Battery maximum available capacity estimation method based on actual operation condition of energy storage power station

Abstract

The invention discloses a battery maximum available capacity estimation method based on the actual operation condition of an energy storage power station, and relates to the technical field of lithium battery energy storage, and the method comprises the following steps: carrying out a battery life test on an energy storage lithium battery to collect data information of the lithium battery; according to voltage segments of operation of the energy storage power station, dividing the voltage into a plurality of voltage segments on the charging data, and performing feature information extraction; selecting the predicted maximum available capacity characteristics of the battery in each voltage segment according to the extracted characteristic information; constructing a maximum available capacity estimation model of the battery by adopting a Gaussian process regression (GPR) model, and training the estimation model; and inputting the predicted maximum available capacity characteristic of the battery into the trained estimation model to obtain a maximum available capacity estimation value of the battery. Under the condition of crossing multiple intervals, Bayesian filtering is used for iterating multiple model outputs, the maximum available capacity estimation result of the battery is obtained, the maximum available capacity of the energy storage battery can be estimated in real time, good estimation precision is achieved, and powerful data support is provided for safe and reliable operation and maintenance of an energy storage power station.

Description

Technical Field

[0001] The present invention relates to the field of lithium battery energy storage technology, and in particular to a method and system for estimating the maximum available capacity of a battery based on actual operating conditions of an energy storage power station. Background Art

[0002] The energy storage system is the basis for the smart microgrid to achieve smooth transition of electric energy, load balancing, frequency and voltage regulation, and peak shaving and valley filling. It is an important technology to improve the quality and efficiency of electric energy and thus solve the future energy security and environmental pollution. As a clean energy, lithium batteries are widely used in energy storage power stations due to their high output voltage, long life, and high energy density. The performance of lithium-ion batteries is reflected in the maximum available capacity of the battery, which will decrease with use and is closely related to the service life and safety of the battery. In order to ensure the safety and reliability of the energy storage system and improve the economic benefits of the energy storage power station, it is necessary to monitor the maximum available capacity of the battery in real time.

[0003] Since the maximum available capacity of a battery cannot be measured directly, the industry usually uses the ampere-hour integration method and the model-based method. The ampere-hour integration method is highly dependent on the accuracy of the current sensor, and it is easy to have error accumulation, resulting in inaccurate estimation of the maximum available capacity of the battery. The defect of the model-based method is that the parameters of the model are often determined under fixed operating conditions, which makes it unsuitable for the variable operating conditions of energy storage power stations. In addition, the electrochemical impedance spectroscopy (EIS) method can only measure the maximum available capacity of the battery when the battery is stationary, and cannot be applied to real-time measurements when the battery is in operation.

[0004] The advantage of the data-driven method is that it is suitable for real-time battery usage scenarios. By collecting the characteristic information of the battery on site, mining the implicit information on the maximum available capacity of the battery and its evolution law, the maximum available capacity of the battery is predicted, avoiding modeling and parameter identification problems. For the data-driven method, the selection and appropriate feature input and machine learning algorithm are the key. The currently used data-driven battery maximum available capacity estimation method extracts feature inputs based on the battery full cycle data. However, considering that under the working conditions of the energy storage battery, the battery charging and discharging data is often not a full cycle, and the amount of information is incomplete. Therefore, this application designs a method for estimating the maximum available capacity of the battery based on the actual operating conditions of the energy storage power station. Summary of the invention

[0005] In view of one or more of the above-mentioned existing problems, the present invention proposes a method for estimating the maximum available capacity of a battery based on the actual operating conditions of an energy storage power station.

[0006] According to a first aspect of the present application, a method for estimating the maximum available capacity of a battery based on the actual operating conditions of an energy storage power station is provided, the estimation method comprising the following steps:

[0007] S100. Performing a battery life test on the energy storage lithium battery to collect lithium battery data information, wherein the lithium battery data information includes the voltage, current, temperature, and maximum available capacity of each charge and discharge cycle of the lithium battery;

[0008] S200. According to the voltage segment of the energy storage power station operation, the voltage is divided into multiple voltage segments on the charging data, and the characteristic information of the lithium battery data information of each voltage segment is extracted;

[0009] S300. Selecting the predicted maximum available capacity feature of the battery in each voltage segment using the Deep Lasso algorithm based on the extracted feature information;

[0010] S400. Using a Gaussian process regression (GPR) model and selecting a corresponding kernel function to construct a battery maximum available capacity estimation model, and using the acquired battery maximum available capacity feature data and the corresponding battery maximum available capacity data to complete the training of the estimation model;

[0011] S500. Determine the voltage segment range covered by the energy storage battery according to the data information of the energy storage battery in actual operation, input the predicted maximum available capacity characteristics of the battery into the trained estimation model, and obtain the estimated value of the maximum available capacity of the battery.

