Battery charge state estimation method based on memory elimination recursive least square method
By introducing the memory recursive least squares method and Kalman filtering algorithm in the battery state of charge estimation, the accuracy and calculation efficiency problems of battery state of charge estimation are solved, and accurate online estimation of battery state of charge is realized. Optimized scheduling and system security of energy storage networks.
Patent Information
- Application Number
- CN202410975018.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-19
- Publication Date
- 2024-11-12
AI Technical Summary
The existing battery residual capacity detection methods have problems such as inaccurate estimation and huge data, resulting in difficulty in solving and too long solution time, and lack accurate battery state of charge estimation methods, which affects the optimization scheduling and safety of the power system.
Based on the memory recursive least squares method, the identification model is constructed, the memory factor is introduced, and the Kalman filtering algorithm is combined to realize the online accurate estimation of the battery's state of charge. By constructing a first-order RC model, Larchite transform, discretization, Z-inverse transform and Introduce forgetting factors to form a joint algorithm for parameter identification and state estimation.
Improve the accuracy and calculation speed of battery state of charge estimation, narrow the gap between simulation model and actual model, adapt to different working states, enhance the real-time and accuracy of state estimation, and support the optimization scheduling and economics of energy storage networks. Safe system operation.
Smart Images

Figure CN118938003A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of power system operation and control, and in particular relates to a battery charge state estimation method based on a memory elimination recursive least squares method. Background Art
[0002] In recent years, the large-scale grid connection of renewable energy in the power system has brought a series of uncertainties. The energy storage system has a good effect in solving such problems. Its wide application in the power system will become an inevitable trend in the development of smart grids. However, the current single energy storage equipment still lacks the means of accurate estimation, which brings difficulties to the optimization and scheduling. At present, the commonly used methods for detecting the remaining capacity of batteries at home and abroad include ampere-hour measurement method, power accumulation method, open circuit voltage method, linear model method, Kalman filter method, neural network method, etc. However, these methods may have the problem of inaccurate estimation, or the problem of large data causing difficulty in solving and too long solving time.
[0003] In recent years, the large-scale grid connection of renewable energy in the power system has brought a series of uncertainties. The energy storage system has a good effect in solving such problems. Its wide application in the power system will become an inevitable trend in the development of smart grids. However, the current single energy storage equipment still lacks the means of accurate estimation, which adds unsafe and uneconomic factors to the optimization and dispatching of the power system. At present, the commonly used methods for detecting the remaining capacity of batteries at home and abroad include ampere-hour measurement method, power accumulation method, open circuit voltage method, linear model method, Kalman filter method, neural network method, etc. However, these methods may have the problem of inaccurate estimation, or the problem of large data causing difficulty in solving and too long solving time. Summary of the invention
[0004] The technical problem solved by the present invention is to provide a battery state of charge estimation method based on the memory elimination recursive least squares method, which is based on constructing a first-order circuit model of an energy storage battery, constructing an identification model by using the memory elimination recursive least squares method, solving the problem of data packets and by introducing a memory factor, and then forming a joint algorithm with the Kalman filter to achieve online and accurate estimation of the battery state of charge.
[0005] Technical solution: In order to solve the above technical problems, the technical solution adopted by the present invention is as follows:
[0006] A battery state of charge estimation method based on memory elimination recursive least squares method is characterized by: firstly, a first-order RC model is selected to be constructed under the premise of weighing the amount of calculation and the accuracy of the model, and then, according to the circuit time domain topology equation, a memory elimination recursive least squares method is constructed to construct an identification model through Laplace transformation, discretization, Z inverse transformation, and introduction of memory factors; based on the open circuit voltage-state of charge relationship curve, a joint algorithm is formed with Kalman filtering to complete the online estimation of the state of charge at that moment while identifying the model parameters. The following steps are included:
[0007] Step 1: Collect historical data of the energy storage battery under rated operating conditions;
[0008] Step 2: Based on the charging and discharging characteristics of the energy storage battery, a battery internal model containing resistors and capacitors is constructed, and the most practical network topology is selected through simulation analysis;
[0009] Step 3: Establish the network topology time domain equation of the first-order circuit, discretize it using the forward Euler transformation method after the Lagrange transformation, and then obtain a difference equation through Z transformation and write it in the form of least squares to construct a recursive least squares identification model;
[0010] Step 4: Based on the above least squares identification model, the forgetting factor is introduced to obtain the recursive least squares method for eliminating memory;
[0011] Step 5: The memory-eliminating recursive least squares method obtained above is combined with the unscented Kalman filter to form a joint algorithm. The joint algorithm can be used to estimate the parameters and status of the battery in real time online.
