load current; P is the parameter set to be identified; S is the search space of the parameter set; N is the number of data points on the voltage curve with time.

Parameter identification can be realized by genetic algorithm or least square method.

3. Some concepts of particle filter algorithm

(1) State variables

According to the parameter set and its changes in the training phase, L mechanism model parameters closely related to aging are selected as state variables, which are denoted as state vector X. Other parameters are fixed to take the average value of multiple identifications.

$\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span>\left[\begin{array}{c}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}\\ {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\\ <span\; class="notranslate">\; .</span>\\ <span\; class="notranslate">\; .</span>\\ <span\; class="notranslate">\; .</span>\\ {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; L</span>}}\end{array}\right]<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

Wherein, X ₁ ˜X _L are L parameters related to aging in the model parameter set P.

(2) Observation

The observed quantity used is the capacity of the battery, the measured value is Q, and the estimated value is

(3) Observation equation

Combine the battery terminal voltage simulation process formula (1) with the capacity calculation process formula (2) to obtain the estimated value of the observed quantity under the specific charging and discharging condition I(t) This observation process is represented by h[·] in the observation equation formula (5). In addition, the calculation of the observation value also takes into account the addition of observation noise, namely:

$\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; Q</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; \&lsqb;</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&rsqb;</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; ~</span>\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; v</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$

Among them, k is the number of cycles, and _v is the observation noise, which obeys the Gaussian distribution with mean value 0 and variance σv.

(4) Equation of state

In the training phase, it is not clear how the state variables change. The state equation can be regarded as the following recursive relationship:

$\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left[\begin{array}{c}{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>\\ {\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>\\ <span\; class="notranslate">\; .</span>\\ <span\; class="notranslate">\; .</span>\\ <span\; class="notranslate">\; .</span>\\ {\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}}_{\mathrm{<span\; class="notranslate">\; L</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>\end{array}\right]<span\; class="notranslate">\; =</span>\left[\begin{array}{c}{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span>\\ {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span>\\ <span\; class="notranslate">\; .</span>\\ <span\; class="notranslate">\; .</span>\\ <span\; class="notranslate">\; .</span>\\ {\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}}_{\mathrm{<span\; class="notranslate">\; L</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span>\end{array}\right]<span\; class="notranslate">\; +</span>\left[\begin{array}{c}{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span>\\ {\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span>\\ <span\; class="notranslate">\; .</span>\\ <span\; class="notranslate">\; .</span>\\ <span\; class="notranslate">\; .</span>\\ {\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; L</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span>\end{array}\right]<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ~</span>\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 6</span>}<span\; class="notranslate">\; )</span>$

In the formula, k is the number of cycles; is the estimated value of the state variable X _i in the kth cycle; wi is the system process noise of the state variable X _i , which obeys the Gaussian distribution with mean value 0 and variance σ _w,i .

Specific embodiment 2: This embodiment is a further description of the life prediction method of a secondary battery combined with a particle filter and a mechanism model described in specific embodiment 1. In this embodiment, in step 1, the secondary battery The mechanism model is:

U(t)=f[I(t), P(k)] (Formula 2),

In the formula, I(t) is the given current, f is the function map, P(k) is the parameter set of the secondary battery, k is the number of charge and discharge cycles, and the parameter set P changes with the increase of the number of charge and discharge k, U(t) is the external measurable voltage of the secondary battery.

Specific embodiment 3: This embodiment is a further description of a secondary battery life prediction method that combines a particle filter and a mechanism model described in specific embodiment 1. In this embodiment, in step 2, the objective function is:

In the formula, I(t) is the given current, P is the parameter set to be identified, S is the search space of the parameter set, N is the number of data points on the curve of voltage versus time, is the simulation output voltage of the mechanism model, and U is the actual battery output voltage.

Embodiment 4: This embodiment is to further explain the life prediction method of a secondary battery combined with a particle filter and a mechanism model described in Embodiment 1. In this embodiment, in step 2, the state vector X is :

In the formula, X ₁ ～X _L are L parameters related to aging in the model parameter set P.

