Battery life prediction method based on density outlier factor elimination

Abstract

The invention discloses a battery life prediction method based on density outlier factor elimination. The method comprises the following steps: firstly, carrying out charge-discharge circulation on a lithium battery, analyzing a voltage-current change curve of the lithium battery in a charge-discharge stage, and selecting a plurality of HIs capable of representing battery degradation from the voltage-current change curve; then, the correlation degree between each HI sequence and the battery capacity is quantitatively analyzed, and meanwhile, standardization processing is carried out on the extracted HI sequences; then, on the basis of a local outlier factor algorithm, outliers are removed in sequence, redundancy and information bias are reduced, and the data density is improved; and finally, the cleaned HI sequence is used as an input and output sequence of the model, so that the effectiveness and accuracy of the battery life prediction model are improved. According to the method, the case library is constructed by using the actual operation battery attenuation data, and powerful guidance is provided for an operator to find the battery life prediction accuracy.

Description

Technical Field

[0001] The present invention belongs to the field of battery fault diagnosis and battery health management, and specifically relates to a battery life prediction method based on density outlier factor elimination. Background Art

[0002] In recent years, with the development of technologies such as artificial intelligence and big data, the scale of data has experienced explosive growth. Massive data sets contain more information, and the increase in the amount of lithium battery sample data can effectively improve the accuracy of the constructed prediction or classification model. However, in the actual engineering practice of lithium batteries, outliers will be generated, which makes the collected data sets usually have some inconsistent data samples. For example, in the battery data set, the monitoring data will be affected by factors such as external environmental interference, equipment failure restart, battery life attenuation, sensor measurement errors, etc., which will affect the performance of the final prediction model. Therefore, when predicting the life of the battery, removing outlier factors from the data set can effectively improve the prediction ability of the model.

[0003] Based on the comparative analysis of the battery health indicator (HI) sequence data set and common data cleaning methods, considering that the battery HI sequence does not satisfy the normal distribution, and due to the accelerated decay rate of the battery in the later stage, the number of monitoring data points in the early stage is significantly more than that in the later stage, making the distribution density of the sampled HI sequence uneven. If the traditional method (such as the quartile method, etc.) is used to directly predict the battery life, the overall decay level cannot be accurately represented, resulting in the model often having poor prediction ability, poor generalization ability, and poor robustness when actually predicting. Summary of the invention

[0004] To solve the above problems, the present invention uses the local outlier factor (LOF) to calculate the abnormal score of each data point in the sequence according to the distribution form and characteristics of the outliers in the battery HI sequence, and identifies and removes the outliers by setting the score threshold to enhance the accuracy and performance of the battery prediction model. Therefore, the invention proposes a battery life prediction method based on density outlier factor removal.

[0005] The purpose of the present invention is to adopt the following technical solution to achieve the battery life prediction method based on density outlier factor elimination proposed by the present invention, comprising the following steps:

[0006] Step 1: Charge and discharge the lithium battery, and analyze the voltage and current change curves during the charge and discharge stages, and select multiple HIs that can characterize battery degradation;

[0007] Step 2: Quantitatively analyze the correlation between each HI sequence and the battery capacity, and standardize the extracted HI sequence;

[0008] Step 3: Based on the local outlier factor algorithm, outliers are removed one by one to reduce redundancy and information bias and improve data density;

[0009] Step 4: Use the cleaned HI sequence as the input and output sequence of the model to improve the effectiveness and accuracy of the battery life prediction model.

[0010] Furthermore, step one specifically includes:

[0011] Select a lithium battery for testing, and perform charge and discharge cycles on the lithium battery. Select four groups of HI sequences from the battery charge and discharge curve to reflect the degradation state of the battery, namely: constant current charging stage voltage drop end point value HI1, constant current charging time HI2, constant current discharge time HI3 and battery average internal resistance HI4, where:

[0012]

[0013]

[0014]

[0015]

[0016] In the formula, It is the value of the lowest point of voltage drop after the constant current charging stage ends; It is the staged constant current charging time; is the constant current discharge time; is the measured average internal resistance of the battery.

