New energy vehicle power battery thermal runaway early warning method and system based on big data

Abstract

The invention discloses a new energy vehicle power battery thermal runaway early warning method and system based on big data, and belongs to the technical field of new energy vehicle battery monitoring, and the method comprises the steps: obtaining the voltage continuous time sequence data of each single battery in a battery pack, and forming a voltage matrix; obtaining a voltage data set A through a sliding window principle; if the column voltage data value of the voltage data set A is unique, adding the same periodic noise to obtain a voltage data set B; measuring-based geometric probability distribution statistics and information entropy calculation are carried out on each column of voltage data in the voltage data set A or the voltage data set B, or measuring-based geometric probability distribution statistics and frequency-based probability distribution statistics are carried out, and weighted mutual entropy calculation is carried out; abnormal value identification and positioning are carried out on information entropy or mutual entropy based on a local distance anomaly detection method in combination with a threshold value method, so that the battery fault diagnosis and thermal runaway prediction effect is not affected by the number of calculation windows and geometric probability partitions.

Description

Technical Field

[0001] The present invention relates to the technical field of battery monitoring for new energy vehicles, and in particular to a thermal runaway early warning method and system for power batteries for new energy vehicles based on big data. Background Art

[0002] Thermal runaway of lithium-ion batteries has many potential causes and complex evolution patterns, which are highly concealed before an accident occurs. Therefore, studying the accurate and early warning of thermal runaway is an important means to improve the safety of power battery systems. When thermal runaway occurs, battery parameters (such as temperature, voltage, current, and gas concentration) will change. Therefore, using the changing battery parameters as identification parameters for thermal runaway and studying the evolution of the behavior characteristics of power batteries in the short term before thermal runaway occurs can provide an important research basis for studying the early warning mechanism of thermal runaway.

[0003] At present, various entropy and improved algorithms combined with voltage data have been applied to the field of battery fault diagnosis and have achieved certain results. However, there are also some disadvantages. For example, the current battery fault diagnosis method based on Shannon entropy of histogram frequency statistics is related to the calculation window of Shannon entropy and the optimal number of histogram partitions, while the battery fault diagnosis method based on sample entropy is affected by the scale factor, which is specifically related to the reconstruction vector size m and the similarity r value. The calculation window or reconstruction vector m and similarity r that play a decisive role in fault diagnosis are mainly obtained in the form of trial and error, which is labor-intensive. Summary of the invention

[0004] In order to solve the problems of the prior art, the present invention provides a new energy vehicle power battery thermal runaway early warning method and system based on big data;

[0005] On the one hand, a new energy vehicle power battery thermal runaway early warning method based on big data is provided, the method comprising:

[0006] Obtain the continuous time series data of the voltage of each single cell in the power battery pack of new energy vehicles to form a voltage matrix;

[0007] According to a window with a preset length of m and a step size of 1, the voltage matrix is ​​slid from top to bottom to obtain a voltage data set A under the current sliding window;

[0008] The voltage data value of each column of the voltage data set A is judged, and if there is a unique voltage data value in a column, the same periodic noise is added to the voltage data of all columns in the voltage data set A to obtain a voltage data set B with noise added in the current window;

[0009] If it does not exist, it will not be processed;

[0010] Performing geometric probability distribution statistics based on measurement and information entropy calculation on each column of voltage data in the voltage data set A or the voltage data set B to obtain the information entropy of each column of voltage data in the current window; or performing geometric probability distribution statistics based on measurement and probability distribution statistics based on frequency, and performing weighted mutual entropy value calculation to obtain the mutual entropy of each column of voltage data in the current window;

[0011] An improved local distance-based anomaly detection method combined with a threshold method is used to identify and locate outliers on the information entropy or mutual entropy of the voltage data set A or the voltage data set B, thereby completing the prediction of thermal runaway of the power battery.

