Lithium battery pack multi-fault diagnosis method

Abstract

The invention belongs to the field of lithium battery pack fault diagnosis, and relates to a lithium battery pack multi-fault diagnosis method, which comprises the steps that a diagnosis topological structure is established, the diagnosis topological structure comprises N sub-modules with the same structure, and the N sub-modules are connected in series; performing computer simulation on the diagnosis topological structure to find out important features in fault classification; taking the important characteristics as input, taking different fault types corresponding to a change curve of the important characteristics as output, training an ET algorithm, and in the training process, adjusting parameters of the ET algorithm by adopting a particle swarm algorithm to obtain an optimal classification positioning model; polynomial transformation is carried out on measured values collected by the sensor in the diagnosis process, important features are obtained, and the important features are input into the optimal classification and positioning model for fault classification and positioning. The problem that in an existing testing method, each battery cell needs to be monitored, and a large number of monitoring sensors are needed is solved, and the safety and stability of the lithium battery pack can be improved.

Description

Technical Field

[0001] The present invention belongs to the field of lithium battery pack fault diagnosis, and specifically relates to a lithium battery pack multi-fault diagnosis method. Background Art

[0002] As the proportion of renewable energy in new power systems increases, the use of lithium battery packs becomes more common. Since thermal runaway in lithium-ion batteries can cause irreversible damage to the entire system, it is important to analyze and detect faults that may lead to thermal runaway. At the same time, timely identification of the location of the faulty battery also improves the overall service life of electric vehicles.

[0003] Thermal runaway in lithium-ion batteries mainly includes internal short circuit (ISC) faults and cell open circuit (COC) faults. In practical applications, when the battery system is running, the fault type is not fixed. Current detection methods require detection time of up to several hours, or every unit in the system must be monitored, which means that many sensors must be used and a large amount of data must be processed, which is very inconvenient for fault diagnosis. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a multi-fault diagnosis method for a lithium battery pack, which solves the problem that each battery cell needs to be monitored and a large number of monitoring sensors are required in the existing testing method.

[0005] The present invention is achieved in this way.

[0006] A lithium battery pack multi-fault diagnosis method, the method comprising:

[0007] Establish a diagnostic topology structure, the diagnostic topology structure includes N submodules with the same structure, the submodule includes 16 lithium iron phosphate batteries, a first current sensor, a second current sensor and a voltage sensor, two lithium iron phosphate batteries are connected in series to form a series group, the connection end between the two lithium iron phosphate batteries forms a node, all the series groups are connected in parallel, four series groups form a first array group, and the other four series groups form a second array group, one measuring end of the first current sensor is connected to four nodes of the first array group, and the other measuring end of the first current sensor is connected to four nodes of the second array group; one measuring end of another second current sensor is connected to one of the nodes of the first array group, and the other measuring end is connected to one of the nodes of the second array group; the two ends of the voltage sensor are connected between any two nodes in the first array group or the second array group;

[0008] N submodules are connected in series;

[0009] The diagnostic topology is simulated by a computer, and the curves of the measured values ​​of the current sensor and the voltage sensor and the polynomial transformation of the measured values ​​over time under different fault types are simulated to find the prominent change curves corresponding to different faults and the current sensor, voltage sensor or / and polynomial transformation corresponding to the change curves; the importance analysis of the curves of the measured values ​​of the current sensor and the voltage sensor and the polynomial transformation of the measured values ​​over time under multiple faults is performed to find the important features in the fault classification;

[0010] The important characteristics are used as input, and the different fault types corresponding to the change curves of the important characteristics are used as output to train the ET algorithm. During the training process, the particle swarm algorithm is used to adjust the parameters of the ET algorithm to obtain the optimal classification and positioning model.

[0011] The measurement values ​​collected by the sensors in the diagnosis process are transformed into polynomials, and important features are obtained. The important features are input into the optimal classification and positioning model for fault classification and positioning.

[0012] Furthermore, the polynomial transformation refers to expanding the measured values ​​of the current sensor and the voltage sensor into multiple features by making differences and ratios between each other.

[0013] Furthermore, the important features refer to features that contribute to all faults when a fault occurs and have different contribution forms for different faults.

[0014] Furthermore, a pre-judgment is performed before the polynomial transformation is performed on the measurement values ​​collected by the sensors during the diagnosis process. The pre-judgment includes determining to which submodule the measurement values ​​of the sensors belong the changed measurement values.

[0015] Furthermore, the pre-judgment also includes determining that no fault has occurred when none of the measured values ​​in the judgment submodule has changed.

[0016] Furthermore, the pre-judgment also includes determining the number of sudden changes in the measured values ​​of the sensors in the submodule. When the number is equal to 1, it is considered as a sensor failure. When the number is greater than or equal to 2, it is considered as a battery failure.