[0017] Furthermore, step 2 specifically includes the following process:

[0018] Step 2-1: Combine the extracted HI sequence and its corresponding capacity value to construct a battery degradation feature sequence data set {cycle, HI1, HI2, HI3, HI4, capacity}. Combined with the battery degradation feature sequence data set, the Spearman correlation analysis method is used to quantify the correlation between each HI sequence and the battery capacity. The correlation range is [-1, 1], and the calculation method is as follows:

[0019]

[0020] Where n is the number of elements in variables X and Y; d i For data point x i and i The level difference, if ρs The closer the value is to 1 or -1, the stronger the correlation between the two variables. s The closer the value is to 0, the weaker or no correlation between the two variables;

[0021] Step 2-2: Use the Min-max standardization method to standardize the battery degradation feature sequence data set: Min-max standardization uses the maximum and minimum values ​​in the sequence to process the data in the sequence so that the data is mapped to the [0, 1] interval. The following is its calculation formula:

[0022]

[0023] Where x is the original HI input data; X max is the maximum value of HI data; X min is the minimum value of HI data; Y is the battery capacity after standardization;

[0024] Step 2-3: Use the lithium battery degradation feature sequence data set {cycle, HI1, HI2, HI3, HI4, capacity} that has undergone association analysis and standardization as the input and output of the battery prediction model.

[0025] Furthermore, step three specifically includes the following process:

[0026] Step 3-1: Calculate the distance between data points: Assume that there are data points p and o in the battery degradation feature sequence data set D, where data point o is the set D / {p} other than p in the data set D. Use the Euclidean distance to calculate the distance between data point p(x p ,y p ) and the data point o(x o ,y o ) is the distance d(p,o):

[0027]

[0028] Step 3-2: Calculate the kth distance of data point p: Sort the distance calculation results from small to large, and find the set of k adjacent data points O = {o1, o2, ..., o k}, the distance from data point p to the farthest k-th adjacent data point is d(p, o k ), which is the kth distance of data point p, which can also be recorded as d k (p);

[0029] Step 3-3: Determine the kth distance neighborhood of data point p: Set the kth distance d of data point p k The collection of all data points q in (p) is called the kth distance neighborhood of p, denoted by Nk (p), which is expressed as:

[0030] N k (p) = {q|d(p,q)≤d k (p)}

[0031] Step 3-4: Determine the reachable distance of data point p relative to data point o: The reachable distance of data point p relative to data point o is the larger value of "the kth distance of data point o" and "the Euclidean distance between data point p and data point o", and its specific expression is:

[0032] reach·dist(p,o)=max{d k (o), d(p, o)}

[0033] Step 3-5: Calculate the local outlier factor of data point p: The local outlier factor of data point p is expressed as the neighborhood N of data point p k The inverse of the average reachable distance of other data points in (p) is expressed as:

[0034]

[0035] It can be seen from the above formula that the larger the average reachable distance, the smaller the local reachable density of the data point p;

[0036] Step 3-6: Calculate the local outlier factor of data point p: The local outlier factor of data point p is expressed as the neighborhood N of data point p k The ratio of the average local reachable density of other data points in (p) to the local reachable density of data point p is expressed as:

[0037]

[0038] According to the above formula, the closer the LOF value is to 1, the more similar the average density of the data point p is to the surrounding data points o, and the more likely it is to be a normal value; the larger the LOF value is than 1, the farther the data point p is from the surrounding data points, and the more likely it is to be an outlier.

[0039] Furthermore, step 4 specifically includes the following process:

[0040] Step 4-1: Take the HI sequence after standardization and outlier elimination in step 3, perform prediction analysis on the battery life, and use the mean square error to evaluate the later model. The specific calculation formula is as follows:

[0041]

[0042] in, and iis the predicted life and actual life of the i-th battery tuple;

[0043] Step 4-2: According to the method of step 4-1, the original data and the data after the density outlier factor is eliminated are used as the input of the model to predict the battery life;

[0044] Step 4-3: Analyze the results of step 4-2, and compare the two methods to evaluate the accuracy of battery life prediction under the model evaluation index MSE.

[0045] The beneficial effects of the present invention based on its technical solution are:

[0046] (1) The present invention provides a battery life prediction method based on density outlier elimination, which uses a Spearman correlation analysis method to extract relevant feature sequences.

[0047] (2) The battery life prediction method based on density outlier factor elimination of the present invention uses actual operation data to build a case library, and the case library has a high credibility.

[0048] (3) The present invention provides a battery life prediction method based on density outlier factor elimination, which eliminates outlier factors and improves the accuracy of the model in predicting battery life.

