Method for estimating state of charge of lithium battery on line under working condition

Abstract

The invention discloses a method for estimating the state of charge of a lithium battery on line under working conditions, which is suitable for estimating the state of charge of the lithium battery by acquiring OCV-SOC characteristics based on a fractional order equivalent circuit model and a coevolution particle swarm algorithm, and belongs to the field of battery equivalent parameter identification. The method is suitable for obtaining the accurate open-circuit voltage so as to improve the SOC estimation precision of the table look-up method. The problems that the identification result is inconsistent and fluctuates due to the fact that the battery open-circuit voltage identification result is often associated with the working condition, and the accuracy of the SOC estimation result is not high due to the fact that the original OCV-SOC characteristic has the hysteretic characteristic are solved. According to the method, the stability of an open-circuit voltage identification result is enhanced, and in addition, linear processing is carried out on the hysteretic characteristics of the SOC generated by dynamic current integration. According to the processing results of the open-circuit voltage and the SOC, an updatable OCV-SOC mapping relation is established, and the accuracy of the mapping relation on the SOC estimation of the lithium battery is verified under different working conditions.

Description

Technical Field

[0001] The present invention relates to lithium battery power estimation, and in particular to a method for online lithium battery charge state estimation under working conditions. Background Art

[0002] Lithium-ion batteries have the advantages of high energy density, high power density and long cycle life, and are widely used in the new energy vehicle industry. Among them, open circuit voltage (OCV) is one of the key characteristic parameters of lithium-ion batteries. The relationship between open circuit voltage and state of charge (SOC) plays an important role in the estimation of the state of charge of lithium-ion batteries. By establishing an equivalent circuit model (ECM) to simulate the dynamic and static characteristics of the battery, the state estimation algorithm can be used to estimate the SOC. Using the model-based SOC estimation method, it is first necessary to identify the model parameters to obtain the model parameters. As an important parameter in the ECM, researchers have found through a large number of experiments that there is a strong linear relationship between the open circuit voltage of lithium-ion batteries and the SOC of the battery. The OCV-SOC mapping curve is used to establish a connection between the measured electrical parameters and the SOC, so that the current open circuit voltage value can be measured to map the current SOC of the battery. This method is also called the open circuit voltage method. However, due to the polarization phenomenon inside the lithium-ion battery, the battery needs to be left alone for a long enough time to eliminate the polarization phenomenon, so that the open circuit voltage of the battery can be accurately obtained. The open circuit voltage method requires obtaining the open circuit voltage value of a certain SOC state by statically placing the vehicle, so it cannot be used for online SOC estimation, and there is a problem that the research model has poor accuracy within the local SOC range. Therefore, obtaining an accurate OCV-SOC curve is crucial to improving the accuracy of SOC estimation.

[0003] When using open circuit voltage to estimate SOC, the accuracy of open circuit voltage parameter identification will directly affect the accuracy of SOC estimation. There are two types of SOC parameter identification algorithms: offline and online, corresponding to offline modeling and online modeling respectively. Offline parameter identification does not require the identification and calculation of model parameters during the SOC estimation process of the battery, and is simple and convenient. However, the model accuracy of this method will decrease under complex working conditions and after battery aging. The main offline parameter identification algorithms are the least squares algorithm. The online parameter identification method can estimate the model parameters in real time according to the specific situation of the model, avoid the shortcomings of the offline parameter identification method, and obtain higher model accuracy. However, this method has large computational complexity and high cost. The main online parameter identification algorithms are the recursive least squares algorithm, the extended Kalman filter algorithm, the particle swarm algorithm, the genetic algorithm, etc.

[0004] Due to its simple calculation and easy correction of the initial SOC, the highly reliable method of combining the ampere-hour integration method with OCV has been successfully applied to SOC estimation in BMS. The filtering technology of the state space model is used to enhance the SOC estimation. Based on the Kalman filter algorithm, a variety of filtering algorithms that can be used for nonlinear systems have been developed. Including the extended Kalman filter algorithm (EKF), the unscented Kalman filter algorithm (UKF), the volumetric Kalman filter algorithm (CKF), etc. Some scholars have proposed a new polynomial parameterization method for the OCV curve. As the lithium-ion battery ages, the parameters are adjusted to ensure the accurate fitting of the polynomial to the OCV curve. The use of the OCV-SOC model based on fractional calculus improves the accuracy of the model, which can well represent the strong nonlinear relationship between the OCV and SOC of the lithium-ion battery. However, the hysteresis characteristics of the OCV-SOC curve have an impact on the SOC estimation. Therefore, it is necessary to systematically study the OCV-SOC hysteresis characteristics of lithium-ion batteries to improve the accuracy of SOC estimation. Summary of the invention

[0005] In view of the technical problems existing in the prior art, the present invention proposes a method for online estimation of the state of charge of a lithium battery under working conditions, which has a reasonable design, overcomes the shortcomings of the prior art, and has good effects.