Specific Embodiment 5: This embodiment is to further explain the life prediction method of a secondary battery combined with a particle filter and a mechanism model described in Specific Embodiment 1. In this embodiment, in step 2, the battery capacity Q The equation is:

Q(k)=q[I(t), P(k)] (Formula 5),

In the formula, I(t) is the current used to measure the capacity, P(k) is the model parameter set, and q[ ] represents the process of calculating the discharge capacity according to the discharge curve and discharge time;

In step 2, the observation equation is:

In the formula, is the estimated value of the observed quantity under a specific charge-discharge condition I(t), k is the number of charge-discharge cycles, and _v is the observation noise, which obeys a Gaussian distribution with a mean of 0 and a variance of σv.

Specific embodiment six: This embodiment is to further explain the life prediction method of a secondary battery combined with particle filter and mechanism model described in specific embodiment one. In this embodiment, in step two, the particle filter algorithm is used Estimates of the state vectors at various stages in the aging process The specific process of approaching the real value X is:

Step A1, use the particle filter algorithm to initialize particles: set the process noise variance σ _w,i of the state vector X _i , 1≤i≤L; set the observation noise variance σ _v ; set the particle number M;

Step A2, preliminary estimation of the state vector during the kth charge-discharge cycle: according to the formula:

Recursively obtain the preliminary estimated value of the state vector at the kth cycle from the final estimated value of the state vector at the k-1th cycle,

In the formula, is the estimated value of the state variable X _i in the kth cycle; w _i is the system process noise of the state variable X _i , which obeys the Gaussian distribution with mean value 0 and variance σ _w,i ;

Step A3, particle sampling during the k-th charge-discharge cycle: the state vector during each cycle corresponds to M particles, and according to formula 1, determine the particles during the k-th charge-discharge cycle;

Step A4. Calculation of importance weights: Bring M particles into the observation equation in Formula 6 to obtain M estimated values of capacity observations in, v～N(0,σ _v ), X _p,j (k) is the jth particle, is the estimated value of the observed quantity corresponding to the jth particle, 1≤j≤M,

According to M estimates The error with the measured observation Q(k), according to the formula:

Get the importance weights of different particles,

In the formula, W _j (k) is the importance weight of the jth particle, σ _v is the Gaussian distribution with variance σ _v ;

Step A5, weight normalization: according to the formula:

Divide the importance weight of each particle by the sum of all particle importance weights to realize the normalization of each particle importance weight,

In the formula, is the normalized importance weight;

Step A6, particle resampling: resampling the particles so that the probability of each particle being resampled is equal to its normalized importance weight;

Step A7, particle update: according to the formula:

Taking the average value of each dimension of the M particles after resampling as the final estimated value of the corresponding state vector, in the training phase, for each charge-discharge cycle, repeat the above steps A2 to A7, so that the estimated value of the state vector becomes more and more closer to its true value.

In this embodiment, for the sake of convenience, k represents the index of the observed data sequence, Q represents the observed quantity, w represents the process noise, v represents the observation noise, σ _w represents the variance of the process noise, and σ _v represents the variance of the observation noise. Particle sampling: According to the state equation, from the M state variable particles of the k-1th time, the M state variable particles of the kth time are deduced recursively.

Calculate the importance weight: bring the state quantity corresponding to each particle into the observation equation, and obtain the estimated value of the observation quantity The importance weights of different particles are obtained according to the error between each particle and the observed quantity Q.

The importance weight of each particle is divided by the sum of all particle importance weights to realize the normalization of each particle importance weight.

Embodiment 7: This embodiment is to further explain the life prediction method of a secondary battery combined with particle filter and mechanism model described in Embodiment 1. In this embodiment, in step 3, the specific details of the prediction process The process is:

Step B1, State Equation Update: At the end of the training phase, according to the change trend of historical state vector estimation values, obtain their regression polynomial equations that change with the number of charge and discharge cycles k, and add the system process error to obtain the updated state equation , the updated state equation is:

X(k)=f _regression (k)+w(k), w～N(0,σ _w ) (Formula 11),

In the formula, f _regression (k) represents the regression equation of the state quantity with respect to k, and w(k) is the system process noise;