[0012] As a preferred method that can be realized, step S100. performs a battery life test on the energy storage lithium battery to collect lithium battery data information, wherein the data information includes the voltage, current, temperature, and maximum available capacity of each charge and discharge cycle of the lithium battery. The specific steps include:

[0013] S101 adopts a constant current and constant voltage charging method to charge the lithium battery, and collects the battery voltage, current and temperature values ​​during the process at a certain time interval;

[0014] S102. After standing still, discharge with a small current until the discharge reaches the cut-off voltage, which is considered to be the end of the discharge; after the discharge is completed, stand still and calculate the capacity value data corresponding to the battery by the integral value of the current over time;

[0015] S103. Repeat steps S101-S102 until the maximum available capacity data of the battery reaches 70% of the nominal capacity of the battery, which is considered the end of the battery life and the data collection is completed.

[0016] As a preferred method that can be realized, step S200. According to the voltage segment of the energy storage power station, the voltage is divided into multiple voltage segments on the charging data, and the characteristic information of the lithium battery data information of each voltage segment is extracted; the specific steps include:

[0017] S201. Data trimming of the collected lithium battery data information, including removing the instantaneous voltage drop data at the start of charging and the instantaneous voltage recovery data at the end of charging;

[0018] S202. Set the voltage minimum point U according to the voltage segment where the energy storage power station operates. L and the highest voltage point U H , with a fixed ΔU as the voltage interval segment, divided into N voltage segments: [U0, U1], [U1, U2], …, [U k-1 , U k ],…,[U N-1 , U N ], where U0 = U L , U N =U H , and at the same time, according to the time segment where the voltage segment is located, the corresponding temperature and current data segments are segmented;

[0019] S203. According to the discrete voltage and current values ​​in each charging data, the sampling interval is adjusted to obtain the differential voltage dV value, and the differential capacity dQ value is obtained by using the current-time trapezoidal approximate integral;

[0020] S204. Using a Savitzky-Golay filter, by setting a window width and removing high-frequency noise, obtain a battery ICA curve, and determine a voltage segment where a peak is located according to the battery ICA curve;

[0021] S205. Extract characteristic information of lithium battery according to battery charge and discharge curve and ICA curve of each voltage segment. Characteristic information includes maximum value, minimum value, mean value, variance of temperature in the segment; maximum value, minimum value, mean value, variance and time integral value of current; mean value, variance and time integral value of voltage. Voltage value of IC peak, value of IC peak, kurtosis and skewness of IC peak, etc. A total of 15 characteristic values ​​can be selected.

[0022] As an achievable preferred method, step S300. selects the predicted maximum available capacity features of the battery in each voltage segment using the Deep Lasso algorithm according to the extracted feature information, specifically including:

[0023] S301. For n sets of data, m features of input X and battery capacity output Y, the Deep Lasso method is used to set the coefficients of some features to zero, and eliminate features that have little impact on the prediction target;

[0024] S302. Using Deep Lasso method to obtain feature extraction objective function;

[0025]

[0026] in, is the loss function of the model, θ is the parameter of the model, is a conventional loss term that measures the difference between the model’s prediction and the true label; is the penalty term of Deep Lasso, which is equivalent to Represents the partial derivative of the loss function with respect to the jth feature in the ith sample; α is used to control the weights of the loss term and the penalty term.

[0027] S303. Adjust the weight α so that the model has sparse gradients for features that have little impact on the output during training;

[0028] S304. For the j-th battery maximum available capacity feature, the loss function is calculated for the input feature gradient, which is expressed as:

[0029]

[0030] Set a suitable α and use the Deep Lasso method for training. After the training is completed, the importance of the jth feature is:

[0031]

[0032] S305. If the importance of the j-th feature is greater than the set threshold, the feature is extracted as the maximum available battery capacity feature; if the importance of the j-th feature is less than the set threshold, the feature is discarded.

[0033] As an achievable preferred method, step S400. adopts a Gaussian process regression (GPR) model and selects a corresponding kernel function to construct a battery maximum available capacity estimation model, and uses the acquired battery maximum available capacity characteristics and the corresponding battery maximum available capacity data to complete the training of the estimation model; specifically includes:

[0034] S401. The relationship between the maximum available battery capacity feature x selected in step S3 and the battery maximum available capacity data y is described as:

[0035] y=f(x)+ε;

[0036] Among them, f(x) is the target output, x is the input vector, ε is Gaussian noise, and satisfies ε~N(0,σ n 2 );

[0037] S402. The maximum available capacity decay process of the battery is described using the Gaussian process expression, which is specifically:

[0038] f(x)~GP(m(x),k(x,x'));

[0039] Among them, GP(m(x),k(x,x')) is a Gaussian process; m(x) is the mean function of the Gaussian process;

[0040] k(x,x') is the kernel function, and f(x)~GP(m(x),k(x,x')) means that f(x) obeys the Gaussian process;