[0006] The present invention adopts the following technical solutions: A method for estimating the state of charge of a lithium battery online under working conditions comprises the following steps: Step 1: Use the experimental platform to perform characteristic tests on ternary lithium batteries under different working conditions to obtain test data.

[0007] Step 2: Establish a second-order fractional equivalent circuit model, input the test data into the second-order fractional equivalent circuit model and use the co-evolutionary particle swarm algorithm to calculate the open circuit voltage and parameter Δu of the second-order fractional equivalent circuit model. oc Online identification.

[0008] Step 3: Obtain the parameter Δu by adjusting the length of the parameter identification window in the co-evolutionary particle swarm algorithm oc The length of the parameter identification window that is lower than 0.1mv and has no fluctuation is the optimal parameter identification window length. Under the condition of the optimal parameter identification window length, the operating condition test data is identified to obtain the open circuit voltage data, and the charge state data is obtained by the ampere-hour integration method.

[0009] Step 4: Use the mean and filtering methods to process the open-circuit voltage data to eliminate errors. In order to avoid random errors caused by the identification results of the open-circuit voltage data, the open-circuit voltage data obtained by the conditional identification of the optimal parameter identification window length for 5 times is averaged and filtered using the recursive average filtering method. The state of charge data obtained by the ampere-hour integration method is filtered using the recursive average filtering method.

[0010] Step 5: The open circuit voltage data and state of charge data processed in step 4 form an OCV-SOC curve. Under working conditions, the state of charge data corresponding to the open circuit voltage is obtained by a table lookup method to complete the lithium battery power estimation.

[0011] Preferably, the operating conditions are FUDS conditions and DST conditions, the FUDS conditions are federal urban driving test conditions, and the DST conditions are dynamic stress test conditions.

[0012] Preferably, the second-order fractional-order equivalent circuit model utilizes two constant-phase elements to replace the capacitor in the integer-order equivalent circuit model.

[0013] Preferably, the parameter Δu oc is the change in open circuit voltage between two adjacent sampling points in the parameter identification window, and the parameter Δu oc The numerical change of can directly reflect the fluctuation of the open circuit voltage identification result.

[0014] Preferably, the optimal parameter identification window length is 1200.

[0015] The present invention has the following beneficial effects: The present invention studies a method for online lithium battery power estimation under working conditions based on a co-evolutionary particle swarm algorithm. According to the method, a selection criterion for a parameter identification window that can resist the influence of working conditions is determined, the stability of the open circuit voltage identification result is enhanced, and the hysteresis effect of the OCV-SOC characteristic generated by the current fluctuation under dynamic working conditions is smoothed, thereby increasing the engineering application value. It is suitable for obtaining accurate open circuit voltage to improve the SOC estimation accuracy of the table lookup method. It solves the problem that the battery open circuit voltage identification result is often associated with the working conditions, resulting in inconsistency and fluctuation in the identification results, and the original OCV-SOC characteristic has hysteresis characteristics, resulting in low accuracy of the SOC estimation result. An updateable OCV-SOC mapping relationship is established based on the processing results of the open circuit voltage and SOC, and the accuracy of the mapping relationship for the SOC estimation of the lithium battery, that is, the accuracy of the lithium battery power estimation, is verified under different conditions. The results show that the SOC estimation values ​​of the processed OCV-SOC curve under different working conditions can accurately follow the true value of the SOC, have good accuracy and stability, and the SOC estimation value can quickly converge to the vicinity of the true value in any interval, with good consistency. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a fractional-order equivalent circuit model.

[0017] Figure 2 This is the principle diagram of the CPSO algorithm.

[0018] Figure 3 This is the OCV-SOC curve calibration of the small current charge and discharge method; (a) is the OCV-SOC curve at different temperatures, and (b) is the OCV-SOC curve at 25 degrees.

[0019] Figure 4 (a) is the load current diagram of the DST condition, and (b) is the load current diagram of the FUDS condition.

[0020] Figure 5 (a) shows the test data and OCV identification results of the DST condition, and (b) shows the test data and OCV identification results of the FUDS condition.

[0021] Figure 6 (a) is the original OCV-SOC curve under two working conditions, and (b) is the hysteresis characteristic in the original OCV-SOC mapping relationship.

[0022] Figure 7 is the impact of load current fluctuation on the original SOC.

[0023] Figure 8 (a) is the OCV identification result under different window length conditions, and (b) is a local enlarged view of the OCV identification result under different window length conditions.

[0024] Fig. 9 (a) is Δu under different identification window length conditions oc Identification results, (b) Δu under different window length conditions oc A partial enlarged view of the identification results.

[0025] Fig.10 (a) is the Δu corresponding to the high SOC range oc Identification results, (b) is the Δu corresponding to the middle SOC interval oc Identification results, (c) is the Δu corresponding to the low SOC interval oc Identify the results.

[0026] Fig.11 This is the process of recursive average filtering.

[0027] Fig.12 (a) shows the 5 OCV identification results and their average value, and (b) shows the SOC filtering result.

[0028] Fig.13 (a) is the OCV average filtering result, and (b) is the filtering error.