[0012] Furthermore, the same periodic noise is one of a discrete square wave signal, a sine signal and a cosine signal, the amplitude of the sine signal or the cosine signal is between one thousandth and one ten-thousandth of the average amplitude of the current voltage signal, and the added discrete square wave signal or sine signal or cosine signal period is the sliding window length m; the discrete square wave signal function is:

[0013]

[0014] Where f(t) is a discrete square wave signal function with an amplitude of ω and a period of m.

[0015] Further, the geometric probability distribution statistics based on the measure are probability distribution statistics based on quantile intervals;

[0016] The specific method is: taking each column of voltage data of the voltage data set A in the current window or each column of voltage data of the voltage data set B with noise in the current window as the object, arranging the voltage data in ascending order and then dividing them into K equal intervals, and obtaining the quantile corresponding to each equal division point at the same time, calculating the difference of quantiles corresponding to adjacent equal division points, and the ratio of the difference of quantiles corresponding to adjacent equal division points to the sum of the differences of quantiles corresponding to all adjacent equal division points is the probability distribution P based on the quantile interval.

[0017] Furthermore, the information entropy calculation specifically includes:

[0018] Based on the probability distribution P of each single cell based on the quantile interval under each sliding window, the information entropy H of each single cell under the window is calculated using the entropy formula. The formula is as follows:

[0019]

[0020] in, It represents the ratio of the difference between the quantile corresponding to the i+1th equal division point of the Nth single cell in the Cth window and the quantile corresponding to the i-th equal division point to the sum of the differences between the quantiles corresponding to all adjacent equal division points of the Nth single cell in the window.

[0021] Furthermore, the specific method of frequency-based probability distribution statistics is:

[0022] Take each column of voltage data of the voltage data set A in the current window or each column of voltage data of the voltage data set B with noise in the current window as the object, and take the maximum voltage value v in the current window. max and minimum voltage value v min As the upper limit of the preset interval v upper and the lower limit v lower , set the lower limit v lower and upper limit v upper The area between is divided into K equal intervals, and the distribution frequency of each column of voltage data values ​​in each interval is calculated as the frequency-based probability distribution Q.

[0023] Furthermore, the weighted mutual entropy value calculation specifically includes:

[0024] Based on the probability distribution P of each column of voltage data based on the quantile interval in the current window and the probability distribution Q of each column of voltage data value based on the frequency, the weighted divergence D of the two probability distributions is calculated. JS (P||Q), the formula is as follows:

[0025] D JS (P||Q)=(1-a)D KL (P||M)+aD KL (Q||M)

[0026] Where M is the average distribution of P and Q, a is the weighting coefficient, which takes values ​​between [0, 1]; D KL (P||M) represents the relative entropy of distribution P and distribution M, D KL (Q||M) represents the relative entropy of distribution Q and distribution M;

[0027] d=1-D JS (P||Q) is the mutual entropy of the similarity probability of each column of voltage data in the current window under distribution P and distribution Q.

[0028] Furthermore, the value of m is between 30 and 300;

[0029] If the geometric probability distribution statistics based on the measure are used and the information entropy is calculated, the value of m ranges from 30 to 300; if the geometric probability distribution statistics based on the measure and the probability distribution statistics based on the frequency are used and the mutual entropy value is calculated, the value of m ranges from 250 to 300.

[0030] Furthermore, the improved local distance-based anomaly detection method is a local outlier factor detection method based on a clustering algorithm, specifically:

[0031] First, a clustering method is used to classify the information entropy of all single cells in the voltage data set A or the voltage data set B in the current window, or a clustering method is used to classify the mutual entropy of the voltage of all single cells in the voltage data set A or the voltage data set B in the current window;

[0032] Calculate the average distance r from all points in each class to the current cluster center under the current window;

[0033] Eliminate the points whose distance from each class to the current cluster center is less than the average distance r in the current window;

[0034] Make local anomaly factors for the remaining points retained in all classes under the current window.