[0041] S403. The kernel function is expressed in the form of multiplying a linear kernel function and an RBF kernel function. The formula of the kernel function is as follows:

[0042]

[0043] Among them, ||·|| 2 represents the L2 norm of the vector, C is the linear kernel function parameter, and l is the characteristic length scale of the kernel function;

[0044] S404. Divide the data set of the maximum available battery capacity feature of each segment and the data set of the maximum available battery capacity into a training set (x, y) and a test set (x * ,y * ), and train them separately;

[0045] According to the training set (x, y), the prior distribution of the maximum available capacity of the battery is obtained as:

[0046] y~N(0,K(x,x)+σ n 2 I n );

[0047] Where K(x, x)=(k(x i , x j )) n×n , σ n 2 I n is the noise covariance matrix, I n is an n-dimensional identity matrix;

[0048] The Gaussian process is a random process in which any finite subset of random variables obeys a joint normal distribution. When the input vector of the test is added, the joint prior distribution of the training set and the test set can be expressed as:

[0049]

[0050] As a preferred method that can be realized, step S500. Determine the voltage segment range covered by the energy storage battery according to the data information of the energy storage battery in actual operation, input the predicted maximum available capacity feature of the battery into the trained estimation model of the corresponding voltage segment, and obtain the estimated value of the maximum available capacity of the battery, specifically including:

[0051] S501. Collect the voltage, current and temperature values ​​of the battery fragment during the charging process; according to the terminal voltage of the battery when charging starts and the terminal voltage when charging stops, determine which one or more of the N fragments in step S2.1 the voltage range of the battery in operation is covered, and cut the voltage charging curve data according to the upper and lower limits of the fragment;

[0052] S502. According to the feature of the maximum available capacity of the battery selected in S3, feature extraction is performed, and the feature data is further input into the estimation model trained in S400. For the case where the charging segment covers 1 segment, the output of the GPR model is the estimated value of the maximum available capacity of the battery; for the case where the charging segment covers a segments, according to the idea of ​​Bayesian filtering, the output value of the GPR estimation model of the first segment is expressed as a priori, and the output value of the GPR estimation model of the second segment is used for updating, and the obtained posterior value is their joint distribution. The specific maximum available capacity estimation method is as follows:

[0053] For the first segment GPR estimation model, the output estimate y1 * Satisfies Gaussian distribution

[0054] For the second segment GPR estimation model, the output estimate y2 * Satisfies Gaussian distribution

[0055] With y2 * Update y1 * , the updated estimate y 12 * Satisfies Gaussian distribution in:

[0056]

[0057] For the a-th segment GPR estimation model, the output estimate y a * , satisfying the Gaussian distribution

[0058] With y a * Update y a-1 * ,y a-2 *…, a total of (a-1) iterations are made to get the estimated result y of the maximum available capacity of the battery 12...a * , the corresponding Gaussian distribution is as follows:

[0059]

[0060] in, is the mean of the maximum available capacity estimation after iteration, (σ 12...a 2 ) * is the variance of the maximum available capacity estimate after iteration;

[0061] S503. According to the obtained mean and variance of the maximum available capacity estimate, the 95% confidence segment of the maximum available capacity estimate of the battery is expressed as:

[0062]

[0063] As an achievable preferred method, the specific steps of cutting the voltage charging curve data according to the upper and lower limits of the segment in step S501 are:

[0064] For data covering 1 of the N segments, the current, voltage, temperature, and time data are cropped as follows:

[0065] [U k-1 ,U k ],[I k-1 , I k ],[T k-1 , T k ],[t k-1 , t k ];

[0066] For data covering a segments out of N segments, the current, voltage, temperature, and time data are trimmed as follows:

[0067] [U k-1 , U k ],[U k , U k+1 ], ..., [U k+a-1 , U k+a ];

[0068] [I k-1 , I k ],[I k , I k+1 ],...,[I k+a-1 , I k+a ];;

[0069] [T k-1 , T k ],[Tk , T k+1 ], ..., [T k+a-1 , T k+a ];;

[0070] [t k-1 , t k ],[t k , t k+1 ],...,[t k+a-1 , t k+a ];;

[0071] For data covering less than 1 segment, the maximum available capacity of the battery is not calculated because its impact on the maximum available capacity of the battery is small.

[0072] The beneficial effects of the present invention are:

[0073] The present invention provides a method for estimating the maximum available capacity of a battery based on the actual operating conditions of an energy storage power station. The method builds a model according to the actual operating conditions of the energy storage power station. In the case of spanning multiple intervals, multiple model outputs are iterated using Bayesian filtering to obtain an estimation result of the maximum available capacity of the battery. The method can not only estimate the maximum available capacity of the energy storage battery in real time, but also has good estimation accuracy, thus providing strong data support for the safe and reliable operation and maintenance of the energy storage power station. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 It is a flow chart of a method for estimating the maximum available capacity of a battery based on the actual operating conditions of an energy storage power station;

[0075] Figure 2 A diagram showing the estimation result of the maximum available capacity of a battery using a method for estimating the maximum available capacity of a battery based on the actual operating conditions of an energy storage power station according to the present application;

[0076] The above drawings show clear embodiments of the present disclosure, which will be described in more detail below. These drawings and text descriptions are not intended to limit the scope of the present disclosure in any way, but to illustrate the concepts of the present disclosure to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0077] The technical solution of the application is further described in detail below with reference to the accompanying drawings.