[0029] Fig.14 This is the OCV-SOC curve after treatment.

[0030] Fig.15 (a) is the comparison of the estimated SOC values ​​under DST conditions, and (b) is the estimation error of SOC.

[0031] Fig.16 (a) is the comparison of the estimated SOC values ​​under FUDS conditions, and (b) is the estimation error of SOC.

[0032] Fig.17 (a) is the comparison of SOC estimation under five FUDS conditions, and (b) is a partially enlarged view of the comparison of SOC estimation under five FUDS conditions.

[0033] Fig.18 Figure 2 is the convergence of SOC estimation under different initial values, where (a) shows the SOC estimation result when the initial value is 80%, and (b) shows the SOC estimation result when the initial value is 60%.

[0034] Fig.19 (a) is the comparison of SOC estimation value with the small current method, and (b) is the SOC estimation error. DETAILED DESCRIPTION

[0035] The specific implementation of the present invention is further described below in conjunction with the accompanying drawings and specific embodiments: The electrochemical impedance spectrum of the ternary lithium battery has obvious fractional-order characteristics. Therefore, we use components with fractional-order characteristics to replace the capacitors in the integer-order circuit and construct a fractional-order equivalent circuit model to describe the impedance characteristics of the battery, which can accurately extract the battery state and parameters.

[0036] Second-order fractional equivalent circuit model: The present invention uses two constant phase elements CPE1 and CPE2 to replace the two integer capacitors of the equivalent circuit model to establish a fractional equivalent circuit model (FECM) as shown in Figure 1 shown.

[0037] exist Figure 1 in, u oc is the open circuit voltage of the lithium battery, u t is the terminal voltage, R0 is the ohmic internal resistance, CPE1 and CPE2 are two fractional capacitors with values ​​of C1 and C2, R1 and R2 are the parallel resistances of the fractional capacitors, and i is the load current. In order to reduce the computational consumption and the subsequent identification time, the fractional orders α and β of the constant phase elements CPE1 and CPE2 are determined as fixed values.

[0038] After a large number of experiments and data analysis, the fractional orders α and β are set to 0.7 and 0.9 respectively, so as to reduce the computational consumption and shorten the time in the subsequent identification process.

[0039] The characteristics of the co-evolutionary particle swarm algorithm (CPSO) are: in order to solve the problem of low real-time performance of the bionic optimization algorithm, the particle swarm algorithm and the parameter identification window (PIW) are combined to form a co-evolutionary particle swarm algorithm. By advancing the parameter identification window, the online identification of open circuit voltage and other equivalent parameters is realized.

[0040] The length of the parameter identification window (PIW), as a key parameter of CPSO, affects the stability and consistency of the equivalent parameter identification results. The algorithm is tested by setting different values ​​to achieve the effect of the operating conditions on the open circuit voltage by adjusting the length of the parameter identification window. It has been verified that under the DST operating condition standard, the time period for each operating condition test is 360 seconds, so the number of data points sampled under a time length of 360s is 720. The length of the parameter identification window affects the optimization results of the particle swarm and thus affects the identification results. The longer the window length, the more data information is stored, and it tends to express the steady-state characteristics of the battery. Therefore, a longer parameter identification window is conducive to obtaining stable parameter identification results. In order to better analyze the relationship between the length of the parameter identification window and the open circuit voltage fluctuation, Δu is added to the parameter set identified by the CPSO algorithm. oc Parameters. Δu oc It is the change of the open circuit voltage between two adjacent sampling points in the parameter identification window. Parameter Δu oc The numerical change of can directly reflect the fluctuation of OCV identification results. The analysis shows that when the window length is longer than the data length within the time period of a DST condition, the algorithm has a better suppression effect on the fluctuation of the open circuit voltage identification results caused by the condition. Therefore, the window length should be greater than 720. However, the longer the window length, the longer the identification time will be. When the window is too long, the algorithm will no longer have a significant suppression effect on the fluctuation when the open circuit voltage identification value is relatively smooth. It has been verified that when the parameter identification window length is 900-1200, the algorithm has a more obvious suppression effect on the fluctuation and the open circuit voltage identification result is relatively smooth.

[0041] CPSO uses PIW with a window length of w to save a small piece of data before the current moment of the battery, and identifies the equivalent parameters of the battery in PIW. PIW moves with the current sampling data point, and CPSO completes an evolutionary step each time PIW moves. In each evolutionary step, CPSO only identifies and updates one parameter in the parameter set P. As the sampling point steps, PIW stores the latest data and discards the oldest data. CPSO identifies and updates each parameter in P in sequence. When the parameters in P are identified, the identification starts from OCV in a loop until PIW moves to the last original data point. Figure 2 The schematic diagram of CPSO principle is given. In the figure, PIW(n) represents the current identification window, PIW(n-1) represents the previous identification window; L S Indicates the prediction step size of the window forward; ut(kL S ) and ut(L S ) represent the voltage data loaded within the prediction step.