[0035] Furthermore, the use of the improved local distance-based anomaly detection method combined with the threshold method to identify and locate outliers on the information entropy or mutual entropy of the voltage data set A or the voltage data set B specifically includes:

[0036] Setting thresholds: setting sliding windows with lengths in the interval [10, 300] respectively, and calculating the local abnormal factors of the information entropy or mutual entropy of each normal single cell in the power battery pack under each length window;

[0037] The maximum value of the local anomaly factor of each length window under all sliding windows is counted, and the maximum value of the local anomaly factor under all length windows is taken as the set threshold;

[0038] Outlier identification and location: If in the current time window, the local abnormal factor of the information entropy of the single cell in the power battery pack or the local abnormal factor of the mutual entropy of the single cell does not exceed the set threshold, then the single cells in the current power battery pack are in a safe state; otherwise, it means that there is a single cell at risk of thermal runaway at the current time, and the number of the single cell whose local abnormal factor exceeds the threshold is recorded.

[0039] On the other hand, a new energy vehicle power battery thermal runaway warning system based on big data is provided, comprising: a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the steps of the new energy vehicle power battery thermal runaway warning method based on big data can be implemented.

[0040] The technical solution provided by the embodiment of the present invention has the following beneficial effects: the present invention utilizes a single information entropy value calculation method based on a measure of geometric probability distribution, so that the battery fault diagnosis and thermal runaway prediction effects are not affected by the number of calculation windows and geometric probability partitions, and has strong robustness and versatility; and a weighted mutual entropy value calculation method is performed by combining the measure-based geometric probability distribution statistics and the frequency-based probability distribution statistics. This entropy measurement method that is not biased towards sample size can avoid the problem of insensitivity of sparse samples and low-probability areas to outliers caused by only calculating entropy values ​​through frequency-based probability distribution statistics.

[0041] In addition, the present invention utilizes an improved local distance-based anomaly detection method, and utilizes a clustering method to first screen out suspicious entropy values ​​to be further diagnosed from all entropy values. There is no need to calculate the local anomaly coefficients of all entropy values, which greatly reduces the amount of calculation of the local anomaly coefficients and greatly shortens the algorithm operation time. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0043] Figure 1 It is a flow chart of a thermal runaway early warning method for power batteries of new energy vehicles based on big data provided by the present invention;

[0044] Figure 2 The present invention provides a timing diagram of the voltage of all single cells on the day when a thermal runaway accident occurs;

[0045] Figure 3 The present invention provides a total of 37 frames of all voltage time series data before and after a thermal runaway accident occurs;

[0046] Figure 4 The invention provides a weighted mutual entropy graph of geometric probability distribution statistics based on measurement and probability distribution statistics based on frequency under all windows;

[0047] Figure 5 The invention provides an improved local outlier factor value map based on mutual entropy under all windows. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0049] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0050] Embodiment 1

[0051] A thermal runaway warning method for power batteries of new energy vehicles based on big data, see Figure 1 , specifically including the following steps:

[0052] Step (1): Obtain the continuous time series data of the voltage of each single cell in the power battery pack of the new energy vehicle to form a voltage matrix. The structure of the voltage matrix is ​​as follows (1):

[0053]

[0054] In formula (1), T is the number of samples, N is the total number of single cells in the power battery pack, and u TN Indicates the voltage value of the Nth single cell at the Tth sampling time.

[0055] Step (2): According to a window with a preset length of m and a step size of 1, slide the voltage matrix from top to bottom to obtain the voltage data set A under the current sliding window.

[0056] It should be noted that the specific value range of the preset length m is determined by calculating the information entropy based on the measurement-based geometric probability distribution statistics or calculating the mutual entropy value based on the measurement-based geometric probability distribution statistics and the frequency-based probability distribution statistics in the following step (4).

[0057] If the geometric probability distribution statistics based on the measure are used and the information entropy is calculated, the value of m is between 30 and 300; if the geometric probability distribution statistics based on the measure and the probability distribution statistics based on the frequency are used and the mutual entropy value is calculated, the value of m is between 250 and 300, m< <T。

[0058] According to the sliding window principle, according to the window with a preset length of m and a step size of 1, the voltage data set A under the current sliding window is obtained, and the structure is as follows:

[0059]

[0060] In formula (2), c is the number of windows, c = T-m+1, represents the voltage value sequence of all single cells in the cth window with a length of m, It represents the voltage value of the nth single cell at time t in the cth window with a length of m.

