BP neural network dynamic parameter identification method based on discrete time sequence

Abstract

The invention discloses a BP neural network dynamic parameter identification method based on a discrete time sequence. The method specifically comprises the following steps: acquiring indexes of various dynamic parameters to form a dynamic parameter index data set; performing data normalization on the dynamic parameter index data set; calculating the index weight of the dynamic parameter index data set after data normalization; calculating an entropy value of each piece of dynamic parameter index data and determining an optimization weight of each piece of dynamic parameter index data; calculating a combined weight of the optimization weights of the dynamic parameter index data according to the optimization weights of the dynamic parameter index data; calculating a lithium battery operation state evaluation value according to the dynamic parameter index data set after data normalization and the combined weight; and performing battery pack state classification and making a control strategy according to the lithium battery operation state evaluation value. The method has the beneficial effect that the health state of the battery can be quickly and accurately predicted.

Description

Technical Field

[0001] The present invention relates to the technical field related to lithium batteries, and in particular to a BP neural network dynamic parameter identification method based on discrete time series. Background Art

[0002] The fire hazards of lithium battery energy storage systems can be roughly divided into two aspects: early thermal runaway inside lithium batteries and fire hazards of electrical equipment. The charging and discharging of lithium batteries are mainly completed by chemical reactions. Heat energy will inevitably be generated during the charging and discharging process. If the heat energy generated by the battery itself exceeds the heat dissipation capacity of the battery, the lithium battery cannot be dissipated in time, and the heat will accumulate, causing the battery to overheat, triggering thermal runaway, and generating flammable gases, which can easily cause explosions when encountering open flames. Therefore, in view of the fire and explosion hazards caused by thermal runaway and heat diffusion caused by early heat accumulation in lithium-ion batteries during the charging and discharging process, it is urgent to take complete systematic steps to prevent the occurrence of lithium battery disasters.

[0003] In the prior art, there are corresponding monitoring methods for thermal runaway of lithium-ion batteries. Such technical solutions usually design corresponding sensors for lithium-ion batteries, including sensors such as temperature and gas concentration, and monitor whether the lithium-ion battery has abnormal temperature rise or gas leakage to monitor whether the lithium-ion battery has thermal runaway. However, in the actual implementation process, since the above scheme only makes a simple judgment on a single factor, such as whether the gas concentration reaches the leakage threshold, whether the temperature rise reaches the temperature rise threshold, etc., to make an alarm, this leads to a relatively single judgment factor that is easy to misjudge, and cannot effectively predict the future status of the lithium-ion battery, resulting in insufficient judgment accuracy, and the operation method is extremely complicated. Summary of the invention

[0004] The present invention aims to overcome the above-mentioned deficiencies in the prior art and provides a BP neural network dynamic parameter identification method based on discrete time series that can quickly and accurately predict the battery health state.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A BP neural network dynamic parameter identification method based on discrete time series specifically includes the following steps:

[0007] (1) Collect indicators of multiple dynamic parameters to form a dynamic parameter indicator data set;

[0008] (2) Normalize the dynamic parameter indicator data set;

[0009] (3) Calculate the indicator weights of the dynamic parameter indicator data set after data normalization;

[0010] (4) Calculate the entropy value of each dynamic parameter index data to determine the optimization weight of each dynamic parameter index data;

[0011] (5) calculating the combined weight of the optimization weights of each dynamic parameter indicator data according to the optimization weights of each dynamic parameter indicator data;

[0012] (6) Calculate the lithium battery operating status evaluation value based on the dynamic parameter indicator data set and combined weight after data normalization;

[0013] (7) Classify the battery pack status and formulate control strategies based on the lithium battery operating status evaluation value.

[0014] The present invention establishes a multi-objective optimization function for the earliest warning threshold of the current lithium battery fault state on the timeline and automatically seeks the optimum, which can quickly and accurately predict the battery health state.

[0015] Preferably, in step (1), specifically: collect 10 indicator data of the lithium battery pack, including calorific value, gas pressure value, gas concentration value, sound wave value, membrane ion concentration value, voltage parameter, current parameter, charge state parameter, health state parameter and environmental state parameter, to form a dynamic parameter indicator data set.