[0042] Example 1, the battery used for testing is a ternary lithium battery, the lithium battery model tested is NCR18650BE, and the test is carried out at a constant temperature of 25°C. The detailed parameters are shown in Table 1. The experimental platform adopts the main controller model STM32F407ZGT6, and the DST load condition is defined by the 12-bit DA converter integrated in the MCU, and the charge and discharge current is controlled by the power amplifier to provide a programmable LoadProfile to the battery. ADS1274 is used to collect the terminal voltage and current of the battery. The collected data is pre-processed by the MCU and sent to the PC (Intel Core i7-7500U CPU, 2.70GHz, 12GB, 64bit, Windows 10) to save the experimental data. After the original experimental data is tested and stored, MATLAB ((R2020b) is used to verify the parameter identification algorithm.

[0043] Table 1 Technical parameters of ternary lithium battery

[0044] Lithium battery characteristic test: (1) Low current charge and discharge test. Low current charge and discharge uses a current of C / 20 to charge and discharge the battery. At a very low discharge rate, the polarization effect of the battery is very weak. The average voltage of the discharge and charge process is recorded as OCV. The average voltage can reduce the hysteresis effect of the battery and the influence of the ohmic internal resistance.

[0045] At a fixed temperature, using the terminal voltage curve measured by charging and discharging, the OCV-SOC curve at different temperatures is as follows Figure 3 The OCV-SOC curve at 25 degrees Celsius is obtained by interpolating the median value. Figure 3As shown in (b).

[0046] (2) Dynamic operating condition test. The dynamic operating condition test of lithium batteries is carried out according to the Federal Urban Driving Schedule (FUDS) and Dynamic Stress Test (DST) standards in the "USABC Battery Experimental Method Manual". The test data is saved in the database. The data points stored in the database are sorted according to the sampling time. The actual data sampling is 0.5s, which is used for algorithm analysis and verification; the FUDS operating condition is the standard time-speed curve of urban driving vehicles in the automotive industry, and the time cycle for one completion is 1372 seconds. The DST operating condition is a simplified version based on the power-time requirements in FUDS, and the time cycle for one completion is 360 seconds. The load current under the two dynamic electrical conditions is as follows Figure 4 shown.

[0047] Figure 5 Indicates the battery discharge test results under two dynamic conditions. Figure 5 (a) is the DST working condition test curve. Figure 5 (b) is the FUDS working condition test curve, where the red line is the OCV result identified by the CPSO algorithm.

[0048] Under dynamic working conditions, the identified OCV-SOC curve has hysteresis characteristics, which affects engineering applications. The commonly used OCV-SOC calibration methods in engineering, including the HPPC method and the low current charge and discharge method, usually obtain a monotonic OCV-SOC mapping relationship. However, the CPSO algorithm is based on the standard particle swarm algorithm. When testing the algorithm, the part with large operating fluctuations will show hysteresis characteristics. Figure 5 It shows the identification results of OCV under DST and FUDS conditions using the second-order fractional equivalent circuit model and the typical CPSO algorithm structure characteristics. Due to the different discharge times of the two conditions, it is difficult to compare their characteristics using time as the horizontal axis, so the OCV-SOC curves of the two conditions are used for comparative analysis. Figure 6 (a) shows the OCV-SOC identification results of two working conditions. Figure 6 (b) indicates that there is a hysteresis characteristic in the OCV-SOC identification result. The hysteresis characteristic refers to the situation where multiple OCV values ​​in the OCV-SOC curve correspond to the same SOC value, which will result in the SOC value obtained when using the table lookup method for SOC estimation to be non-unique, and thus it is impossible to obtain an accurate SOC value. This not only reduces the estimation accuracy of the SOC value, but also greatly weakens the engineering application value of the OCV-SOC curve.

[0049] The open circuit voltage is related to the operating conditions and algorithm conditions: (1) The operating conditions affect the identification results of the open circuit voltage. Figure 5 It can be seen that OCV can change with the terminal voltage u t However, under different working conditions, the OCV identification results will change. This is because Figure 1 In the circuit shown, OCV is converted into u due to the discharge current and the internal resistance of the battery. t Therefore, the open circuit voltage is affected by the working conditions and has a certain relationship with the terminal voltage working conditions. In the dynamic working condition test, the battery open circuit voltage data cannot be directly obtained, so the change of the terminal voltage can reflect the influence of the working conditions on the battery open circuit voltage.