[0012] Furthermore, the vertical data in step 1 includes a terminal voltage variation curve, a terminal current variation curve, and a power variation curve in the entire process from a fully charged state to the end of discharge.
[0013] Furthermore, in step 3, the network topology time domain equation of the first-order circuit is formulated as follows in the time domain:
[0014]
[0015] u(t)=u OCV (t)+u RC (t)+R 0 I(t)
[0016] Where: u RC It represents the voltage across the parallel capacitor and resistor model in a first-order circuit;
[0017] After Laplace transformation, we get:
[0018]
[0019] u(s)=u OCV (s)+u RC (s)+R 0 I(s)
[0020] Where s represents the complex variable in Laplace transform, usually expressed as s = σ + jω, where σ is the real part and ω is the imaginary part;
[0021] Lianlide:
[0022]
[0023] Discretized by forward Euler transform method:
[0024]
[0025] Among them, T S is the sampling time, z -1 Represents a complex variable in the z-transform, which has real and imaginary parts and is often expressed in polar coordinate form.
[0026] Combining (1) and (2), we get:
[0027]
[0028] Furthermore, perform Z inverse transformation on equation (3):
[0029] u(k)=θ 1 (k)[u(k-1)-u ocv (k-1)]+θ 2 (k)I(k)+θ 3 (k)I(k-1)+u ocv (k)
[0030] Where: k represents the discrete variable of the discrete sequence obtained by z inverse transformation, k = 0, 1, 2, 3...;
[0031]
[0032] θ 2 (k) = R 0
[0033]
[0034] Written in recursive least squares form:
[0035] u(k)=θ T Φ(k)+ε(k)
[0036] θ=[θ 1 (k),θ 2 (k),θ 3(k),u OCV (k)] T
[0037] Φ(k)=[u(k-1)-u OCV (k-1),I(k),I(k-1),1] T
[0038] Where ε(k) is the model fitting residual caused by measurement and noise.
[0039] Furthermore, in step 4, based on the least squares identification model, the forgetting factor is introduced to obtain the least squares formula for eliminating memory:
[0040]
[0041] Where K(k) is the correction factor, P(k-1) is the covariance matrix at time k-1; P(k) is the covariance matrix at time k; e(k) is the forecast error, is the predicted value of terminal voltage at time k, is the estimated value of θ at time k and is given by Corrected, is the estimated value of θ at time k-1; 0<λ≤1 is the forgetting factor which is adjusted according to the actual situation.
[0042] Furthermore, in step 5, the memory-eliminating recursive least squares method and the unscented Kalman filter form a joint algorithm, and the specific contents include:
[0043] First, the battery terminal voltage is measured in the initial state, and the initial SOC value is obtained according to the OCV-SOC curve; the battery model parameters are identified by the recursive least squares method with memory elimination using the voltage, current and e(k) at a certain moment, and the parameters are substituted into the extended Kalman filter to calculate the optimal estimate of SOC at that moment, and the parameters are passed to the stored fitting function of open circuit voltage and SOC, and the e(k) and open circuit voltage at this moment are obtained using the OCV-SOC curve; then they are passed as parameters into the adaptive recursive least squares method, and then the operation at the previous moment is repeated at the next moment, thus completing the cyclic recursive estimation of lithium ion parameters and state.
[0044] Beneficial effects: Compared with the prior art, the present invention has the following advantages:
[0045] The battery state of charge estimation method based on the memory elimination recursive least squares method of the present invention establishes an imaginary first-order internal model for the battery based on the memory elimination least squares method, introduces a forgetting factor to solve the data saturation problem, and forms a joint algorithm with the Kalman filter to achieve online estimation of the battery state of charge. The present invention improves the estimation accuracy while reducing the amount of data and meeting the rapidity of calculation, further narrows the gap between the simulation model and the actual model, has good adaptability to batteries in different working states, and takes into account the real-time changes inside the battery, etc., meets the state estimation under various conditions, and enhances the possibility of practical application.