[0017] Furthermore, the particle swarm algorithm is used to adjust the parameters of the ET algorithm including the number of decision trees, the maximum depth of the decision tree, the minimum number of samples required for node splitting, and the minimum number of samples required for leaf nodes.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] The topological structure of the present invention can detect and locate each submodule using one voltage sensor and two current sensors, and each submodule can detect internal faults, open circuit faults and sensor faults in a battery pack consisting of 16 lithium iron phosphate batteries. With a very small number of sensors, the method of the present invention reduces the amount of calculation while improving the accuracy and efficiency of fault diagnosis. At the same time, the present invention adds the diagnosis of sensor faults to ensure the accuracy and reliability of monitoring data. The method of the present invention will help improve the safety and stability of lithium battery packs and reduce the risks and losses caused by lithium battery failures. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A schematic diagram of the structure of the diagnostic topology in the lithium battery pack multi-fault diagnosis method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0022] like Figure 1 As shown, a multi-fault diagnosis method for a lithium battery pack is described by taking two submodules as an example. The method can simultaneously detect and locate the faults of 32 lithium iron phosphate batteries. The method includes:

[0023] A diagnostic topology is established, wherein the diagnostic topology includes a first submodule and a second submodule having the same structure, wherein the first submodule includes 16 lithium iron phosphate batteries, a first current sensor A1, a second current sensor A2, and a first voltage sensor V1, wherein two lithium iron phosphate batteries are connected in series to form a series group, wherein the connection end between the two lithium iron phosphate batteries forms a node, wherein all series groups are connected in parallel, wherein four series groups form a first array group, and the other four series groups form a second array group, wherein one measuring end of the first current sensor A1 is connected to four nodes of the first array group, and another measuring end of the first current sensor A1 is connected to four nodes of the second array group; wherein one measuring end of another second current sensor A2 is connected to one of the nodes of the first array group, and the other measuring end is connected to one of the nodes of the second array group; and both ends of the first voltage sensor V1 are connected between any two nodes in the first array group or the second array group;

[0024] The second submodule includes a third current sensor A3, a third current sensor A4 and a second voltage sensor V2, wherein a measuring end of the third current sensor A3 is connected to four nodes of the first array group, and another measuring end of the third current sensor A3 is connected to four nodes of the second array group; a measuring end of another fourth current sensor A4 is connected to one of the nodes of the first array group, and another measuring end is connected to one of the nodes of the second array group; two ends of the second voltage sensor V2 are connected between any two nodes in the first array group or the second array group;

[0025] The first submodule and the second submodule are in series connection. When there are multiple submodules, the submodules are all connected in series. In this embodiment, two submodules are used as an example, but it is not limited to two.

[0026] The diagnostic topology is simulated by computer. In this embodiment, MATLAB is used for simulation. The measured values ​​of the current sensors and voltage sensors in the diagnostic topology under different fault types and the polynomial transformation of the measured values ​​are simulated over time. For example, the measured value I1 of the first current sensor A1 is changed over time. The change curve here can also be a straight line with constant value. The change curves with prominent changes corresponding to different faults and the current sensors, voltage sensors or / and polynomial transformations corresponding to the change curves are found. It should be noted that there are multiple change curves corresponding to one fault. For all the current sensors and voltage sensors in the diagnostic topology, some have a large impact and some have a small impact. It is necessary to find the current sensors, voltage sensors or / and polynomial transformations corresponding to the change curves with large impacts. The polynomial transformation here refers to the difference and comparison between the measured values ​​of the first current sensor A1, the second current sensor A2, the first voltage sensor V1, the third current sensor A3, the fourth current sensor A4 and the second voltage sensor V2 to expand them into multiple features.

[0027] Through the above process, it can be simulated that for different faults, multiple corresponding change curves and features corresponding to the change curves can be obtained. The feature here refers to which current sensor or voltage sensor or polynomial transformation it belongs to.

[0028] The importance analysis of the measured values ​​of current sensors and voltage sensors under multiple faults and the polynomial transformation of the measured values ​​over time is performed to find out the important features in fault classification;

[0029] The important feature here refers to the feature that contributes to all faults when a fault occurs, and has different contribution forms for different faults. For example: for fault A and fault B, the change curves corresponding to the first current sensor A1 have changed, and the changes are significant, and for fault A and fault B, the change curve of the current sensor A1 can distinguish whether it belongs to fault A or fault B. Then the first current sensor A1 can be used as an important feature. This important feature can be a single feature or a combination of multiple features.