[0016] Preferably, in step (2), the following steps are specifically included:

[0017] (21) Classify various indicator data in the dynamic parameter indicator data set in a discrete time series, define the data that meets the battery health status as the criterion layer, and define the data that does not meet the battery health status as the solution layer;

[0018] (22) According to the discrete degree of each indicator data in the scheme layer relative to the criterion layer, a threshold judgment matrix is ​​constructed;

[0019] (23) The threshold judgment consistency of the jth indicator corresponding to the i-th dynamic parameter is calculated according to the threshold judgment matrix.

[0020] Preferably, in step (22), the formula of the threshold judgment matrix is:

[0021]

[0022] Among them, r ij It represents the discrete degree of the jth indicator corresponding to the ith dynamic parameter, and r∈{0,1}; m is the total number of indicators of the collected lithium battery pack, n is the number of discrete time series intervals; R represents the output result of the threshold judgment matrix.

[0023] Preferably, in step (23), the calculation formula for judging the consistency of the threshold value of the jth indicator corresponding to the i-th dynamic parameter is: ij =r ij -r ji +0.5, where r ij Indicates the discrete degree of the jth indicator corresponding to the ith dynamic parameter, r ji Indicates the discrete degree of the ith indicator corresponding to the jth dynamic parameter, l ij Indicates the consistency of the threshold judgment of the jth indicator corresponding to the ith dynamic parameter.

[0024] Preferably, in step (3), the calculation formula of the indicator weight of the dynamic parameter indicator data set after data normalization is:

[0025]

[0026]

[0027] Among them, t represents the dimension of the indicator; a represents the weight difference factor; v x represents the weight of the xth indicator; v y Represents the weight of the y-th indicator.

[0028] Preferably, in step (4), the following steps are specifically included:

[0029] (41) Calculate the weight of a single indicator using the following formula:

[0030]

[0031] (42) Calculate the entropy value of each indicator in the indicator data set. The calculation formula is:

[0032]

[0033] (43) Calculate the optimization weight of each index entropy value, the calculation formula is:

[0034]

[0035] where f ij Represents the feature weight of a single indicator; l ij H represents the consistency of the threshold judgment of the jth indicator corresponding to the i-th dynamic parameter; j Represents the entropy value of each indicator in the indicator data set; W ij Represents the weight of each index entropy value; 0≤W ij ≤1, m is the total number of indicators of the collected lithium battery packs, and n is the number of discrete time series intervals.

[0036] Preferably, in step (5), the calculation formula for calculating the combined weight of the optimization weights of each dynamic parameter indicator data is as follows:

[0037]

[0038] Among them, W ij represents the weight of each index entropy value; v1, v2, v3, v4 represent the index weight of each dynamic parameter, that is, v x Represents the weight of the xth indicator.

[0039] Preferably, in step (6), the calculation formula for calculating the lithium battery operating status evaluation value based on the dynamic parameter index data set and the combined weight after data normalization is: Where F represents the evaluation value of the lithium battery operating status; b ij The combined weight representing the optimization weight of each dynamic parameter indicator data; ij Indicates the consistency of the threshold judgment of the jth indicator corresponding to the ith dynamic parameter.

[0040] Preferably, in step (7), the battery pack status is classified and a control strategy is formulated according to the lithium battery operating status evaluation value, including classifying the status in the battery pack into four levels: normal, caution, abnormal, and severe.

[0041] F = 0 ~ 0.2 is a normal state, the index is relatively stable and does not exceed the warning value;

[0042] F = 0.2-0.4 is the attention state, the index is relatively stable, and continuous observation is still needed;

[0043] F = 0.4 to 0.7 is an abnormal state, the indicator has approached or exceeded the attention value, and the indicator change is developing towards the attention value;

[0044] F=0.7~1.0 is a serious state. The indicator seriously exceeds the warning value. It is necessary to immediately cut off the current charging and discharging process of the lithium battery and issue an alarm message.