[0061] Step (3): judge and process the voltage data value of each column in the voltage data set A under the current sliding window.

[0062] It should be noted that in the following step (4), when performing geometric probability distribution statistics based on measurement for each column of voltage data in the current window, in order to avoid the situation where a column of voltage data in the current window has a unique value, the quantile interval of the column is calculated to be 0, resulting in subsequent calculation problems. It is necessary to judge whether the value of each column of voltage data in the current window is unique and perform corresponding processing. The specific measures are as follows:

[0063] If there is a unique column voltage data value, add the same periodic noise to the voltage data of all columns in the current window to obtain the voltage data set B with noise in the current window, where the same periodic noise added is specifically one of a discrete square wave signal, a sine signal and a cosine signal, and the amplitude of the sine signal or the cosine signal is between one thousandth and one ten-thousandth of the average amplitude of the current voltage signal, and the added discrete square wave signal, sine signal or cosine signal period is the sliding window length m; wherein the discrete square wave signal function is:

[0064]

[0065] In formula (3), f(t) is a discrete square wave signal function with an amplitude of ω and a period of m. In this embodiment, ω is 1.

[0066] In this embodiment, the added periodic noise signal is specifically a sinusoidal signal, specifically f=a*sin(w)+b, where w=1, 2, ..., m; a=0.0001, b=0; then the voltage data set B with noise is obtained as shown in the following formula:

[0067]

[0068] In formula (4), c is the number of windows, c = T-m+1, represents the voltage value sequence of all single cells with noise in the cth window of length m, represents the noisy voltage value of the nth single cell at time t in the cth window with a length of m. In this embodiment,

[0069] If there is no unique column voltage data value, the voltage data set A under the current sliding window does not need to be processed.

[0070] Step (4): Perform geometric probability distribution statistics based on measurement and information entropy calculation on each column of voltage data in the voltage data set A or voltage data set B in the current window to obtain the information entropy of each column of voltage data in the current window; or perform geometric probability distribution statistics based on measurement and probability distribution statistics based on frequency, and perform weighted mutual entropy value calculation to obtain the mutual entropy of each column of voltage data in the current window.

[0071] Among them, the measurement-based geometric probability distribution statistics are probability distribution statistics based on quantile intervals. The specific method is: taking each column of voltage data in the voltage data set A or the voltage data set B in the current window as the object, the voltage data is arranged in ascending order and then divided into K equal intervals, and the quantile corresponding to each equal division point is obtained at the same time, and the difference between the quantiles corresponding to adjacent equal division points is further calculated. The ratio of the difference between the quantiles corresponding to adjacent equal division points to the sum of the differences between the quantiles corresponding to all adjacent equal division points is the probability distribution P based on the quantile interval.

[0072] In this embodiment, the value of K is equal to the number of cells N in the power battery pack. In Matlab, the sort function is first used to sort each column of voltage data in the current window in ascending order. Then, the linspace function is used to divide the interval 0-1 into N+1 equal intervals, that is, t=linspace(0, 1, N+1). Then, the quantile function is used to obtain the quantile value of each equal interval of each column of voltage data in the current window after ascending order, that is, S i =quantile(x i , t), where x i is the voltage data sequence of the i-th column in the current window in ascending order, S i S is the corresponding quantile value of the voltage data sequence in the i-th column in the current window after being divided into N+1 equal parts in ascending order; i Subtract the adjacent quantile values ​​in L i =S i (2: end)-S i (1: end-1); further divide the value obtained by subtracting the adjacent quantile values ​​in Si by the sum of the values ​​obtained by subtracting all the adjacent quantile values ​​in Si, that is, P = L i. / sum(L i ), which is the probability distribution P based on the quantile interval of the voltage data in the i-th column in the current window after being sorted in ascending order.