[0046] The present invention can more accurately estimate the real-time state of charge changes of the battery during the battery charging and discharging process, and estimate the future state. More accurate energy storage state information will be beneficial to the optimization and scheduling of the energy storage network, while avoiding overcharging and overdischarging, ensuring the economy and safety of the system operation while giving full play to the energy storage potential. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 It is a structural schematic diagram of a battery state of charge estimation method based on memory elimination recursive least squares method;
[0048] Figure 2 It is a first-order RC topology diagram;
[0049] Figure 3 It is the algorithm flow chart of the joint algorithm. DETAILED DESCRIPTION
[0050] The present invention is further illustrated below in conjunction with specific examples. The examples are implemented based on the technical solutions of the present invention. It should be understood that these examples are only used to illustrate the present invention and are not used to limit the scope of the present invention.
[0051] like Figure 1 As shown, the battery state of charge estimation method based on the memory elimination recursive least squares method of the present invention comprehensively considers the port voltage, current and power curves and coupling relationships during battery charging and discharging, and realizes a circuit structure that is more accurate than actual circuit characteristics.
[0052] First, under the premise of weighing the amount of calculation and the accuracy of the model, a first-order RC model is chosen to be constructed, and then the circuit time-domain topology equation is listed. After Laplace transform, discretization, Z inverse transform, introduction of memory factor and other steps, the memory elimination recursive least squares method is constructed to construct the identification model. Based on the open circuit voltage-state of charge relationship curve, a joint algorithm is formed with the Kalman filter, which completes the online estimation of the state of charge at that moment while identifying the model parameters. The present invention adopts a recursive online estimation method for each moment of battery charging and discharging. Compared with the traditional least squares method, it better reflects the dynamic characteristics of the battery, and solves the problem of huge data by introducing a memory factor. Compared with the software simulation prediction simulation and the actual output process of the battery, this model can accurately estimate the state of charge of the energy storage battery.
[0053] As shown in Title 1, this is a flow chart of the present invention, which specifically includes the following steps:
[0054] Step 1: Collect historical data of the target model energy storage battery under rated working conditions (rated temperature, rated load, etc.), including the terminal voltage change curve, terminal current change curve, power change curve, etc. from the full charge state to the end of discharge.
[0055] Step 2: Build a battery internal model containing resistors and capacitors based on the charging and discharging characteristics of the energy storage battery, and select the most practical network topology through simulation analysis.
[0056] Since the internal characteristics of the battery are different at different charging and discharging times, the actual battery system should be a nonlinear system. Therefore, the battery linear structure constructed using the Thevenin model cannot be equivalent to the actual model. To this end, several resistor-capacitor loops are added on the basis of the Thevenin linear circuit to realize the nonlinear transformation of the model to better approximate the actual battery state. Considering the compromise between model accuracy and algorithm complexity, it is chosen to add a resistor-capacitor loop to construct a first-order RC topology consisting of an ohmic internal resistance and an RC loop in series.
[0057] Specific as Figure 2 As shown, R 0 is the internal resistance of the battery; R 1 is the polarization resistance; C 1 is the polarization capacitance; u OCV (t) is the overall open circuit voltage at time t; I(t) is the current flowing through the internal resistance at time t; u(t) is the voltage at the battery terminal.
[0058] Step 3: Establish the network topology time domain equation of the first-order circuit, discretize it using the forward Euler transformation method after the Lagrange transformation, and then obtain a difference equation through Z transformation and write it in the form of least squares to construct a recursive least squares identification model.
[0059] The formula of this model in the time domain is:
[0060]
[0061] u(t)=u OCV (t)+u RC (t)+R 0 I(t)
[0062] Where: u RC It represents the voltage across the parallel capacitor and resistor model in a first-order circuit;
[0063] After Laplace transformation, we get:
[0064]
[0065] u(s)=u OCV (s)+u RC (s)+R 0 I(s)
[0066] Where s represents the complex variable in Laplace transform, usually expressed as s = σ + jω, where σ is the real part and ω is the imaginary part;
[0067] Lianlide:
[0068]
[0069] Discretized by forward Euler transform method:
[0070]
[0071] Among them, T S is the sampling time, z -1 Represents a complex variable in the z-transform, which has real and imaginary parts and is often expressed in polar coordinate form.