[0045] The beneficial effects of the present invention are as follows: a multi-objective optimization function is established for the earliest warning threshold of the current lithium battery fault state on the timeline and the optimization is automatically sought, so that the battery health state can be predicted quickly and accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0047] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0048] like Figure 1In the embodiment described, a BP neural network dynamic parameter identification method based on discrete time series specifically includes the following steps:

[0049] (1) Collect indicators of multiple dynamic parameters to form a dynamic parameter indicator data set;

[0050] Specifically: 10 indicator data including calorific value, gas pressure value, gas concentration value, sound wave value, membrane ion concentration value, voltage parameter, current parameter, charge state parameter, health state parameter and environmental state parameter of the lithium battery pack are collected to form a dynamic parameter indicator data set.

[0051] (2) Normalizing the dynamic parameter indicator data set; specifically, the following steps are included:

[0052] (21) Classify various indicator data in the dynamic parameter indicator data set in a discrete time series, define the data that meets the battery health status as the criterion layer, and define the data that does not meet the battery health status as the solution layer;

[0053] (22) According to the discrete degree of each indicator data in the scheme layer relative to the criterion layer, a threshold judgment matrix is ​​constructed; the formula of the threshold judgment matrix is:

[0054]

[0055] Among them, r ij It represents the discrete degree of the jth indicator corresponding to the ith dynamic parameter, and r∈{0,1}; m is the total number of indicators of the collected lithium battery pack, n is the number of discrete time series intervals, m=10, n=6; R represents the output result of the threshold judgment matrix.

[0056] (23) calculating the threshold judgment consistency of the jth indicator corresponding to the i-th dynamic parameter according to the threshold judgment matrix;

[0057] The calculation formula for the consistency of the threshold value of the jth indicator corresponding to the ith dynamic parameter is: ij =r ij -r ji +0.5, where r ij Indicates the discrete degree of the jth indicator corresponding to the ith dynamic parameter, r ji Indicates the discrete degree of the ith indicator corresponding to the jth dynamic parameter, l ij Indicates the consistency of the threshold judgment of the jth indicator corresponding to the ith dynamic parameter.

[0058] Advantages of this step: The threshold judgment matrix R is established based on discrete time series. The formula below is actually based on the BP neural network dynamic parameter identification method. The original method is to do a dynamic identification for one parameter. The method of the present invention is to do a dynamic identification for 10 indicators, which can effectively improve the precision and accuracy of identification; the purpose of this step is to prepare for the final evaluation value of the lithium battery operation status.

[0059] (3) Calculate the indicator weights of the dynamic parameter indicator data set after data normalization;

[0060] The calculation formula of the indicator weight of the dynamic parameter indicator data set after data normalization is:

[0061]

[0062]

[0063] Among them, t represents the dimension of the indicator; a represents the weight difference factor, a=t / 2, the larger the value, the more obvious the weight difference of each indicator; v x represents the weight of the xth indicator; v y Represents the weight of the y-th indicator.

[0064] The advantage of this step is that the indicator weight v will be used in the formula of step (5). The innovation of this step is to add the weight factor a, because some of the 10 indicators are more meaningful than other indicators, such as gas concentration and charge state; so the weight factor is added to distinguish different indicators, making the final judgment basis more reasonable.

[0065] (4) Calculating the entropy value of each dynamic parameter index data to determine the optimization weight of each dynamic parameter index data; specifically including the following steps:

[0066] (41) Calculate the weight of a single indicator using the following formula:

[0067]

[0068] (42) Calculate the entropy value of each indicator in the indicator data set. The calculation formula is:

[0069]

[0070] (43) Calculate the optimization weight of each index entropy value, the calculation formula is:

[0071]

[0072] where f ij Represents the feature weight of a single indicator; l ijH represents the consistency of the threshold judgment of the jth indicator corresponding to the i-th dynamic parameter; j Represents the entropy value of each indicator in the indicator data set; W ij Represents the weight of each index entropy value; 0≤W ij ≤1, m is the total number of indicators of the collected lithium battery packs, n is the number of discrete time series intervals, n=6; m=10.

[0073] The advantages of this step are: the optimized weight W will be used in the formula of step (5) to obtain the combined weight. The purpose of using the entropy method here is that the algorithm is simple, no other software analysis is required, and the computing memory consumption is reduced; and the entropy weight method can deeply reflect the distinguishing ability of the indicator and determine the better weight.