[0050] The fluctuation characteristics of the working condition are the main reason affecting the local hysteresis characteristics of the OCV-SOC curve. Under the typical algorithm test scheme, the two working conditions obtained highly consistent estimation results. Due to the different cycle periods and current distribution intensities of the DST and FUDS working conditions, Figure 6 The OCV-SOC identification results in (a) also show the corresponding fluctuation period. Figure 6 (b) shows the part where the OCV-SOC of the two working conditions differ greatly. The blue box shows the open circuit voltage difference at the same SOC, with a maximum of about 7mV. The yellow box shows the SOC difference at the same OCV, with a maximum difference of about 0.7. The differences in both places are within a reasonable range. Therefore, it can be considered that the OCV-SOC curves of the DST working condition and the FUDS working condition have a high consistency. In addition, Figure 6 In (b), both OCV-SOC curves show obvious hysteresis characteristics. On the one hand, the SOC of the battery is calculated using the ampere-hour integration method in the test. Since the load current has dynamic characteristics, when the current is negative, the SOC decreases; when the current is positive, the SOC increases. Therefore, the final SOC is related to the load current, that is, while the overall decrease is accompanied by a periodic fluctuation characteristic. The impact of load current fluctuation on SOC is shown in Figure 2. Figure 7 As shown. On the other hand, the OCV identification results also show fluctuation characteristics related to the algorithm and working conditions. In summary, the above two aspects are the main reasons why the OCV-SOC curve under dynamic conditions has local hysteresis characteristics.

[0051] The OCV-SOC curve will change with the temperature. Figure 3 It can be seen that as the SOC increases, the OCV of the battery increases monotonically from 2.7V to 4.2V. In the 20%-80% SOC range, the temperature has no obvious effect on the OCV-SOC curve. At both ends of the SOC, the effect of low temperature on the ternary lithium battery is more obvious, and the OCV-SOC curve at 15°C has a significant shift compared to that at 25°C and 35°C.

[0052] (2) The open circuit voltage identification result is related to the algorithm conditions, and the parameter identification window length affects the smoothness of the OCV identification result. The OCV identification result is related to the algorithm conditions. The length of the parameter identification window (PIW), as a key parameter of CPSO, affects the stability and consistency of the equivalent parameter identification result. The algorithm is tested by setting the PIW length of 300-1200, and the influence of the PIW weakening condition on the open circuit voltage is discussed. The identification results of the open circuit voltage under different PIWs are shown in Figure 2. Figure 8 shown.

[0053] The longer the parameter identification window is, the more it can suppress the open circuit voltage fluctuation and make the open circuit voltage result smoother. Figure 8 (a) shows the open circuit voltage identification results corresponding to different PIW lengths. When PIW is 300-1200, the consistency of the open circuit voltage is generally high. From the enlarged figure, it can be seen that the longer the PIW, the more stable the open circuit voltage identification results, especially at the end of battery discharge, the shorter the PIW, the more obvious the open circuit voltage fluctuation. Figure 8 (b) is a local enlarged view. In order to explore the influencing factors of open circuit voltage fluctuation, the terminal voltage u t The results show that the fluctuation period of the open circuit voltage is basically consistent with the operating period of the working condition, especially when the PIW length is 300 and 600, the fluctuation of the open circuit voltage increases significantly when the working condition changes drastically. However, as the PIW becomes longer, the identification result of the open circuit voltage almost becomes a uniform decrease, and the correlation with the working condition fluctuation is weakened. That is, the longer the PIW, the more it can suppress the fluctuation of the open circuit voltage caused by the working condition change.

[0054] In order to better analyze the relationship between PIW length and open circuit voltage fluctuation, Δu is added to the parameter set identified by the CPSO algorithm. oc Parameters. Δu oc It is the change of OCV between two adjacent sampling points in the parameter identification window.

[0055] Set the PIW length to 300-1200, test the algorithm while keeping other algorithm parameters the same, and obtain Δu oc Parameters such as Fig. 9 shown. Fig. 9 (a) is the Δu obtained by identifying the complete working condition test data under the PIW with a length of 300-1200 oc Parameters. From Fig. 9 In (a), it can be seen that in the entire working condition test, the Δu identified under the condition of PIW=300 is oc The values ​​are basically the largest, which reflects that the OCV value fluctuation obtained under this condition is the most obvious. Fig. 9Middle (b) is a local enlarged view in the middle stage of identification when the data identification results are relatively stable. Fig. 9 In (b), as the identification window length increases, it can be seen that Δu oc The value of gradually decreases, indicating that as the identification window length increases, the change in OCV between two adjacent sampling points becomes smaller, that is, the fluctuation of OCV identification results decreases, which is also consistent with Figure 8 The longer the window length, the smoother the OCV recognition result and the more consistent the features. Fig. 9 It can also be seen from (b) that when PIW=900, Δu oc The value of is already small, stably distributed around 0.1mv, compared with Δu of PIW=300 and 600 oc When PIW=1200, Δu oc The value is stable below 0.1mv and has little fluctuation. Figure 8 Correspondingly, the OCV identification result at this time is already very smooth, meeting the requirements for constructing a smooth OCV-SOC characteristic.