[0073] Based on the probability distribution P of each single cell based on the quantile interval under each sliding window, the information entropy formula is used to calculate the information entropy H of each single cell under the window. The formula is as follows:

[0074]

[0075] In formula (5), represents the information entropy of the Nth single cell in the Cth window, It represents the ratio of the difference between the quantile corresponding to the i+1th equal division point of the Nth single cell in the Cth window and the quantile corresponding to the i-th equal division point to the sum of the differences between the quantiles corresponding to all adjacent equal division points of the Nth single cell in the window.

[0076] Based on the probability distribution statistics of frequency, the specific method is as follows: take each column of voltage data in the voltage data set A or voltage data set B under the current window as the object, and take the maximum voltage value v in the current window max and minimum voltage value v min As the upper limit of the preset interval v upper and the lower limit v lower , set the lower limit v lower and upper limit v upper The area between is divided into K equal intervals, and the distribution frequency of each column of voltage data values ​​in each interval is calculated as the frequency-based probability distribution Q.

[0077] In this embodiment, the voltage interval in the current window is divided into K intervals with equal spacing, and the voltage range of each interval is:

[0078] Where j=0, 1, 2, ..., K-1.

[0079] Count the frequency F of each column of voltage data in each interval under the current window, that is, where i = 1, 2, ..., N, Indicates that the voltage data of the ith single cell in the current window falls within the Kth interval voltage range, that is, Calculate the frequency Q of each column of voltage data in each interval under the current window i .

[0080] in Q i That is, the probability distribution of the voltage data of the i-th single cell in the current window based on the frequency.

[0081] The weighted mutual entropy value is calculated based on the geometric probability distribution statistics of the measure and the probability distribution statistics based on the frequency. The specific method is: based on the probability distribution P of each column of voltage data based on the quantile interval in the current window and the probability distribution Q of each column of voltage data value based on the frequency, the weighted divergence D of the two probability distributions is calculated. JS (P||Q), the formula is as follows:

[0082] DJS (P||Q)=(1-a)D KL (P||M)+aD KL (Q||M) (6)

[0083] In formula (6), M is the average distribution of P and Q, a is the weighting coefficient, which takes values ​​between [0, 1]; D KL (P||M) represents the relative entropy of distribution P and distribution M, D KL (Q||M) represents the relative entropy of distribution Q and distribution M, and the formula is as follows:

[0084]

[0085]

[0086] d=1-D JS (P||Q) is the mutual entropy of the similarity probability of each column of voltage data in the current window under the probability distribution P based on quantile spacing and the probability distribution Q based on frequency. In this embodiment, a=0.5.

[0087] Step (5): Use the improved local distance-based anomaly detection method combined with the threshold method to identify and locate outliers on the information entropy or mutual entropy of the voltage data set A or the voltage data set B in the current window to complete the thermal runaway prediction of the power battery.

[0088] Among them, the improved local distance-based anomaly detection method is specifically a local outlier factor detection method based on clustering. The clustering algorithm can be any one of K-Means clustering, AP clustering, hierarchical clustering, DBSCAN clustering, and GMM clustering. In this embodiment, K-Means clustering is preferred.

[0089] Based on the improved local distance-based anomaly detection method, the specific method is as follows:

[0090] Firstly, a clustering method is used to classify the information entropy of the voltages of all the single cells in the voltage data set A or the voltage data set B in the current window; or the mutual entropy of the voltages of all the single cells in the voltage data set A or the voltage data set B in the current window is classified.

[0091] Calculate the average distance r from all points in each class in the current window to the current cluster center point; remove the points in each class in the current window whose distance to the current cluster center point is less than the average distance r; and use the remaining points of all classes in the current window as local anomaly factors.

[0092] Among them, the local anomaly factor algorithm, namely the LOF algorithm, is an unsupervised learning method that does not require any prior labels or markers. In this embodiment, the local anomaly factor of the remaining points in the current window can be directly called by the LOF function in Matlab. In its LOF function, it is necessary to set the number of k neighboring points of each point in advance to calculate the reachable distance of each point based on the k neighboring points, and further calculate its local reachable density and finally the local anomaly factor of each point. In this embodiment, k is 15.