[0078] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0079] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0080] The following is combined with Figure 1-2 , some embodiments of the present invention are described in detail. In the absence of conflict, the following embodiments and features in the embodiments may be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0081] The present application provides a method for estimating the maximum available capacity of a battery based on the actual operating conditions of an energy storage power station, the method comprising the following steps:

[0082] S100. Perform a battery life test on the energy storage lithium battery to collect lithium battery data information, wherein the lithium battery data information includes the voltage, current, temperature, and maximum available capacity of each charge and discharge cycle of the lithium battery; the specific method includes:

[0083] S101 adopts a constant current and constant voltage charging method to charge the lithium battery, and collects the battery voltage, current and temperature values ​​during the process at a certain time interval;

[0084] S102. After standing still, discharge with a small current until the discharge reaches the cut-off voltage, which is considered to be the end of the discharge; after the discharge is completed, stand still and calculate the capacity value data corresponding to the battery by the integral value of the current over time;

[0085] S103. Repeat steps S101-S102 until the maximum available capacity data of the battery reaches 70% of the nominal capacity of the battery, which is considered the end of the battery life and the data collection is completed.

[0086] The specific explanation is as follows:

[0087] Take the 18650 lithium-ion battery cell with a nominal maximum available capacity of 2Ah as an example. a. At room temperature of 24°C, the CC-CV charging method is used to charge the energy storage battery at a constant current of 2A to a cut-off voltage of 4.2V, and then charge it at a constant voltage until the current is lower than a certain value. During the process, the voltage, current, and temperature values ​​of the battery are collected at certain time intervals.

[0088] b. After standing for a while, discharge with a small current until the cut-off voltage reaches 2.7V, which is considered to be the end of the discharge. After the discharge is completed, stand still. Calculate the corresponding capacity value of the battery by the integral value of the current over time.

[0089] c. Repeat steps ab until the maximum available capacity of the battery reaches 70% of the nominal capacity of the battery, which is considered the end of the battery life and the experiment is stopped.

[0090] B. Fill in the missing values ​​in the voltage, current, temperature and capacity data recording process and remove abnormal values.

[0091] S200. According to the voltage segment of the energy storage power station, the voltage is divided into multiple voltage segments on the charging data, and the characteristic information of the lithium battery data information of each voltage segment is extracted; it should be noted that the specific steps include:

[0092] S201. Data trimming is performed on the collected lithium battery data information, including removing the instantaneous voltage drop data at the beginning of charging and the instantaneous voltage recovery data when the charging is completed and the temperature is left at rest; that is, for each complete charging voltage, current, and temperature data, data trimming is first performed to remove the instantaneous voltage drop data at the beginning of charging and the instantaneous voltage recovery data when the charging is completed and the temperature is left at rest.

[0093] S202. Set the voltage minimum point U according to the voltage segment where the energy storage power station operates. L and the highest voltage point U H , with a fixed ΔU as the voltage interval segment, divided into N voltage segments: [U0, U1], [U1, U2], …, [U k-1 , U k ],…,[U N-1 , U N ], where U0 = U L , U N =U H At the same time, according to the time segment where the voltage segment is located, the corresponding temperature and current data segments are segmented; this application gives the lowest voltage point U according to the interval where the energy storage power station is operating. L =2.7V and the highest voltage point U H=4.2V, with a fixed ΔU = 0.3V as the voltage interval, n = 5 voltage segments are intercepted: [U0, U1], [U1, U2], …, [U k-1 , U k ],…,[U k-1 , U n ], thereby completing the division of temperature and current intervals according to the time interval of the voltage segment.

[0094] S203. According to the discrete voltage and current values ​​in each charging data, the sampling interval is adjusted to obtain the differential voltage dV value, and the differential capacity dQ value is obtained by using the current-time trapezoidal approximate integral;

[0095] S204. Using a Savitzky-Golay filter, by setting a window width and removing high-frequency noise, obtain a battery ICA curve, and determine a voltage segment where a peak is located according to the battery ICA curve;

[0096] S205. Extract characteristic information of lithium battery according to battery charge and discharge curve and ICA curve of each voltage segment. Characteristic information includes maximum value, minimum value, mean value, variance of temperature in the segment; maximum value, minimum value, mean value, variance and time integral value of current; mean value, variance and time integral value of voltage. Voltage value of IC peak, value of IC peak, kurtosis and skewness of IC peak, etc. A total of 15 characteristic values ​​can be selected.