[0056] According to the analysis of the original OCV identification results in the previous article, at the end of discharge, the open circuit voltage fluctuation in the low SOC section is the most intense, so it is necessary to analyze the different SOC intervals Δu oc of changes. Fig.10 Shows the Δu corresponding to different SOC intervals oc Changes. Fig.10 In (a), it can be seen that in the low SOC section, when PIW=300, Δu oc The value of is large, and the maximum value exceeds 0.5mv, which is consistent with the characteristics of the violent fluctuation of the OCV identification results in the low SOC section at the end of discharge. Under this condition, the fluctuation error in the identification process significantly reduces the accuracy of the open circuit voltage. However, as the window length increases, when PIW=1200, the Δu in the low SOC section oc The value can still be stably distributed within 0.1mv. This proves that even in the low SOC range, the increase in PIW length still has a significant effect on suppressing the open circuit voltage fluctuation caused by operating condition changes. Fig.10 (b) and (c) show the Δu corresponding to the medium and high SOC sections respectively. oc Changing trends. Fig.10 It can be seen from (b) and (c) that in the medium and high SOC sections, Δu oc Although the value of is lower than that of the low SOC section, the data value obtained when PIW=300 is still much larger than the values ​​obtained under other window conditions. In the three SOC regions, the Δu oc The value is not only always lower than 0.1mv but also fluctuates stably.

[0057] By comparing OCV and Δu oc Analysis of the identification results. For the window length PIW, it can be seen that when its value is 900, the identified OCV and Δu oc The fluctuation of the value has been significantly reduced; when its value is 1200, Δu oc The value is basically stable, and the OCV value obtained by identification is basically smooth as a whole. The reason why a longer window length can suppress the open circuit voltage fluctuation caused by the change of load current is that there are more data points identified in a longer window, the data information is more sufficient, and the particles in the algorithm are more likely to maintain the consistency of the optimal particle position gbest when searching for the best. The optimal particle position affects the update of the model parameter value obtained by identification, so a longer window is more likely to obtain parameter identification results with slow fluctuations and higher smoothness.

[0058] The OCV data is processed using methods such as averaging and filtering to eliminate errors and increase the smoothness of the OCV data. The CPSO algorithm is an improvement on the standard particle swarm algorithm. Bionic optimization algorithms such as PSO seek the optimal solution among many parameter values. Due to the nonlinear characteristics of lithium batteries and the fitness function of the CPSO algorithm itself, the optimization results of the parameters are not completely consistent, but fluctuate within a reasonable range. The experiment tested the identification results of the equivalent parameters of lithium batteries by the CPSO algorithm 5 times. The OCV identification results of the 5 experiments have a certain error, and the maximum difference is approximately within 0.5mV. The identification results of OCV under DST conditions are continuously decreasing.

[0059] Since the actual working state of the battery is very complex and will be affected by the surrounding environment, the design of the battery SOC estimator must ensure the SOC estimation accuracy under more complex working conditions. Therefore, the accuracy of the OCV-SOC mapping relationship is crucial to the SOC estimation based on the table lookup method.

[0060] The open circuit voltage data obtained by five identifications under the condition of the optimal parameter identification window length are averaged. Fig.12 (a) is a local enlargement of the average value of OCV and the original value of five tests. When testing the original data, the CPSO algorithm uses the terminal voltage fitting error as the fitness function. When each particle approaches the optimal solution, there is a probability of obtaining a similar solution. Therefore, the OCV identification results will have the same value. When calibrating the OCV-SOC mapping relationship, if the same OCV value appears, there will be a situation in the lookup table where one OCV value corresponds to multiple SOC values, which will increase the error. Therefore, it is necessary to further process the SOC data and OCV data to retain the original data trend while trying to avoid the same value in the constructed OCV-SOC mapping relationship.

[0061] The recursive average filtering method is used to process the SOC data and the averaged OCV data. The CPSO algorithm has high stability, and the identified OCV results rarely have large deviations. Therefore, the recursive average filtering method is used to process the OCV data. Fig.11 As shown in the figure, this method follows the first-in-first-out principle when reading data. A queue of length N is used as the OCV data storage. As the queue continues to move forward, the oldest data at the head of the queue is discarded and the latest data is placed at the end of the queue, so that there is always the "latest" data in the queue. When calculating the arithmetic mean, you only need to perform arithmetic averaging on the N OCV data in the queue to get a new arithmetic mean.

[0062] The load current of the DST condition changes frequently. In order to avoid the hysteresis characteristics caused by current integration when calibrating the OCV-SOC mapping relationship, the SOC obtained by the ampere-hour integration method is filtered. When the battery is discharged, the SOC of the battery decreases continuously. Since the dynamic conditions tested all have periodic characteristics, it can be considered that the SOC of the DST condition decreases continuously and evenly under certain fluctuations. When conducting dynamic condition tests, the sampling frequency of the battery data is 100Hz. Due to the small sampling frequency, although the original SOC data fluctuates, the overall trend is stable and the fluctuation value is small. Therefore, the recursive average filtering method, which can effectively suppress periodic interference, make the signal smoother and maintain the basic shape of the signal, is used to process the OCV data and SOC data. The selection of the filter window length N is determined by the operating condition characteristics. A cycle of the DST condition is 360 seconds, and the number of sampling data points in a cycle under the 0.5s sampling condition is 720. Therefore, the data points in the filter window should be selected as 720, which is consistent with the number of data points in a single cycle of the condition. Fig.12 (b) shows the filtered SOC data.