[0093] After calculating the local outlier factors of the remaining points in the current window, it is necessary to further detect abnormal local outlier factors. The detection method can use the Laida criterion, i.e., the 3σ principle, the threshold method, or the clustering method (such as GMM clustering). In this embodiment, the threshold method is preferably used to predict thermal runaway of the power battery.

[0094] The specific method of calibrating the threshold in the threshold method is:

[0095] The battery time series data of several vehicles with normal power batteries within a certain time span are extracted, and a sliding window is set in the interval of [10, 300] with a sliding window length interval of 20. The local abnormality factor of the information entropy of each single battery in the power battery pack or the mutual entropy of each single battery under each length window is calculated for the battery time series data of several vehicles with normal batteries within a certain time span, and the maximum value of the local abnormality factor of each single battery or the mutual entropy of each single battery under all length windows of the battery time series data of several vehicles with normal batteries within a certain time span is set as the threshold.

[0096] If in the current time window, the local abnormal factor of the information entropy of the single cell in the power battery pack or the local abnormal factor of the mutual entropy of the single cell does not exceed the set threshold value, then the single cells in the current power battery pack are in a safe state; otherwise, it means that there is a single cell at risk of thermal runaway at the current time, and the number of the single cell whose local abnormal factor exceeds the threshold value is recorded at the same time.

[0097] Embodiment 2

[0098] A new energy vehicle power battery thermal runaway warning system based on big data includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the steps of the new energy vehicle power battery thermal runaway warning method based on big data in embodiment one can be implemented.

[0099] It should be noted that see Figure 2 , Figure 2This is the voltage timing diagram of all the single cells (156 cells) of the vehicle on the day when the thermal runaway accident occurred. There are a total of 1724 frames of data on that day, and the interval between each data is 20s. It can be roughly seen that the voltage timing data is in the normal voltage amplitude range for most of the time in the front, and the consistency is strong. Only at the end moment, some single cells have voltage fluctuations, and one cell has a gradual voltage drop. At the last moment, the voltage of several single cells increases and decreases sharply.

[0100] See also Figure 3 , Figure 3 Based on Figure 2 The partial enlarged diagram of the last 37 frames of voltage timing data of all single cells (156 cells) of the vehicle with thermal runaway accident clearly shows that at the moment of 1690 frame, the voltage of single cell No. 148 began to deviate from the voltage of other single cells, and continued until the 173th frame when the battery voltage dropped to 0. Between the 1690th frame and the 1721st frame, the voltages of other single cells were normal. At the 1722th frame, the voltages of single cells No. 147, No. 149, and No. 150, which were adjacent to the 148th battery, all mutated. As of the 1724th frame, no voltage data was transmitted, indicating that thermal runaway has caused the internal circuit of the power supply to burn out. Among them, through the battery temperature backtracking, the battery temperature reached 122 degrees at the moment of 1720 frame, which can be considered as thermal runaway.

[0101] See also Figure 4 , Figure 4 Based on Figure 2 The weighted mutual entropy calculation results of the voltage timing diagram of all single cells (156 cells) of the vehicle on the day of the thermal runaway accident were made based on the measurement-based geometric probability distribution statistics and the frequency-based probability distribution statistics. The window length was set to WS=300, the measurement-based geometric probability distribution equal interval and the frequency-based probability distribution equal interval K were set to 156, and the weighting coefficient a was set to 0.1. The mutual entropy values ​​of each column of voltage under all windows were obtained. It can be seen that after 1392 iterations, the mutual entropy of cell No. 148 deviated significantly from the mutual entropy of other single cells.

[0102] See also Figure 5 , Figure 5 Based on Figure 4The mutual entropy value of each column voltage under all windows is based on the improved local outlier factor value. In this embodiment, k in the LOF function is 15, and the threshold is set to 10. It can be clearly seen that before the thermal runaway occurred on that day, the local outlier factor values ​​of all single cells were less than 10, and the voltage area was normal. As the iteration progressed, at the 1392th iteration, that is, when the actual data was at the 1691th frame, the No. 148 single cell suddenly exceeded the threshold value of 10 and continued, indicating that the vehicle was at risk of thermal runaway. Combined with the previous analysis, thermal runaway occurred at the 1720th frame. It can be seen that this method can effectively provide early warning of possible thermal runaway phenomena.