[0049] (4) The present invention uses actual battery attenuation data to build a case library, providing powerful guidance for operators to find the accuracy of battery life prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flowchart of a method for predicting battery life based on density outlier factor elimination according to an embodiment of the present invention;

[0051] Figure 2 It is a graph of the number of battery cycles versus actual battery capacity;

[0052] Figure 3 It is a statistical chart of battery life prediction and comparison experiments. DETAILED DESCRIPTION

[0053] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0054] The present invention provides a battery life prediction method based on density outlier factor elimination, such as Figure 1 As shown, the specific steps include:

[0055] Step 1: Perform charge and discharge cycles on the lithium battery, and analyze the voltage and current change curves during the charge and discharge stages, and select multiple HIs that can characterize battery degradation.

[0056] Step 1 specifically includes:

[0057] The test battery uses a 18650 commercial lithium iron phosphate battery (APR18650M1A) manufactured by A123 Systems. The nominal capacity of the test battery is 1.1Ah, the rated voltage is 3.3V, and the capacitance failure threshold is 0.88Ah (80%). In order to ensure that the selected health factor contains sufficient battery degradation information, the present invention selects four groups of HI sequences from the battery charge and discharge curve to reflect the degradation state of the battery, namely: the voltage drop end point value of the constant current charging stage (HI1), the constant current charging time (HI2), the constant current discharge time (HI3) and the average internal resistance of the battery (HI4).

[0058] in:

[0059]

[0060]

[0061]

[0062]

[0063] In the formula, It is the value of the lowest point of voltage drop after the constant current charging stage ends; It is the staged constant current charging time; is the constant current discharge time; is the measured average internal resistance of the battery.

[0064] Step 2: Quantitatively analyze the correlation between each HI sequence and the battery capacity, and standardize the extracted HI sequence.

[0065] Step 2 specifically includes the following processes:

[0066] Step 2-1: Combine the extracted HI sequence and its corresponding capacity value to construct a battery degradation feature sequence data set {cycle, HI1, HI2, HI3, HI4, capacity}. Combined with the battery degradation feature sequence data set, the Spearman correlation analysis method is used to quantify the correlation between each HI sequence and the battery capacity. The correlation range is [-1, 1], and the calculation method is as follows:

[0067]

[0068] Where n is the number of elements in variables X and Y; d i For data point x i and i If ρ sThe closer the value is to 1 or -1, the stronger the correlation between the two variables. s The closer the value is to 0, the weaker the correlation between the two variables, or even no correlation.

[0069] Step 2-2: Use the Min-max standardization method to standardize the battery degradation feature sequence data set. Min-max standardization uses the maximum and minimum values ​​in the sequence to process the data in the sequence so that the data is mapped to the [0, 1] interval. The purpose is to make the influence of each dimensional feature in the input data set on the weight in the output function consistent. The following is the calculation formula:

[0070]

[0071] Where X is the original HI input data; X max is the maximum value of HI data; X min is the minimum value of HI data; Y is the battery capacity after standardization.

[0072] Step 2-3: Use the lithium battery degradation feature sequence data set {cycle, HI1, HI2, HI3, HI4, capacity} that has undergone association analysis and standardization as the input and output of the battery prediction model.

[0073] Step 3: Based on the local outlier factor algorithm, outliers are eliminated one by one to reduce redundancy and information bias and improve data density.

[0074] Step 3 specifically includes the following processes:

[0075] Step 3-1: Calculate the distance between data points. Assume that there are data points p and o in the battery degradation feature sequence data set D, where data point o is the set D / {p} other than p in the data set D. Use Euclidean distance to calculate the distance between data points p(x p ,y p ) and the data point o(x o ,y o ) is the distance d(p,o):

[0076]

[0077] Step 3-2: Calculate the kth distance of data point p. Sort the distance calculation results from small to large, and find the set of k adjacent data points O = {o1, o2, ..., o k}. The distance from data point p to the farthest k-th adjacent data point is d(p, o k ), which is the kth distance of data point p, which can also be recorded as dk (p).

[0078] Step 3-3: Determine the kth distance neighborhood of data point p. Set the kth distance d of data point p to k The collection of all data points q in (p) is called the kth distance neighborhood of p, denoted by N k (p), which is expressed as:

[0079] N k (p) = {q|d(p,q)≤d k (p)}

[0080] Step 3-4: Determine the reachable distance of data point p relative to data point o. The reachable distance of data point p relative to data point o is the larger value of "the kth distance of data point o" and "the Euclidean distance between data point p and data point o", and its specific expression is:

[0081] reach·dist(p,o)=max{d k (o), d(p, o)}

[0082] Step 3-5: Calculate the local outlier factor of data point p. The local outlier factor of data point p is expressed as the neighborhood N of data point p k The inverse of the average reachable distance of other data points in (p) is expressed as:

[0083]

[0084] It can be seen from the above formula that the larger the average reachable distance, the smaller the local reachable density of the data point p.