[0063] The processed reference characteristics have representative value. The OCV processing results are as follows Fig.13 As shown, Fig.13 (a) shows that the processed OCV results maintain the dynamic trend of the original data. Fig.13 It can be seen from the enlarged figure in (a) that the recursive average filtering method can effectively suppress the periodic fluctuations of the OCV results caused by the periodic changes of dynamic working conditions and increase the smoothness of the OCV data. Fig.13 The middle (b) figure shows the error between the original data and the filtered data, which is basically within 0.5mV.

[0064] The OCV-SOC curve after treatment is as follows Fig.14As shown, we can see that the SOC range is 4%-98%, and the OCV range is 3.2V -4.2V, which basically covers the normal use range of ternary lithium batteries and has representative value.

[0065] The processed reference characteristics can improve the accuracy of SOC estimation (applicability verification). The SOC estimation values ​​of the processed OCV-SOC curve under DST and FUDS working conditions can accurately follow the true value of SOC. The SOC estimation errors under the two working conditions are within 0.5% and 1% respectively, with good accuracy and stability.

[0066] Under DST conditions, the accuracy of SOC estimation based on reference characteristics is verified by table lookup method. It is verified that the stability of SOC estimation is good and the estimation accuracy is high. The estimation results are as follows: Fig.15 As shown, Fig.15 (a) represents the true value and estimated value of SOC, Fig.15 (b) represents the error between the estimated SOC value and the true value. The results show that under the DST condition, the estimated SOC value can accurately follow the true value of the SOC. Since the OCV-SOC mapping relationship in the lookup table is obtained by processing the OCV results of the DST condition identification, the SOC estimation accuracy under the DST condition is relatively high, with an error within 0.5%. In addition, when the SOC drops to about 10%, the error increases significantly. There are two main reasons: First, at the end of the discharge of the ternary lithium battery, the internal polarization of the battery is more intense and the terminal voltage changes faster; second, during the OCV preprocessing, the error between the filtered OCV and the original value increases significantly, which also leads to an increase in the SOC estimation error at the end of the discharge.

[0067] The OCV-SOC calibrated by the CPSO algorithm has good applicability and can be used under various working conditions. It has been verified that the SOC estimation error under FUDS conditions is within 1%. Fig.16 It represents the SOC estimation result of OCV lookup table identified by CPSO algorithm under FUDS condition. Fig.16 (a) is the comparison between the estimated SOC value and the true value. Fig.16 (b) is the SOC estimation error. Fig.16 It can be seen in (a) that under the FUDS condition, the estimated value of SOC can still accurately follow the true value of SOC. Fig.16 (b) shows the estimated error of SOC under FUDS condition. The error is basically within 1% and shows certain periodic characteristics, which is related to the periodic fluctuation of the true value of SOC obtained by integrating the load current.

[0068] Under FUDS conditions, the average error of SOC estimation in multiple tests does not exceed 0.3%, and the stability is good. In order to avoid contingency, CPSO was used to identify the equivalent parameters of 5 FUDS conditions, and the SOC estimation values ​​of 5 FUDS conditions were obtained by looking up the table. Fig.17 As shown, Fig.17 (a) shows the estimated and true SOC values ​​of the five FUDS conditions. It can be seen that the estimated SOC values ​​of the five FUDS conditions can accurately follow the true value of SOC. Table 2 lists the average error and root mean square error of the five SOC estimates under the FUDS condition. The average error is less than 0.25%, and the root mean square error is less than 0.35%. Within a reasonable error range, the CPSO algorithm is verified for the accuracy and consistency of OCV identification and the feasibility of table-based SOC estimation.

[0069] Table 2 SOC estimation error of FUDS condition

[0070] The SOC identification results have good convergence and consistency. The SOC estimation can quickly converge to the true value in any interval, with good consistency and stability. In any discharge test interval, the SOC estimation results converge quickly, with a convergence error of less than 1% within 10 seconds. In order to explore the convergence of SOC estimation under algorithm restart or different initial conditions, the DST working condition was used to test three times under the initial SOC conditions of 80% and 60%. The results are as follows: Fig.18 shown. Fig.18 (a) in the figure shows the SOC estimation result under DST condition, where the initial value is 80%. Fig.18 (b) shows the SOC estimation result when the initial value is 60%. It can be seen that the SOC estimation can quickly converge to the true value in any interval, and the consistency and stability of the SOC estimation are good.

[0071] The OCV-SOC mapping relationship estimated by the CPSO algorithm under DST working conditions is similar to that measured by the low current method. Based on the results of the low current test, the OCV is brought into the OCV-SOC table calibrated by the CPSO algorithm, and the results are as follows: Fig.19 As shown. Fig.19 (a) is the test comparison result. Fig.19 (b) is the error between the two.