[0072] Combining (1) and (2), we get:
[0073]
[0074] Perform inverse Z transform on (3):
[0075] u(k)=θ 1 (k)[u(k-1)-u ocv (k-1)]+θ 2 (k)I(k)+θ 3 (k)I(k-1)+u ocv (k)
[0076] Where: k represents the discrete variable of the discrete sequence obtained by z inverse transformation, k = 0, 1, 2, 3...;
[0077]
[0078] θ 2 (k) = R 0
[0079]
[0080] Written in recursive least squares form:
[0081] u(k)=θ T Φ(k)+ε(k)
[0082] θ=[θ 1 (k),θ 2 (k),θ 3 (k),u OCV (k)] T
[0083] Φ(k)=[u(k-1)-u OCV (k-1),I(k),I(k-1),1] T
[0084] Where ε(k) is the model fitting residual caused by measurement and noise.
[0085] Step 4: Introduce the forgetting factor based on the above least squares identification model
[0086] Since the recursive least squares method makes corrections based on the previous parameter identification to obtain further results, the former is an online form of the latter and is more accurate than the latter. However, under the condition of intensive sampling time, data saturation will occur, resulting in the inability of the identification results to track parameter changes. Therefore, the forgetting factor is introduced on the basis of the above least squares identification model, and the utilization rate of new data is improved by reducing the influence of old data in the iterative process, thereby solving the problem of data saturation.
[0087] On the basis of the above least squares identification model, the forgetting factor is introduced to obtain the least squares formula for eliminating memory:
[0088]
[0089] Where K(k) is the correction factor, P(k-1) is the covariance matrix at time k-1, P(k) is the covariance matrix at time k, e(k) is the forecast error, is the predicted value of terminal voltage at time k, is the estimated value of θ at time k and is given by Corrected, is the estimated value of θ at time k-1; 0<λ≤1 is the forgetting factor which is adjusted according to the actual situation.
[0090] Step 5: In order to achieve online estimation of battery parameters and status, a joint algorithm is introduced to combine the above-obtained memory-eliminating recursive least squares method with the unscented Kalman filter.
[0091] In actual operation, the internal resistance, polarization resistance, and polarization capacitance of energy storage batteries are not constant, and have different sizes at different charging and discharging times. The traditional least squares method is to fit the various parameters of the battery equivalent model into corresponding curves according to the changes in the entire charging and discharging process. When the battery reaches the corresponding state, the model parameters are taken according to the curve. This cannot reflect the dynamic characteristics of the battery well, and the amount of data to be stored is huge, so the recursive least squares method for eliminating memory obtained above is combined with the unscented Kalman filter to form a joint algorithm. The specific algorithm flow is as follows Figure 3 As shown:
[0092] First, measure the battery terminal voltage in the initial state, and get the initial SOC value according to the OCV-SOC curve. Use the voltage, current and e(k) at a certain moment to identify the battery model parameters through the memory-eliminating recursive least squares method, substitute the parameters into the extended Kalman filter to calculate the optimal SOC estimate at that moment, and pass it to the stored fitting function of open circuit voltage and SOC, and use the OCV-SOC curve to get e(k) and open circuit voltage at this moment. Then pass it as a parameter to the adaptive recursive least squares method (see steps 1-4 for the adaptive recursive least squares method), and then repeat the operation at the next moment, thus completing the cyclic recursive estimation of lithium ion parameters and state. While identifying the model parameters, the charge state estimation at that moment is completed. It can be seen that the joint algorithm can estimate the battery parameters and state online in real time.
[0093] Kalman filtering is an algorithm that uses system models and observation data to recursively estimate system states. Extended Kalman filtering is an improved algorithm based on Kalman filtering and is an existing algorithm. It improves the robustness and accuracy of the system by increasing the complexity of the system model. Specifically, the extended Kalman filter adds one or more Gaussian noises to the Kalman filter and performs secondary filtering on the system state, thereby improving the accuracy of the estimated value.
[0094] The present invention uses historical data and measured data to fit the parameters in the second-order circuit model inside the energy storage battery, and adopts the memory elimination recursive least squares method to seek the parameter index that best meets the port circuit characteristics under a certain error to describe the electromagnetic internal structure. Taking into account the influence of external conditions such as ambient temperature and humidity and the internal aging factors of the battery, the internal second-order model is fine-tuned according to the actual situation, so that the circuit model can always accurately express the battery port output characteristics.
[0095] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.
Claims
1. A method for estimating the state of charge of a battery based on the memory elimination recursive least squares method, characterized in that: Firstly, a first-order RC model is constructed under the premise of balancing the amount of calculation and model accuracy. Then, according to the circuit time-domain topological equation, the memory elimination recursive least squares method is constructed through Laplace transform, discretization, Z inverse transform, and the introduction of memory factor to construct the identification model. Based on the open-circuit voltage-state of charge relationship curve, a joint algorithm is formed with the Kalman filter to complete the online estimation of the state of charge at that moment while identifying the model parameters.