[0030] The ET algorithm is trained by taking important characteristics as input and different fault types corresponding to the change curves of important characteristics as output. During the training process, the particle swarm algorithm is used to adjust the parameters of the ET algorithm to obtain the optimal classification and positioning model. The parameters of the ET algorithm adjusted by the particle swarm algorithm include the number of decision trees, the maximum depth of the decision tree, the minimum number of samples required for node splitting, and the minimum number of samples required for leaf nodes.

[0031] The measurement values ​​collected by the sensors in the diagnosis process are transformed into polynomials, and important features are obtained. The important features are input into the optimal classification and positioning model for fault classification and positioning.

[0032] In another embodiment, before performing the polynomial transformation, all features need to be pre-judged to determine whether the changed measurement value belongs to the measurement value of the sensor in the first submodule or the measurement value of the sensor in the second submodule.

[0033] This can further narrow the scope of the fault.

[0034] The pre-judgment also includes that when none of the measured values ​​in the judgment submodule has changed, it is considered that no fault has occurred.

[0035] The pre-judgment also includes the number of sudden changes in the measured values ​​of the sensors in the judgment submodule. When the number is equal to 1, it is regarded as a sensor failure. When the number is greater than or equal to 2, it is regarded as a battery failure.

[0036] The method of the present invention can extract the measured values ​​collected by the sensor and the polynomial transformation of the measured values ​​as features, and analyze the differences between normal state, internal fault, and open circuit fault. By comparing the change curves under different fault states, the abnormal changes during the fault can be found. Thus, the location of the fault can be accurately located, thereby achieving effective fault diagnosis.

[0037] In this embodiment, the measured values ​​collected by the current sensor and the voltage sensor are represented by A 1 , A 2 , A 3 , A 4 、V 1 and V 2In order to reduce the computational complexity in fault detection, the collected features are transformed into polynomials, and then the importance of the measured values ​​of the current sensor and voltage sensor under multiple faults and the polynomial transformation of the measured values ​​over time are analyzed through simulation to find out the important features in fault classification. 1 ”, “A 1 -A 3 ”, “A 3 -A 4 ”, “V 1 -V 2 ” and “A 1 / A 2 " is used as an important feature for multiple fault detection. By selecting important features and performing feature processing, it plays a role in reducing unnecessary calculations and redundant information, thereby improving the efficiency and accuracy of fault diagnosis. Therefore, in the classification process of the ET algorithm, targeted processing and selection can be performed according to the importance of the features, thereby optimizing the performance of the fault detection system.

[0038] In this embodiment, when an ISC fault (internal short circuit) or a COC fault (cell open circuit) occurs in the first submodule, A 1 , A 2 and V 1 will change simultaneously. Therefore, when only one sensor reading shows a change, this sensor may be faulty. But when all these sensor readings change, it indicates a fault in the first submodule. First, the extracted features are processed by polynomial transformation to obtain “V 1 ”, “A 1 -A 3 ”, “A 3 -A 4 ”, “V 1 -V 2 ” and “A 1 / A 2 "Four important features. Then, these extracted important features and other features are used as input, and the different fault types corresponding to the change curves of the important features are used as output to train the ET algorithm. During the training process, the particle swarm algorithm is used to adjust the parameters of the ET algorithm to obtain the optimal classification and positioning model; it should be noted that when the readings of the three sensors do not change significantly, it means that there is no fault in normal operation.

[0039] The process of adjusting the parameters of the ET algorithm using the particle swarm algorithm is as follows: first, all particles in the space are assigned an initial random position and an initial random velocity. Then, each particle moves forward according to its velocity, the known best global position in the problem space, and the known best position of the particle. As the calculation proceeds, the particles gather or condense around one or more optimal points by exploring and exploiting known favorable positions in the search space.

[0040] The formula used in the ET algorithm planning node is:

[0041]

[0042] Where P mk is the percentage of observations of type k in node m, and H is the Gini coefficient used to divide the nodes of the ET algorithm;

[0043] The formula for adjusting the parameters of the ET algorithm using the particle swarm algorithm is:

[0044]

[0045] in and is the velocity and position of particle i in the kth generation, pbest is the extreme value of a single particle, gbest is the extreme value of all particles, r 1 and r 2 is a random number between [0, 1]; d represents the dimension; c 1 and c 2 is the acceleration coefficient.

[0046] Features need to be normalized when they are input, and the following normalization formula is used:

[0047]

[0048] Among them, x ′ mj is the result of feature normalization, x mj is the i-th item of data of the j-th feature of the battery, is the average value of the jth feature data, S j is the standard deviation of the j-th feature data.

[0049] The method of the present invention can well diagnose ISC faults, COC faults and sensor faults of lithium battery packs and their corresponding locations. The fault diagnosis mechanism can effectively ensure the safety and stability of lithium battery packs. It can help operators to find and handle faults in time, reduce the impact on the entire system, and improve the reliability and life of the battery pack.