[0072] Under two different OCV-SOC calibration systems, the maximum error between the two does not exceed 1%. The accuracy of the OCV identified by CPSO and the accuracy of the OCV-SOC curve calibrated using the identified OCV were verified. In addition, as one of the most basic methods for estimating the SOC of lithium batteries, the open circuit voltage method will be affected by the cycle aging of the battery, and the OCV-SOC mapping relationship needs to be continuously updated. The CPSO algorithm can identify parameters such as OCV in real time during the operation of the battery, and the updated OCV-SOC curve can be obtained by filtering the identified OCV results. Therefore, the OCV-SOC mapping relationship can also be understood as being updated in real time, which can effectively improve the problems such as the decrease in SOC estimation accuracy caused by the change in the OCV-SOC mapping relationship due to battery aging.

[0073] The present invention studies a method for online estimation of the state of charge of a lithium battery under working conditions. The method is suitable for estimating the state of charge of a lithium battery by obtaining OCV-SOC characteristics based on a fractional-order equivalent circuit model and a co-evolutionary particle swarm algorithm. The method belongs to the field of battery equivalent parameter identification. The method is suitable for obtaining accurate open circuit voltage to improve the SOC estimation accuracy of the table lookup method. The problem that the open circuit voltage identification results of the battery are often associated with the working conditions, resulting in inconsistency and fluctuation in the identification results, and the original OCV-SOC characteristics have hysteresis characteristics, resulting in low accuracy of the SOC estimation results. Based on the CPSO algorithm, the correlation between the open circuit voltage identification results of the lithium battery and the working conditions and the weakening method are studied. According to this method, the stability of the open circuit voltage identification results is enhanced, and the hysteresis characteristics of the SOC generated by the dynamic current integration are linearized. An updateable OCV-SOC mapping relationship is established based on the processing results of the open circuit voltage and SOC, and the accuracy of the mapping relationship for lithium battery SOC estimation is verified under different conditions. The results show that:

[0074] (1) Dynamic operating conditions have a clear impact on the open circuit voltage and SOC values. The open circuit voltage identified by the CPSO algorithm shows obvious fluctuation characteristics related to the operating cycle. The longer the parameter identification window length of the CPSO algorithm, the weaker the fluctuation characteristics of the open circuit voltage identification results and the higher the smoothness.

[0075] (2) The hysteresis characteristics in OCV-SOC are eliminated, and a smooth and updateable OCV-SOC reference characteristic with high accuracy is established. The SOC estimation error under two operating conditions is within 1%.

[0076] (3) The convergence of SOC estimation is consistent with that of OCV identification results. Using the reference characteristics to estimate SOC through the table lookup method, the estimation results are consistent, converge quickly, and do not depend on the initial value.

[0077] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the essential scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A method for estimating the state of charge of a lithium battery online under working conditions, characterized in that: The following steps are involved: Step 1: Use the experimental platform to perform characteristic tests on ternary lithium batteries under different working conditions to obtain test data; Step 2: Establish a second-order fractional equivalent circuit model, input the test data into the second-order fractional equivalent circuit model and use the co-evolutionary particle swarm algorithm to calculate the open circuit voltage and parameter Δu of the second-order fractional equivalent circuit model. oc Online identification; Step 3: Obtain the parameter Δu by adjusting the length of the parameter identification window in the co-evolutionary particle swarm algorithm oc The length of the parameter identification window that is less than 0.1mv and has no fluctuation is the optimal parameter identification window length. Under the condition of the optimal parameter identification window length, the operating condition test data is identified to obtain the open circuit voltage data, and the state of charge data is obtained by the ampere-hour integration method; Step 4: Use the mean and filtering methods to process the open circuit voltage data to eliminate errors; in order to avoid random errors caused by the identification results of the open circuit voltage data, the open circuit voltage data obtained by the conditional identification of the optimal parameter identification window length for 5 times is averaged and filtered using the recursive average filtering method, and the state of charge data obtained by the ampere-hour integration method is filtered using the recursive average filtering method; Step 5: The open circuit voltage data and state of charge data processed in step 4 form an OCV-SOC curve. Under working conditions, the state of charge data corresponding to the open circuit voltage is obtained by a table lookup method to complete the lithium battery power estimation.

2. The method for online estimation of the state of charge of a lithium battery under working conditions according to claim 1, characterized in that: The operating conditions are FUDS conditions and DST conditions. The FUDS conditions are federal urban driving test conditions, and the DST conditions are dynamic stress test conditions.

3. The method for estimating the state of charge of a lithium battery online under working conditions according to claim 1, characterized in that: The second-order fractional equivalent circuit model utilizes two constant phase elements to replace the capacitor in the integer-order equivalent circuit model.

4. The method for online estimation of the state of charge of a lithium battery under working conditions according to claim 1, characterized in that: The parameter Δu oc is the change in open circuit voltage between two adjacent sampling points in the parameter identification window, and the parameter Δu oc The numerical change of can directly reflect the fluctuation of the open circuit voltage identification result.

5. The method for online estimation of the state of charge of a lithium battery under working conditions according to claim 1, characterized in that: The optimal parameter identification window length is 1200.

Images