[0097] S300. Selecting the predicted maximum available capacity feature of the battery in each voltage segment using the Deep Lasso algorithm based on the extracted feature information;

[0098] It should be noted that the specific steps are as follows: S301. For n sets of data, m feature inputs X and battery capacity outputs Y, the corresponding matrix data is as follows:

[0099] Y=[y1 y2…y i …y n ] T ;

[0100] At the same time, the Deep Lasso method is used to set the coefficients of some features to zero, eliminating features that have little impact on the prediction target, reducing the complexity of the model, and improving the interpretability and generalization ability of the model;

[0101] S302. Use the Deep Lasso method to set the objective function of feature extraction. The objective function is as follows:

[0102]

[0103] in, is the loss function of the model, θ is the parameter of the model, is a conventional loss term that measures the difference between the model’s prediction and the true label; is the penalty term of Deep Lasso, which is the weighted sum of the L2 norm of the gradient of the loss function with respect to the input features. That is, the gradient of each input feature is first calculated using the loss function, and then the L2 norm of the gradient is calculated. α is used to control the weight of the two. Its form is equivalent to

[0104] S303. Adjust the weight α so that the model has sparse gradients for features that have little impact on the output during training;

[0105] S304. For the j-th battery maximum available capacity feature, the loss function is calculated for the input feature gradient, which is expressed as:

[0106]

[0107] The Deep Lasso method is used for training. After the training is completed, the importance of the jth feature is:

[0108]

[0109] S305. If the importance of the jth feature is greater than the set threshold, the feature is extracted as the maximum available battery capacity feature. If the importance of the jth feature is less than the set threshold, the feature is discarded. In this embodiment, all feature information has 6 features whose importance is greater than the set threshold through the above method, so they are used as the maximum available battery capacity feature.

[0110] S400. Using a Gaussian process regression (GPR) model and selecting a corresponding kernel function to construct a battery maximum available capacity estimation model, and using the acquired battery maximum available capacity feature data and the corresponding battery maximum available capacity data to complete the training of the estimation model;

[0111] It should be noted that the steps are as follows: S401. Describe the relationship between the maximum available battery capacity feature x selected in step S3 and the battery maximum available capacity data y as follows:

[0112] y=f(x)+ε;

[0113] Among them, f(x) is the target output, x is the input vector, ε is Gaussian noise, and satisfies ε~N(0,σ n 2 );

[0114] S402. The maximum available capacity decay process of the battery is described using the expression of the Gaussian process, that is, the Gaussian process of the maximum available capacity decay of the battery can be written as:

[0115] f(x)~GP(m(x),k(x,x'));

[0116] Wherein, GP(m(x), k(x, x')) is a Gaussian process; m(x) is the mean function of the Gaussian process, m(x) = E(f(x)); k(x, x') is the covariance function, which is usually set with a kernel function, k(x, x') = E[(f(x) - m(x))(f(x') - m(x'))], and f(x) ~ GP(m(x), k(x, x')) indicates that f(x) obeys a Gaussian process;

[0117] S403. The kernel function is expressed in the form of multiplying a linear kernel function and an RBF kernel function. The formula of the kernel function is as follows:

[0118]

[0119] Among them, ||·||2 represents the L2 norm of the vector, C is the linear kernel function parameter, and l is the characteristic length scale of the kernel function;

[0120] S404. Divide the data set of the maximum available battery capacity feature of each segment and the data set of the maximum available battery capacity into a training set (x, y) and a test set (x * ,y * ), and train them separately;

[0121] According to the training set (x, y), the prior distribution of the maximum available capacity of the battery is obtained as:

[0122] y~N(0,K(x,x)+σ n 2 I n );

[0123] Where K(x, x)=(k(x i , x j )) n×n , σ n 2 I n is the noise covariance matrix, I n is an n-dimensional identity matrix; k(x i , x j ) is used to describe x i and x j In this application, the maximum available battery capacity feature data set of each interval and the battery maximum available capacity data set are divided into a training set (x, y) and a test set (x) at a ratio of 80% training set and 20% test set. * ,y * ) and train them separately.