[0103] It is worth noting that the present invention utilizes a single information entropy value calculation method based on measurement-based geometric probability distribution, so that the battery fault diagnosis and thermal runaway prediction effects are not affected by the number of calculation windows and geometric probability partitions, and has strong robustness and versatility; and combines the measurement-based geometric probability distribution statistics and the frequency-based probability distribution statistics to perform a weighted mutual entropy value calculation method. This entropy measurement method that is not biased towards sample size can avoid the problem of insensitivity of sparse samples and low-probability areas to outliers caused by only calculating entropy values ​​through frequency-based probability distribution statistics.

[0104] In addition, the present invention utilizes an improved local distance-based anomaly detection method, and utilizes a clustering method to first screen out suspicious entropy values ​​to be further diagnosed from all entropy values. There is no need to calculate the local anomaly coefficients of all entropy values, which greatly reduces the amount of calculation of the local anomaly coefficients and greatly shortens the algorithm operation time.

[0105] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0106] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A thermal runaway early warning method for power batteries of new energy vehicles based on big data, characterized in that: The method comprises: Obtain the continuous time series data of the voltage of each single cell in the power battery pack of new energy vehicles to form a voltage matrix; According to a window with a preset length of m and a step size of 1, the voltage matrix is ​​slid from top to bottom to obtain a voltage data set A under the current sliding window; The voltage data value of each column of the voltage data set A is judged, and if there is a unique voltage data value in a column, the same periodic noise is added to the voltage data of all columns in the voltage data set A to obtain a voltage data set B with noise added in the current window; If it does not exist, it will not be processed; Performing geometric probability distribution statistics based on measurement and information entropy calculation on each column of voltage data in the voltage data set A or the voltage data set B to obtain the information entropy of each column of voltage data in the current window; or performing geometric probability distribution statistics based on measurement and probability distribution statistics based on frequency, and performing weighted mutual entropy value calculation to obtain the mutual entropy of each column of voltage data in the current window; An improved local distance-based anomaly detection method combined with a threshold method is used to identify and locate outliers on the information entropy or mutual entropy of the voltage data set A or the voltage data set B, thereby completing the prediction of thermal runaway of the power battery.

2. The big data-based thermal runaway early warning method for new energy vehicle power batteries according to claim 1 is characterized in that: The same periodic noise is one of a discrete square wave signal, a sine signal and a cosine signal. The amplitude of the sine signal or the cosine signal is between one thousandth and one ten-thousandth of the average amplitude of the current voltage signal. The added discrete square wave signal, sine signal or cosine signal period is the sliding window length m. The discrete square wave signal function is: Where f(t) is a discrete square wave signal function with an amplitude of ω and a period of m.

3. The thermal runaway early warning method for power batteries of new energy vehicles based on big data according to claim 2 is characterized in that: The geometric probability distribution statistics based on the measure are probability distribution statistics based on quantile intervals; The specific method is: taking each column of voltage data of the voltage data set A in the current window or each column of voltage data of the voltage data set B with noise in the current window as the object, arranging the voltage data in ascending order and then dividing them into K equal intervals, and obtaining the quantile corresponding to each equal division point at the same time, calculating the difference of quantiles corresponding to adjacent equal division points, and the ratio of the difference of quantiles corresponding to adjacent equal division points to the sum of the differences of quantiles corresponding to all adjacent equal division points is the probability distribution P based on the quantile interval.

4. The thermal runaway early warning method for power batteries of new energy vehicles based on big data according to claim 3 is characterized in that: The information entropy calculation specifically includes: Based on the probability distribution P of each single cell based on the quantile interval under each sliding window, the information entropy H of each single cell under the window is calculated using the entropy formula. The formula is as follows: in, It represents the ratio of the difference between the quantile corresponding to the i+1th equal division point of the Nth single cell in the Cth window and the quantile corresponding to the i-th equal division point to the sum of the differences between the quantiles corresponding to all adjacent equal division points of the Nth single cell in the window.