[0085] Step 3-6: Calculate the local outlier factor of data point p. The local outlier factor of data point p is expressed as the neighborhood N of data point p k The ratio of the average local reachable density of other data points in (p) to the local reachable density of data point p is expressed as:

[0086]

[0087] According to the above formula, the closer the LOF value is to 1, the more similar the average density of the data point p is to the surrounding data points o, and the more likely it is a normal value; the larger the LOF value is than 1, the farther the data point p is from the surrounding data points, and the more likely it is an outlier.

[0088] Step 4: Use the cleaned HI sequence as the input and output sequence of the model to improve the effectiveness and accuracy of the battery life prediction model.

[0089] Step 4 specifically includes the following processes:

[0090] Step 4-1: Take the HI sequence after standardization and outlier elimination in step 3, perform prediction analysis on the battery life, and use the mean square error (MSE) to evaluate the later model. The specific calculation formula is as follows:

[0091]

[0092] in, and i are the predicted life and actual life of the i-th battery tuple.

[0093] Step 4-2: According to the method in step 4-1, the original data and the data after the density outlier factor is eliminated are used as the input of the model to predict the battery life.

[0094] Step 4-3: Analyze the results of step 4-2, and compare the accuracy of the two methods in predicting battery life under the model evaluation index MSE.

[0095] Example:

[0096] First, the lithium battery is charged and discharged in cycles, and the voltage and current change curves during the charge and discharge stages are analyzed, and multiple HIs that can characterize battery degradation are selected. Then, the correlation between each HI sequence and the battery capacity is quantitatively analyzed, and the extracted HI sequence is standardized. Subsequently, based on the local outlier factor algorithm, the outliers are removed one by one to reduce redundancy and information bias and improve data density. Finally, the cleaned HI sequence is used as the input and output sequence of the model to improve the effectiveness and accuracy of the battery life prediction model. The specific steps are as follows:

[0097] (1) Perform charge and discharge cycles on lithium batteries and select multiple HIs that can characterize battery degradation

[0098] Collect data on HIs such as battery capacity decay and battery life, and analyze the voltage and current change curves during the charge and discharge phases, and select multiple HIs that can characterize battery degradation. Select influencing factors that are highly correlated with battery life as the data set to be collected later, such as Figure 2 As shown, Figure 2 In the figure, the horizontal axis is the number of battery cycles, and the vertical axis is the actual capacity of the battery. Figure 2 It can be seen that the rate of change is high in the middle of the battery life, which is helpful for the extraction of health factors.

[0099] (2) Data preprocessing

[0100] According to step 2, the collected original sample data is preprocessed. Figure 2The curve in the figure is analyzed by Spearman method to find the eigenvalue with the best correlation, and the Min-max method is used to standardize the data.

[0101] (3) Build a case library

[0102] According to the step 3, a case library is constructed. Based on the density outlier factor elimination method, the obtained HI data set is concentrated, the outlier factors are eliminated, and the cleaned model input data is obtained as the case library.

[0103] (4) Comparative experiment of modeling and predicting battery life before and after outlier factors are removed

[0104] According to the steps 3 and 4, the battery feature data set is cleaned by using the outlier factor elimination method, and the prediction experiment is carried out using the established data model. Through comparative analysis, the prediction comparison results corresponding to the four experimental objects in the case library are as follows: Figure 3 As shown. It can be clearly seen that after the density outlier factor is removed, the data set has better correlation. Using the cleaned data, the evaluation index (MSE error value) of the model is smaller than that of the uncleaned data. In summary, the battery life prediction model after using the present invention has better prediction accuracy.

[0105] The results of this embodiment show that the present invention can improve the accuracy of the battery life prediction model and has important economic value and application value.

Claims

1. A battery life prediction method based on density outlier factor elimination, characterized by: The following steps are involved: Step 1: Charge and discharge the lithium battery, and analyze the voltage and current change curves during the charge and discharge stages, and select multiple HIs that can characterize battery degradation; Step 2: Quantitatively analyze the correlation between each HI sequence and the battery capacity, and standardize the extracted HI sequence; Step 3: Based on the local outlier factor algorithm, outliers are removed one by one to reduce redundancy and information bias and improve data density; Step 4: Use the cleaned HI sequence as the input and output sequence of the model to improve the effectiveness and accuracy of the battery life prediction model.