[0124] Because the Gaussian process is a random process, any finite subset of random variables in it obeys the joint normal distribution. When the input vector of the test is added, the joint prior distribution of the training set and the test set can be expressed as:

[0125]

[0126] Observation value y * The posterior distribution of is:

[0127]

[0128] Among them, the predicted mean for:

[0129]

[0130] Prediction variance σ 2 (y * )for:

[0131] σ 2 (y * )=K(x * , x * )-K(x,x * ) T [K(x,x)+σ n 2 I n ] -1 K(x,x * );

[0132] S500. Determine the voltage segment range covered by the energy storage battery according to the data information of the energy storage battery in actual operation, input the predicted maximum available capacity characteristics of the battery into the trained estimation model, and obtain the estimated value of the maximum available capacity of the battery; the specific instructions are as follows:

[0133] S501. Collect the voltage, current and temperature values ​​of the battery fragment during the charging process, and determine which one or more fragments of the N fragments in step S2.1 the voltage range of the running battery is covered by according to the terminal voltage when the battery starts charging and the terminal voltage when the battery stops charging, and cut the voltage charging curve data according to the upper and lower limits of the fragment;

[0134] For data covering 1 of the N segments, the current, voltage, temperature, and time data are cropped as follows:

[0135] [U k-1 , U k ],[I k-1 , I k ],[T k-1 , T k ],[tk-1 , t k ];

[0136] For data covering a segments out of N segments, the current, voltage, temperature, and time data are trimmed as follows:

[0137] [U k-1 , U k ],[U k , U k+1 ], ..., [U k+a-1 , U k+a ];

[0138] [I k-1 , I k ],[I k , I k+1 ],...,[I k+a-1 , I k+a ];

[0139] [T k-1 , T k ],[T k , T k+1 ], ..., [T k+a-1 , T k+a ];

[0140] [t k-1 , t k ],[t k , t k+1 ],...,[t k+a-1 , t k+a ];

[0141] For data covering less than 1 segment, the maximum available capacity of the battery is not calculated because its impact on the maximum available capacity of the battery is small.

[0142] This embodiment collects data of charging terminal voltage of 3.3V to 3.9V, covering two intervals of [3.3V, 3.6V] and [3.6V, 3.9V].

[0143] S502. According to the maximum available capacity feature of the battery selected by S3, feature extraction is performed, and the feature data is further input into the estimation model trained by S4. For the case where the charging segment covers 1 segment, the output of the GPR model is the estimated value of the maximum available capacity of the battery; for the case where the charging segment covers a segments, according to the idea of ​​Bayesian filtering, the output value of the GPR estimation model of the first segment is expressed as a priori, and the output value of the GPR estimation model of the second segment is used for updating, and the obtained posterior value is their joint distribution. The specific maximum available capacity estimation method is as follows:

[0144] For the first segment GPR estimation model, the output estimate y1 * Satisfies Gaussian distribution

[0145] For the second segment GPR estimation model, the output estimate y2 * Satisfies Gaussian distribution

[0146] With y2 * Update y1 * , the updated estimate y 12 * Satisfies Gaussian distribution in:

[0147]

[0148] For the a-th segment GPR estimation model, the output estimate y a * , satisfying the Gaussian distribution

[0149] With y a * Update y a-1 * ,y a-2 * …, a total of (a-1) iterations are made to get the estimated result y of the maximum available capacity of the battery 12...a * , the corresponding Gaussian distribution is as follows:

[0150]

[0151] in, is the mean of the maximum available capacity estimation after iteration, (σ 12...a 2 ) * is the variance of the maximum available capacity estimate after iteration;

[0152] S503. According to the obtained mean and variance of the maximum available capacity estimate, the 95% confidence segment of the maximum available capacity estimate of the battery is expressed as:

[0153]

[0154] Table 1

[0155] MAPE <![CDATA[R 2 ]]> 0.65％ 0.987

[0156] This application randomly selects several running energy storage battery cells for estimation, and the estimation results are shown in the attached Figure 2 As shown, Figure 2The horizontal axis represents the true value of the maximum available capacity, and the vertical axis represents the estimated value of the maximum available capacity. The dotted line in the figure indicates that the estimated value = the true value. It can be seen that the points in the figure are concentrated near the dotted line, indicating that the estimated value is very close to the true value, which means that the estimation has a good effect. At the same time, the estimation accuracy indicators of the model of this method are shown in Table 1 above. The mean absolute percentage error (MAPE) of this method is 0.65%, and the goodness of fit R 2 The MAPE of the existing model-based real-time estimation method of the maximum available capacity of the battery is usually around 1.5%, which shows that the method used in this application is more accurate than the model-based real-time estimation method. The existing data-driven method for estimating the maximum available capacity of the battery uses full cycle data to establish a model, and its MAPE is usually around 0.7%, and its R 2 It is usually around 0.97, but it cannot be applied to the actual operating conditions of energy storage power stations. The method used in this application can achieve this accuracy by using only a model established with incomplete charging data, which means that the method used in this patent can also maintain a high maximum available capacity estimation accuracy under the conditions applicable to the actual operating conditions of energy storage power stations. Therefore, the method described in this article can estimate the maximum available capacity of the battery under the operating conditions of energy storage power stations when there is no complete charge and discharge cycle data, and the estimation accuracy of the maximum available capacity is relatively high.