[0074] (5) calculating the combined weight of the optimization weights of each dynamic parameter indicator data according to the optimization weights of each dynamic parameter indicator data;

[0075] The calculation formula for the combined weight of the optimization weights of each dynamic parameter indicator data is calculated based on the optimization weights of each dynamic parameter indicator data:

[0076]

[0077] Among them, W ij represents the weight of each index entropy value; v1, v2, v3, v4 represent the index weight of each dynamic parameter, that is, v x Represents the weight of the xth indicator.

[0078] (6) Calculate the lithium battery operating status evaluation value based on the dynamic parameter indicator data set and combined weight after data normalization;

[0079] The calculation formula for calculating the lithium battery operating status evaluation value based on the dynamic parameter indicator data set and combined weight after data normalization is: Where F represents the evaluation value of the lithium battery operating status; b ij The combined weight representing the optimization weight of each dynamic parameter indicator data; ij Indicates the consistency of the threshold judgment of the jth indicator corresponding to the ith dynamic parameter.

[0080] The advantage of this step is that the final F is a normalized value between 0 and 1. The closer it is to 1, the closer the state of the battery pack is to a fault state; conversely, the closer it is to 0, the more stable the state of the battery pack is. By normalizing the data, the final judgment basis can be simplified; in this way, there is no need to use complex indicators as the basis for judgment later.

[0081] (7) Classify the battery pack status and formulate control strategies based on the lithium battery operating status evaluation value;

[0082] According to the evaluation value of lithium battery operation status, the battery pack status is classified and the control strategy is formulated, including dividing the status in the battery pack into four levels: normal, caution, abnormal, and severe.

[0083] F = 0 ~ 0.2 is a normal state, the index is relatively stable and does not exceed the warning value;

[0084] F = 0.2-0.4 is the attention state, the index is relatively stable, and continuous observation is still needed;

[0085] F = 0.4 to 0.7 is an abnormal state, the indicator has approached or exceeded the attention value, and the indicator change is developing towards the attention value;

[0086] F=0.7~1.0 is a serious state. The indicator seriously exceeds the warning value. It is necessary to immediately cut off the current charging and discharging process of the lithium battery and issue an alarm message.

[0087] The advantage of this step is that the F value obtained by the method and formula is used for graded management. The advantage of graded management is that sometimes the F value will be distorted due to other factors such as false touch of the sensor and weather conditions; therefore, the battery pack status is graded based on a range of F to avoid unnecessary battery maintenance.

[0088] The present invention establishes a multi-objective optimization function for the earliest warning threshold of the current lithium battery fault state on the timeline and automatically seeks the optimum, which can quickly and accurately predict the battery health state.

Claims

1. A BP neural network dynamic parameter identification method based on discrete time series, characterized in that: The specific steps include: (1) Collect indicators of multiple dynamic parameters to form a dynamic parameter indicator data set; (2) Normalize the dynamic parameter indicator data set; (3) Calculate the indicator weights of the dynamic parameter indicator data set after data normalization; (4) Calculate the entropy value of each dynamic parameter index data to determine the optimization weight of each dynamic parameter index data; (5) calculating the combined weight of the optimization weights of each dynamic parameter indicator data according to the optimization weights of each dynamic parameter indicator data; (6) Calculate the lithium battery operating status evaluation value based on the dynamic parameter indicator data set and combined weight after data normalization; (7) Classify the battery pack status and formulate control strategies based on the lithium battery operating status evaluation value.

2. The method for identifying dynamic parameters of a BP neural network based on discrete time series according to claim 1 is characterized in that: In step (1), specifically: collecting 10 indicator data of the lithium battery pack, including calorific value, gas pressure value, gas concentration value, sound wave value, membrane ion concentration value, voltage parameter, current parameter, charge state parameter, health state parameter and environmental state parameter, to form a dynamic parameter indicator data set.