2. The method for estimating the state of charge of a battery based on the memory elimination recursive least squares method according to claim 1, characterized in that: The following steps are involved: Step 1: Collect historical data of the energy storage battery under rated operating conditions; Step 2: Based on the charging and discharging characteristics of the energy storage battery, a battery internal model containing resistors and capacitors is constructed, and the most practical network topology is selected through simulation analysis; Step 3: Establish the network topology time domain equation of the first-order circuit, discretize it using the forward Euler transformation method after the Lagrange transformation, and then obtain a difference equation through Z transformation and write it in the form of least squares to construct a recursive least squares identification model; Step 4: Based on the above least squares identification model, the forgetting factor is introduced to obtain the recursive least squares method for eliminating memory; Step 5: The memory-eliminating recursive least squares method obtained above is combined with the unscented Kalman filter to form a joint algorithm. The joint algorithm can be used to estimate the parameters and status of the battery in real time online.
3. The method for estimating the state of charge of a battery based on the memory elimination recursive least squares method according to claim 2, characterized in that: The vertical data in step 1 includes the terminal voltage change curve, the terminal current change curve and the power change curve in the whole process from the full-charge state to the end of discharge.
4. The method for estimating the state of charge of a battery based on the memory elimination recursive least squares method according to claim 2, characterized in that: In step 3, the network topology time domain equation of the first-order circuit is formulated in the time domain as follows: u(t)=u OCV (t)+u RC (t)+R0I(t) Where: u RC It represents the voltage across the parallel capacitor and resistor model in a first-order circuit; After Laplace transformation, we get: you(s)=u OCV (s)+u RC (s)+R0I(s) Where s represents the complex variable in Laplace transform, usually expressed as s = σ + jω, where σ is the real part and ω is the imaginary part; Lianlide: Discretized by forward Euler transform method: Among them, T S is the sampling time, z -1 represents a complex variable in z-transform, which has real and imaginary parts; combining (1) and (2), we get:
5. The method for estimating the state of charge of a battery based on the memory elimination recursive least squares method according to claim 4, characterized in that: Perform Z inverse transformation on equation (3): u(k)=θ1(k)[u(k-1)-u ocv (k-1)]+ θ2(k)I(k)+θ3(k)I(k-1)+u ocv (k) Where: k represents the discrete variable of the discrete sequence obtained by z inverse transformation, k = 0, 1, 2, 3...; θ2(k)=R0 Written in recursive least squares form: u(k)=θ T Φ(k)+ε(k) θ=[θ1(k),θ2(k),θ3(k),u OCV (k)] T Φ(k)=[u(k-1)-u OCV (k-1),I(k),I(k-1),1] T Where ε(k) is the model fitting residual caused by measurement and noise.
6. The method for estimating the state of charge of a battery based on the memory elimination recursive least squares method according to claim 2, characterized in that: In step 4, based on the least squares identification model, the forgetting factor is introduced to obtain the least squares formula for eliminating memory: Where K(k) is the correction factor, P(k-1) is the covariance matrix at time k-1, P(k) is the covariance matrix at time k, e(k) is the forecast error, is the predicted value of terminal voltage at time k, is the estimated value of θ at time k and is given by Corrected, is the estimated value of θ at time k-1; 0<λ≤1 is the forgetting factor which is adjusted according to the actual situation.
7. The method for estimating the state of charge of a battery based on the memory elimination recursive least squares method according to claim 2, characterized in that: In step 5, the memory-eliminating recursive least squares method and the unscented Kalman filter form a joint algorithm, and the specific contents include: First, the battery terminal voltage is measured in the initial state, and the initial SOC value is obtained according to the OCV-SOC curve; the battery model parameters are identified by the recursive least squares method with memory elimination using the voltage, current and e(k) at a certain moment, and the parameters are substituted into the extended Kalman filter to calculate the optimal estimate of SOC at that moment, and the parameters are passed to the stored fitting function of open circuit voltage and SOC, and the e(k) and open circuit voltage at this moment are obtained using the OCV-SOC curve; then they are passed as parameters into the adaptive recursive least squares method, and then the operation at the previous moment is repeated at the next moment, thus completing the cyclic recursive estimation of lithium ion parameters and state.