[0157] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

[0158] The above are only some embodiments of the present invention. For those skilled in the art, several modifications and improvements can be made without departing from the creative concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A method for estimating the maximum available capacity of a battery based on the actual operating conditions of an energy storage power station, characterized in that: The estimation method comprises the following steps: S100. Performing a battery life test on the energy storage lithium battery to collect lithium battery data information, wherein the lithium battery data information includes the voltage, current, temperature, and maximum available capacity of each charge and discharge cycle of the lithium battery; S200. According to the voltage segment of the energy storage power station operation, the voltage is divided into multiple voltage segments on the charging data, and the characteristic information of the lithium battery data information of each voltage segment is extracted; S300. Selecting the predicted maximum available capacity feature of the battery in each voltage segment using the Deep Lasso algorithm based on the extracted feature information; S400. A Gaussian process regression (GPR) model is used and a corresponding kernel function is selected to construct a battery maximum available capacity estimation model, and the obtained battery maximum available capacity characteristic data and the corresponding battery maximum available capacity data are used to complete the training of the estimation model; S500. Determine the voltage segment range covered by the energy storage battery according to the data information of the energy storage battery in actual operation, input the predicted maximum available capacity characteristics of the battery into the trained estimation model, and obtain the estimated value of the maximum available capacity of the battery.

2. A method for estimating the maximum available capacity of a battery based on the actual operating conditions of an energy storage power station according to claim 1, characterized in that: Step S100: Perform a battery life test on the energy storage lithium battery to collect lithium battery data information, the data information includes the voltage, current, temperature, and maximum available capacity of each charge and discharge cycle of the lithium battery. The specific steps include: S101 adopts a constant current and constant voltage charging method to charge the lithium battery, and collects the battery voltage, current and temperature values ​​during the process at a certain time interval; S102. After standing still, discharge with a small current until the discharge reaches the cut-off voltage, which is considered to be the end of the discharge; after the discharge is completed, stand still and calculate the capacity value data corresponding to the battery by the integral value of the current over time; S103. Repeat steps S101-S102 until the maximum available capacity data of the battery reaches 70% of the nominal capacity of the battery, which is considered the end of the battery life and the data collection is completed.

3. A method for estimating the maximum available capacity of a battery based on the actual operating conditions of an energy storage power station according to claim 2, characterized in that: Step S200. According to the voltage segment of the energy storage power station, the voltage is divided into multiple voltage segments on the charging data, and characteristic information is extracted from the lithium battery data information of each voltage segment; The specific steps include: S201. Data trimming of the collected lithium battery data information, including removing the instantaneous voltage drop data when charging starts and the instantaneous voltage recovery data when charging ends and the static state; S202. Set the voltage minimum point U according to the voltage segment where the energy storage power station operates. L and the highest voltage point U H , with a fixed ΔU as the voltage interval segment, divided into N voltage segments: [U0, U1], [U1, U2], …, [U k-1 , U k ],…,[U N-1 , U N-2 ], where U0 = U L , U N =U H , and at the same time, according to the time segment where the voltage segment is located, the corresponding temperature and current data segments are segmented; S203. According to the discrete voltage and current values ​​in each charging data, the sampling interval is adjusted to obtain the differential voltage dV value, and the differential capacity dQ value is obtained by using the current-time trapezoidal approximate integral; S204. Using a Savitzky-Golay filter, by setting a window width and removing high-frequency noise, obtain a battery ICA curve, and determine a voltage segment where a peak is located according to the battery ICA curve; S205. Extract characteristic information of the lithium battery according to the battery charge and discharge curve and the ICA curve of each voltage segment.

4. A method for estimating the maximum available capacity of a battery based on the actual operating conditions of an energy storage power station according to claim 3, characterized in that: Step S300: Select the predicted maximum available capacity features of the battery in each voltage segment using the Deep Lasso algorithm based on the extracted feature information, specifically including: S301. For n sets of data, m features of input X and battery capacity output Y, the Deep Lasso method is used to set the coefficients of some features to zero, and eliminate features that have little impact on the prediction target; S302. Use the Deep Lasso method to set the objective function of feature extraction. The objective function is as follows: in, is the loss function of the model, θ are the parameters of the model, is a conventional loss term that measures the difference between the model’s prediction and the true label; is the penalty term of Deep Lasso, α Used to control the weights of loss terms and penalty terms; S303. Adjust the weight α so that the model has sparse gradients for features that have little impact on the output during training; S304. For the j-th battery maximum available capacity feature, the loss function is calculated for the input feature gradient, which is expressed as: Set a suitable α and use the Deep Lasso method for training. After the training is completed, the importance of the jth feature is: S305. If the importance of the j-th feature is greater than the set threshold, the feature is extracted as the maximum available battery capacity feature; if the importance of the j-th feature is less than the set threshold, the feature is discarded.