3. A BP neural network dynamic parameter identification method based on discrete time series according to claim 1 or 2, characterized in that: In step (2), the following steps are specifically included: (21) Classify various indicator data in the dynamic parameter indicator data set in a discrete time series, define the data that meets the battery health status as the criterion layer, and define the data that does not meet the battery health status as the solution layer; (22) Construct a threshold judgment matrix based on the discrete degree of each indicator data in the solution layer relative to the criterion layer; (23) The threshold judgment consistency of the jth indicator corresponding to the i-th dynamic parameter is calculated according to the threshold judgment matrix.

4. The method for identifying dynamic parameters of a BP neural network based on discrete time series according to claim 3 is characterized in that: In step (22), the formula of the threshold judgment matrix is: Among them, r ij It represents the discrete degree of the jth indicator corresponding to the ith dynamic parameter, and r∈{0,1}; m is the total number of indicators of the collected lithium battery pack, n is the number of discrete time series intervals; R represents the output result of the threshold judgment matrix.

5. The method for identifying dynamic parameters of a BP neural network based on discrete time series according to claim 4 is characterized in that: In step (23), the calculation formula for the threshold value judgment consistency of the jth indicator corresponding to the i-th dynamic parameter is: ij =r ij -r ji +0.5, where r ij Indicates the discrete degree of the jth indicator corresponding to the ith dynamic parameter, r ji Indicates the discrete degree of the ith indicator corresponding to the jth dynamic parameter, l ij Indicates the consistency of the threshold judgment of the jth indicator corresponding to the ith dynamic parameter.

6. The method for identifying dynamic parameters of a BP neural network based on discrete time series according to claim 5 is characterized in that: In step (3), the calculation formula of the indicator weight of the dynamic parameter indicator data set after data normalization is: Among them, t represents the dimension of the indicator; a represents the weight difference factor; v x represents the weight of the xth indicator; v y Represents the weight of the y-th indicator.

7. The method for identifying dynamic parameters of a BP neural network based on discrete time series according to claim 5 is characterized in that: In step (4), the following steps are specifically included: (41) Calculate the weight of a single indicator using the following formula: (42) Calculate the entropy value of each indicator in the indicator data set. The calculation formula is: (43) Calculate the optimization weight of each index entropy value, the calculation formula is: where f ij Represents the feature weight of a single indicator; l ij H represents the consistency of the threshold judgment of the jth indicator corresponding to the i-th dynamic parameter; j Represents the entropy value of each indicator in the indicator data set; W ij Represents the weight of each index entropy value; 0≤W ij ≤1, m is the total number of indicators of the collected lithium battery packs, and n is the number of discrete time series intervals.

8. The method for identifying dynamic parameters of a BP neural network based on discrete time series according to claim 7 is characterized in that: In step (5), the formula for calculating the combined weight of the optimization weights of each dynamic parameter indicator data is as follows: Among them, W ij represents the weight of each index entropy value; v1, v2, v3, v4 represent the index weight of each dynamic parameter, that is, v x Represents the weight of the xth indicator.

9. The method for identifying dynamic parameters of a BP neural network based on discrete time series according to claim 8 is characterized in that: In step (6), the calculation formula for calculating the lithium battery operating status evaluation value based on the dynamic parameter indicator data set and the combined weight after data normalization is: Where F represents the evaluation value of the lithium battery operating status; b ij The combined weight representing the optimization weight of each dynamic parameter indicator data; ij Indicates the consistency of the threshold judgment of the jth indicator corresponding to the ith dynamic parameter.

10. The method for identifying dynamic parameters of a BP neural network based on discrete time series according to claim 9, characterized in that: In step (7), the battery pack status is classified and a control strategy is formulated according to the lithium battery operating status evaluation value, including classifying the status in the battery pack into four levels: normal, caution, abnormal, and severe. F = 0 ~ 0.2 is a normal state, the index is relatively stable and does not exceed the warning value; F = 0.2-0.4 is the attention state, the index is relatively stable, and continuous observation is still needed; F = 0.4 to 0.7 is an abnormal state, the indicator has approached or exceeded the attention value, and the indicator change is developing towards the attention value; F=0.7~1.0 is a serious state. The indicator seriously exceeds the warning value. It is necessary to immediately cut off the current charging and discharging process of the lithium battery and issue an alarm message.

Images