Step B2, update of observations: take the estimated value of the state vector X(k) after the kth charge-discharge cycle obtained in step B1 as the initial value, substitute the estimated value into formula 1, obtain M particles, and use the M particles are substituted into formula 6 to obtain estimated values of M observed values, and the median of estimated values of M observed values is used as the predicted value of the capacity at the kth charge-discharge cycle in the future;

Step B3, remaining life prediction: compare the predicted value of the capacity at the kth charge-discharge cycle with the preset minimum capacity of the battery, and as the number of charge-discharge cycles increases, when the predicted capacity starts to be less than the set capacity, the battery The difference between the number of cycles k corresponding to the actual observations of , and the number of cycles used in the training phase in step 2 is the remaining number of cycles available for the battery, thereby obtaining the remaining life of the secondary battery.

Claims (7)

1. A secondary battery life prediction method combining particle filter and mechanism model, characterized in that it comprises the following: Step 1, constructing a mechanism model of the secondary battery, the mechanism model of the secondary battery can simulate the curve of the charging and discharging voltage of the battery as a function of time under any current condition; Step 2. Training phase: Aging the secondary battery in step 1 under normal operating conditions for a period of time, and measuring the charging and discharging curves of the secondary battery during the aging process off-line with a fixed number of charge and discharge cycles at each interval using dynamic conditions. The voltage obtained at this time is the actual secondary battery output voltage U, Input the same dynamic working condition current to the mechanism model simulation of the secondary battery, and use the model to simulate the output To simulate the actual output U of the secondary battery at different aging stages, use the genetic algorithm or the least square method to realize the identification of the secondary battery model parameter set P according to the objective function, and combine the identified multiple aging stages of the secondary battery The parameter set P is used as training data, Select L mechanism model parameters related to the aging process from the multiple parameter sets P used for training as the state vector X, where L is a positive integer, and the battery capacity Q under the actual load current is taken as the observation quantity, and the same load current Simulation of battery mechanism model and estimated value of converted capacity under the condition of The process of the aging process is used as an observation equation, and the particle filter algorithm is used to make the estimated value of the state vector of each stage in the aging process close to the real value X; Step 3. Prediction process: use the state vector estimation value sequence in the training process of the particle filter algorithm in step 2, and use the method of polynomial regression to obtain the regression equation of the state vector X with respect to the number of cycles k, and use this as a new state equation. When k is the number of charge and discharge cycles in the future, the estimated value of the state vector is obtained through the new state equation Substitute into the equation: In the formula, k is the number of cycles, X _P,i,j (k) represents the i-th component of the j-th particle, and the mean value is The variance is σ _{w, the normal distribution of i} , 1≤i≤L, 1≤j≤M, w _i is the system process noise of the state variable Xi, Obtain multiple particles that satisfy the Gaussian distribution, and substitute multiple particles into the observation equation in step 2 to obtain the median of the estimated values of multiple observations as the predicted value of the future battery capacity. When the predicted capacity reaches the pre- When the lower limit of the battery capacity is set, the difference between the corresponding number of cycles and the number of cycles used in the training phase in step 2 is the remaining number of cycles available for the battery, so as to realize the prediction of the remaining life of the secondary battery. 2. The secondary battery life prediction method combining a particle filter and a mechanism model according to claim 1, wherein in step 1, the mechanism model of the secondary battery is: U(t)=f[I(t),P(k)] (Formula 2), In the formula, I(t) is the given current, f is the function map, P(k) is the parameter set of the secondary battery, k is the number of charge and discharge cycles, and the parameter set P changes with the increase of the number of charge and discharge k, U(t) is the external measurable voltage of the secondary battery. 3. The secondary battery life prediction method combining a particle filter and a mechanism model according to claim 1, wherein, in step 2, the objective function is: In the formula, I(t) is the given current, P is the parameter set to be identified, S is the search space of the parameter set, N is the number of data points on the curve of voltage versus time, is the simulation output voltage of the mechanism model, and U is the actual battery output voltage. 