5. A method for estimating the maximum available capacity of a battery based on the actual operating conditions of an energy storage power station according to claim 4, characterized in that: Step S400. Use a Gaussian process regression (GPR) model and select a corresponding kernel function to build a battery maximum available capacity estimation model, and use the acquired battery maximum available capacity characteristics and corresponding battery maximum available capacity data to complete the training of the estimation model; specifically including: S401. The maximum available battery capacity characteristic selected in step S3 x and the maximum available battery capacity data y The relationship is described as: y=f(x)+ε; Among them, f(x) is the target output, x is the input vector, ε is Gaussian noise, and satisfies ε~N(0,σ n 2 ); S402. The maximum available capacity decay process of the battery is described using the Gaussian process expression, which is specifically: f(x)~GP(m(x), k(x, x')) Among them, GP(m(x),k(x,x')) is a Gaussian process; m(x) is the mean function of the Gaussian process; k(x,x') is the kernel function, and f(x)~GP(m(x),k(x,x')) means that f(x) obeys the Gaussian process; S403. The kernel function is expressed in the form of multiplying a linear kernel function and an RBF kernel function. The formula of the kernel function is as follows: Among them, ||·||2 represents the L2 norm of the vector, C is the linear kernel function parameter, and l is the characteristic length scale of the kernel function; S404. Divide the data set of the maximum available battery capacity feature of each segment and the data set of the maximum available battery capacity into a training set (x, y) and a test set (x * ,y * ), and train them separately; According to the training set (x, y), the prior distribution of the maximum available capacity of the battery is: y~N(0, K(x, x)+σ n 2 I n ); Where K(x, x)=(K(x i , x j )) n×n , σ n 2 I n is the noise covariance matrix, I n is an n-dimensional identity matrix; The Gaussian process is a random process in which any finite subset of random variables obeys a joint normal distribution. When the test input vector is added, the joint prior distribution of the training set and the test set can be expressed as:

6. A method for estimating the maximum available capacity of a battery based on the actual operating conditions of an energy storage power station according to claim 5, characterized in that: Step S500: Determine the voltage segment range covered by the energy storage battery according to the data information of the energy storage battery in actual operation, input the predicted maximum available capacity characteristics of the battery into the trained estimation model of the corresponding voltage segment, and obtain the estimated value of the maximum available capacity of the battery, specifically including: S501. Collect the voltage, current and temperature values ​​of the battery fragment during the charging process, and determine which one or more fragments of the N fragments in step S2.1 the voltage range of the running battery is covered by according to the terminal voltage when the battery starts charging and the terminal voltage when the battery stops charging, and cut the voltage charging curve data according to the upper and lower limits of the fragment; S502. According to the feature of the maximum available capacity of the battery selected in S3, feature extraction is performed, and the feature data is further input into the estimation model trained in S400. For the case where the charging segment covers 1 segment, the output of the GPR model is the estimated value of the maximum available capacity of the battery; for the case where the charging segment covers a segments, according to the idea of ​​Bayesian filtering, the output value of the GPR estimation model of the first segment is expressed as a priori, and the output value of the GPR estimation model of the second segment is used for updating, and the obtained posterior value is their joint distribution. The specific maximum available capacity estimation method is as follows: For the first segment GPR estimation model, the output estimate y1 satisfies the Gaussian distribution For the second segment GPR estimation model, the output estimate y2 * Satisfies Gaussian distribution Update y1 with y2 * , the updated estimate y 12 * Satisfies Gaussian distribution in: For the a-th segment GPR estimation model, the output estimate y a , satisfying the Gaussian distribution With y a * Update y a-1 * ,y a-2 * …, a total of (a-1) iterations are made to get the estimated result y of the maximum available capacity of the battery 12...a * , the corresponding Gaussian distribution is as follows: in, is the mean of the maximum available capacity estimation after iteration, (σ 12...a 2 ) * is the variance of the maximum available capacity estimate after iteration; S503. According to the obtained mean and variance of the maximum available capacity estimate, the 95% confidence segment of the maximum available capacity estimate of the battery is expressed as:

7. A method for estimating the maximum available capacity of a battery based on the actual operating conditions of an energy storage power station according to claim 6, characterized in that: The specific steps of cutting the voltage charging curve data according to the upper and lower limits of the segment in step S501 are: For data covering 1 of the N segments, the current, voltage, temperature, and time data are cropped as follows: [U k-1 ,U k ],[I k-1 ,I k ],[T k-1 ,T k ],[t k-1 ,t k ]; For data covering a segments out of N segments, the current, voltage, temperature, and time data are trimmed as follows: [IN k-1 ,IN k ],[IN k ,IN k+1 ],…,[IN k+a-1 ,IN k+a ]; [I k-1 ,I k ],[I k ,I k+1 ],…,[I k+a-1 ,I k+a ]; [T k-1 ,T k ],[T k ,T k+1 ],…,[T k+a-1 ,T k+a ]; [t k-1 ,t k ],[t k ,t k+1 ],...,[t k+a-1 ,t k+a ]; For data covering less than 1 segment, the maximum available capacity of the battery is not calculated because its impact on the maximum available capacity of the battery is small.

Images