[0050] In order to prove that the method of the present invention has a very good performance in multi-fault detection, a test was conducted. The method of the present invention used a data set under NEDC conditions (a standard test process used by European countries to evaluate automobile fuel consumption and emissions. The NEDC test is an experimental method based on a simulated road driving cycle, which covers two different driving modes: low-speed urban driving and highway driving). The trained model performed very well in the verification data set of NEDC conditions and the constant current discharge data set, and the result accuracy was 98% and 100%, respectively.

[0051] In order to verify that the method of the present invention has good generalization performance, three data sets were established under three working conditions: WLTP3 (Europe has launched a new WLTP condition in September 2017. The WLTP condition was jointly developed by experts from Japan, the United States, Europe and other countries. Its characteristic is that real driving condition data is collected worldwide), ARL (urban conditions in Common Artemis DrivingCycles (CADC) conditions, where CADC refers to the statistical analysis of a large database of real-world driving patterns in Europe in the European Atemis project) and AU (rural conditions in CADC conditions), and then the trained data sets were tested. The test results show that the fault diagnosis accuracy under the three working condition data sets of WLTP3, ARL and AU is 94%, 95% and 96% respectively. It proves that the method of the present invention has good fault detection effect under data sets of different working conditions.

[0052] 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 and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A lithium battery pack multi-fault diagnosis method, characterized in that: The method includes: Establish a diagnostic topology structure, the diagnostic topology structure includes N submodules with the same structure, the submodule includes 16 lithium iron phosphate batteries, a first current sensor, a second current sensor and a voltage sensor, two lithium iron phosphate batteries are connected in series to form a series group, the connection end between the two lithium iron phosphate batteries forms a node, all the series groups are connected in parallel, four series groups form a first array group, and the other four series groups form a second array group, one measuring end of the first current sensor is connected to four nodes of the first array group, and the other measuring end of the first current sensor is connected to four nodes of the second array group; one measuring end of another second current sensor is connected to one of the nodes of the first array group, and the other measuring end is connected to one of the nodes of the second array group; the two ends of the voltage sensor are connected between any two nodes in the first array group or the second array group; N submodules are connected in series; The diagnostic topology is simulated by a computer, and the curves of the measured values ​​of the current sensor and the voltage sensor and the polynomial transformation of the measured values ​​over time under different fault types are simulated to find the prominent change curves corresponding to different faults and the current sensor, voltage sensor or / and polynomial transformation corresponding to the change curves; the importance analysis of the curves of the measured values ​​of the current sensor and the voltage sensor and the polynomial transformation of the measured values ​​over time under multiple faults is performed to find the important features in the fault classification; The important characteristics are used as input, and the different fault types corresponding to the change curves of the important characteristics are used as output to train the ET algorithm. During the training process, the particle swarm algorithm is used to adjust the parameters of the ET algorithm to obtain the optimal classification and positioning model. The measurement values ​​collected by the sensors in the diagnosis process are transformed into polynomials, and important features are obtained. The important features are input into the optimal classification and positioning model for fault classification and positioning.

2. The lithium battery pack multi-fault diagnosis method according to claim 1, characterized in that: The polynomial transformation refers to expanding the measured values ​​of the current sensor and the voltage sensor into multiple features by making differences and ratios between each other.

3. The lithium battery pack multi-fault diagnosis method according to claim 1, characterized in that: The important features refer to features that contribute to all faults when a fault occurs and have different contribution forms for different faults.

4. The lithium battery pack multi-fault diagnosis method according to claim 1, characterized in that: Before the measurement values ​​collected by the sensors in the diagnosis process are transformed by polynomials, a pre-judgment is performed, wherein the pre-judgment includes determining the submodule in which the sensor measurement value of the changed measurement value belongs.

5. The lithium battery pack multi-fault diagnosis method according to claim 4, characterized in that: The pre-judgment also includes determining that no fault has occurred when none of the measured values ​​in the judgment submodule has changed.

6. The lithium battery pack multi-fault diagnosis method according to claim 4, characterized in that: The pre-judgment also includes judging the number of mutations in the measured values ​​of the sensors in the submodule. When the number is equal to 1, it is regarded as a sensor failure. When the number is greater than or equal to 2, it is regarded as a battery failure.

7. The lithium battery pack multi-fault diagnosis method according to claim 4, characterized in that: The parameters of the ET algorithm adjusted by the particle swarm algorithm include the number of decision trees, the maximum depth of the decision tree, the minimum number of samples required for node splitting, and the minimum number of samples required for leaf nodes.

Images