4. The secondary battery life prediction method combining a particle filter and a mechanism model according to claim 1, wherein, in step 2, the state vector X is: In the formula, X ₁ ～X _L are L parameters related to aging in the model parameter set P. 5. the secondary battery life prediction method that a kind of particle filter and mechanism model are combined according to claim 1, is characterized in that, in step 2, the equation of battery capacity Q is: Q(k)=q[I(t),P(k)] (Formula 5), In the formula, I(t) is the current used to measure the capacity, P(k) is the model parameter set, and q[ ] represents the process of calculating the discharge capacity according to the discharge curve and discharge time; In step 2, the observation equation is: In the formula, is the estimated value of the observed quantity under a specific charge-discharge condition I(t), k is the number of charge-discharge cycles, and _v is the observation noise, which obeys a Gaussian distribution with a mean of 0 and a variance of σv. 6. The secondary battery life prediction method combining a particle filter and a mechanism model according to claim 1, wherein in step 2, the particle filter algorithm is used to make the estimated value of the state vector of each stage in the aging process The specific process of approaching the real value X is: Step A1, use the particle filter algorithm to initialize particles: set the process noise variance σ _w,i of the state vector X _i , 1≤i≤L; set the observation noise variance σ _v ; set the particle number M; Step A2, preliminary estimation of the state vector during the kth charge-discharge cycle: according to the formula: Recursively obtain the preliminary estimated value of the state vector at the kth cycle from the final estimated value of the state vector at the k-1th cycle, In the formula, is the estimated value of the state variable X _i in the kth cycle; w _i is the system process noise of the state variable X _i , which obeys the Gaussian distribution with mean value 0 and variance σ _w,i ; Step A3, particle sampling during the k-th charge-discharge cycle: the state vector during each cycle corresponds to M particles, and according to formula 1, determine the particles during the k-th charge-discharge cycle; Step A4. Calculation of importance weights: Bring M particles into the observation equation in Formula 6 to obtain M estimated values of capacity observations in, X _p,j (k) is the jth particle, is the estimated value of the observed quantity corresponding to the jth particle, 1≤j≤M, According to M estimates The error with the measured observation Q(k), according to the formula: Get the importance weights of different particles, In the formula, W _j (k) is the importance weight of the jth particle, σ _v is the Gaussian distribution with variance σ _v ; Step A5, weight normalization: according to the formula: Divide the importance weight of each particle by the sum of all particle importance weights to realize the normalization of each particle importance weight, In the formula, is the normalized importance weight; Step A6, particle resampling: resampling the particles so that the probability of each particle being resampled is equal to its normalized importance weight; Step A7, particle update: according to the formula: Taking the average value of each dimension of the M particles after resampling as the final estimated value of the corresponding state vector, in the training phase, for each charge-discharge cycle, repeat the above steps A2 to A7, so that the estimated value of the state vector becomes more and more closer to its true value. 7. The secondary battery life prediction method combining a particle filter and a mechanism model according to claim 1, wherein in step 3, the specific process of the prediction process is: Step B1, State Equation Update: At the end of the training phase, according to the change trend of historical state vector estimates, obtain their regression polynomial equations that change with the number of charge and discharge cycles k, and add the system process error to obtain the updated state equation , the updated state equation is: X(k)＝f _regression (k)+w(k),w～N(0,σ _w ) (Formula 11), In the formula, f _regression (k) represents the regression equation of the state quantity with respect to k, and w(k) is the system process noise; Step B2, update of observations: take the estimated value of the state vector X(k) after the kth charge-discharge cycle obtained in step B1 as the initial value, substitute the estimated value into formula 1, obtain M particles, and use the M particles are substituted into formula 6 to obtain estimated values of M observed values, and the median of estimated values of M observed values is used as the predicted value of the capacity at the kth charge-discharge cycle in the future; Step B3, remaining life prediction: compare the predicted value of the capacity at the kth charge-discharge cycle with the preset minimum capacity of the battery, and as the number of charge-discharge cycles increases, when the predicted capacity starts to be less than the set capacity, the battery The difference between the number of cycles k corresponding to the actual observations of , and the number of cycles used in the training phase in step 2 is the remaining number of cycles available for the battery, thereby obtaining the remaining life of the secondary battery.