5. The big data-based thermal runaway early warning method for new energy vehicle power batteries according to claim 4 is characterized in that: The specific method of probability distribution statistics based on frequency is: Take each column of voltage data of the voltage data set A in the current window or each column of voltage data of the voltage data set B with noise in the current window as the object, and take the maximum voltage value v in the current window. max and minimum voltage value v min As the upper limit of the preset interval v upper and the lower limit v lower , set the lower limit v lower and upper limit v upper The area between is divided into K equal intervals, and the distribution frequency of each column of voltage data values ​​in each interval is calculated as the frequency-based probability distribution Q.

6. The big data-based thermal runaway early warning method for new energy vehicle power batteries according to claim 5 is characterized in that: The weighted mutual entropy value calculation specifically includes: Based on the probability distribution P of each column of voltage data based on the quantile interval in the current window and the probability distribution Q of each column of voltage data value based on the frequency, the weighted divergence D of the two probability distributions is calculated. JS (P||Q), the formula is as follows: D JS (P||Q)=(1-a)D KL (P||M)+aD KL (Q||M) Where M is the average distribution of P and Q, a is the weighting coefficient, which takes values ​​between [0, 1]; D KL (P||M) represents the relative entropy of distribution P and distribution M, D KL (Q||M) represents the relative entropy of distribution Q and distribution M; d=1-D JS (P||Q) is the mutual entropy of the similarity probability of each column of voltage data in the current window under distribution P and distribution Q.

7. The big data-based thermal runaway early warning method for new energy vehicle power batteries according to claim 6 is characterized in that: The value of m is between 30 and 300; If the geometric probability distribution statistics based on the measure are used and the information entropy is calculated, the value of m ranges from 30 to 300; if the geometric probability distribution statistics based on the measure and the probability distribution statistics based on the frequency are used and the mutual entropy value is calculated, the value of m ranges from 250 to 300.

8. The big data-based thermal runaway early warning method for new energy vehicle power batteries according to claim 6 is characterized in that: The improved local distance-based anomaly detection method is a local outlier factor detection method based on a clustering algorithm, specifically: First, a clustering method is used to classify the information entropy of all single cells in the voltage data set A or the voltage data set B in the current window, or a clustering method is used to classify the mutual entropy of the voltage of all single cells in the voltage data set A or the voltage data set B in the current window; Calculate the average distance r from all points in each class to the current cluster center under the current window; Eliminate the points whose distance from each class to the current cluster center is less than the average distance r in the current window; Make local anomaly factors for the remaining points retained in all classes under the current window.

9. The big data-based thermal runaway early warning method for new energy vehicle power batteries according to claim 8, characterized in that: The method of using the improved local distance-based anomaly detection method combined with the threshold method to identify and locate the outliers of the information entropy or mutual entropy of the voltage data set A or the voltage data set B specifically includes: Setting thresholds: setting sliding windows with lengths in the interval [10, 300] respectively, and calculating the local abnormal factors of the information entropy or mutual entropy of each normal single cell in the power battery pack under each length window; The maximum value of the local anomaly factor of each length window under all sliding windows is counted, and the maximum value of the local anomaly factor under all length windows is taken as the set threshold; Outlier identification and location: If in the current time window, the local abnormal factor of the information entropy of the single cell in the power battery pack or the local abnormal factor of the mutual entropy of the single cell does not exceed the set threshold, then the single cells in the current power battery pack are in a safe state; otherwise, it means that there is a single cell at risk of thermal runaway at the current time, and the number of the single cell whose local abnormal factor exceeds the threshold is recorded.

10. A thermal runaway warning system for power batteries of new energy vehicles based on big data, characterized in that: include: A memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the steps of the thermal runaway warning method for power batteries of new energy vehicles based on big data can be implemented as described in any one of claims 1 to 9.

Images