2. The battery life prediction method based on density outlier factor elimination according to claim 1, characterized in that: Step 1 specifically includes: Select a lithium battery for testing, and perform charge and discharge cycles on the lithium battery. Select four groups of HI sequences from the battery charge and discharge curve to reflect the degradation state of the battery, namely: constant current charging stage voltage drop end point value HI1, constant current charging time HI2, constant current discharge time HI3 and battery average internal resistance HI4, where: In the formula, It is the value of the lowest point of voltage drop after the constant current charging stage ends; It is the staged constant current charging time; is the constant current discharge time; is the measured average internal resistance of the battery.

3. The battery life prediction method based on density outlier factor elimination according to claim 1, characterized in that: Step 2 specifically includes the following processes: Step 2-1: Combine the extracted HI sequence and its corresponding capacity value to construct a battery degradation feature sequence data set {cyclc, HI1, HI2, HI3, HI4, capacity}. Combined with the battery degradation feature sequence data set, the Spearman correlation analysis method is used to quantify the correlation between each HI sequence and the battery capacity. The correlation range is [-1, 1], and the calculation method is as follows: Where n is the number of elements in variables X and Y; d i For data point x i and i The level difference, if ρ s The closer the value is to 1 or -1, the stronger the correlation between the two variables. s The closer the value is to 0, the weaker or no correlation between the two variables; Step 2-2: Use the Min-max standardization method to standardize the battery degradation feature sequence data set: Min-max standardization uses the maximum and minimum values ​​in the sequence to process the data in the sequence so that the data is mapped to the [0, 1] interval. The following is its calculation formula: Where x is the original HI input data; X max is the maximum value of HI data; X min is the minimum value of HI data; Y is the battery capacity after standardization; Step 2-3: Use the lithium battery degradation feature sequence data set {cycle, HI1, HI2, HI3, HI4, capacity} that has undergone association analysis and standardization as the input and output of the battery prediction model.

4. The battery life prediction method based on density outlier factor elimination according to claim 1 is characterized in that: Step 3 specifically includes the following processes: Step 3-1: Calculate the distance between data points: Assume that there are data points p and o in the battery degradation feature sequence data set D, where data point o is the set D / {p} other than p in the data set D. Use the Euclidean distance to calculate the distance between data point p(x p ,y p ) and the data point o(x o ,y o ) is the distance d(p,o): Step 3-2: Calculate the kth distance of data point p: Sort the distance calculation results from small to large, and find the set of k adjacent data points O = {o1, o2, ..., o k }, the distance from data point p to the farthest k-th adjacent data point is d(p, o k ), which is the kth distance of data point p, which can also be recorded as d k (p); Step 3-3: Determine the kth distance neighborhood of data point p: k The collection of all data points q in (p) is called the kth distance neighborhood of p, denoted by N k (p), which is expressed as: N k (p)={q|d(p,q)≤d k (p)} Step 3-4: Determine the reachable distance of data point p relative to data point o: The reachable distance of data point p relative to data point o is the larger value of "the kth distance of data point o" and "the Euclidean distance between data point p and data point o", and its specific expression is: reach·dist(p,o)=max{d k (o),d(p,o)} Step 3-5: Calculate the local outlier factor of data point p: The local outlier factor of data point p is expressed as the neighborhood N of data point p k The inverse of the average reachable distance of other data points in (p) is expressed as: From the above formula, we can see that the larger the average reachable distance, the smaller the local reachable density of the data point p; Step 3-6: Calculate the local outlier factor of data point p: The local outlier factor of data point p is expressed as the neighborhood N of data point p k The ratio of the average local reachable density of other data points in (p) to the local reachable density of data point p is expressed as: According to the above formula, the closer the LOF value is to 1, the more similar the average density of the data point p is to the surrounding data points o, and the more likely it is a normal value; the larger the LOF value is, the farther the data point p is from the surrounding data points, and the more likely it is an outlier.

5. The battery life prediction method based on density outlier factor elimination according to claim 1, characterized in that: Step 4 specifically includes the following processes: Step 4-1: Take the HI sequence after standardization and outlier elimination in step 3, perform prediction analysis on the battery life, and use the mean square error to evaluate the later model. The specific calculation formula is as follows: in, and i is the predicted life and actual life of the i-th battery tuple; Step 4-2: According to the method of step 4-1, the original data and the data after the density outlier factor is eliminated are used as the input of the model to predict the battery life; Step 4-3: Analyze the results of step 4-2, and compare the two methods to evaluate the accuracy of battery life prediction under the model evaluation index MSE.

Images