Intelligent rapid testing method and system for power battery life based on cloud-edge collaboration
Through cloud-edge collaboration technology, battery capacity degradation prediction is used using edge-end data and cloud-end models to solve the problem of time-consuming and cost-effectiveness of existing test methods, and realizes intelligent and rapid evaluation of battery life.
Patent Information
- Application Number
- CN202210669341.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-06-14
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2042-06-14
AI Technical Summary
The existing lithium-ion battery cycle life test methods are time-consuming and costly, and require dismantling of batteries, which limits the rapid development of the new energy vehicle industry.
Using the intelligent and rapid test method of power battery life based on cloud-edge collaboration, the battery capacity data is obtained and uploaded to the cloud by performing cyclic charging and discharging tests at the edge. The prediction model is used for retraining, and the battery capacity degradation trend and remaining life prediction are achieved. .
It realizes fast and accurate evaluation of battery life, reduces data transmission delay, enhances data security, avoids battery disassembly, shortens test time, and reduces costs.
Smart Images

Figure CN114879050B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power battery life testing and prediction, and in particular to a method and system for intelligent and rapid testing of power battery life based on cloud-edge collaboration. Background Art
[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.
[0003] Battery life refers to the number of charge and discharge cycles until the available capacity drops to 80% of the initial value. Cycle life testing of batteries can further understand battery characteristics, prove whether the battery meets the design goals, achieve better management and control during battery use, and evaluate the battery's ability to meet the needs of different application scenarios.
[0004] At present, the cycle life test of lithium-ion batteries mainly adopts the national standard GB / T 31484-2015 "Requirements and test methods for cycle life of power batteries for electric vehicles". The standard stipulates that when conducting a standard battery cycle life test, the battery must be continuously charged and discharged until its end-of-life condition, that is, "test to the end". This test is usually very time-consuming. The number of charge and discharge cycles of lithium-ion batteries can reach thousands or even more than 10,000 times, and the test time can reach 1 year or even longer. The battery cycle life requires full-cycle, long-term, and high-cost testing. The new energy vehicle industry requires power batteries to be able to quickly iterate technology and upgrade products, but the lack of rapid test methods for the cycle life of existing batteries has seriously restricted the rapid and high-quality development of the new energy vehicle industry. Therefore, how to shorten the life test time and quickly, accurately and efficiently test and evaluate the battery cycle life has become an important means to break through the key technical bottlenecks in the rapid development of power batteries and related industries.
[0005] The rapid test method of battery life based on prediction is an excellent solution to the above problems, namely "combination of measurement and estimation". By replacing testing with prediction, a lot of time can be saved, and efficient and accurate evaluation of battery cycle life can be achieved. Many scholars have carried out a lot of research. For example, the prior art tests the battery for short-term cycle performance at different cycle times, records the cycle times and capacity retention rate, and then disassembles the battery after different cycle times, uses the XRD method to test the graphitization degree of the graphite negative electrode material, and predicts based on the three data of cycle times, capacity retention rate and graphitization degree. This method requires disassembly and destruction of the battery, and has certain limitations. The prior art discloses a method for predicting the cycle life of a lithium-ion battery, installs a pressure sensor on the surface of the battery, records the capacity information of the battery within a certain number of cycles, and fits the cycle times, discharge capacity and voltage data in the pressure sensor to predict the battery life. This method requires an additional pressure sensor, which will increase the cost in practice and has certain limitations. The prior art also discloses a method for rapid prediction of lithium-ion cycle life, which predicts the battery life by placing the battery under different rates for 500 cycle tests and fitting the equation to obtain a prediction equation. This method does not require sophisticated testing equipment and complex theoretical calculations, but it still requires a long period of cyclic testing and is not very practical.
[0006] The prediction based on data-driven does not need to disassemble the battery or consider the physical and chemical changes inside the battery. It directly uses the test data as the basis for the corresponding analysis and prediction of the battery life. The method is simple and fast. For example, the prior art discloses a data-driven method for predicting the remaining life of the battery. By recording the charging time, temperature, charging times and other data, sampling from these data as the training data of the neural network model, and predicting through a self-set data encapsulator. The prior art discloses a life prediction method based on data-driven and battery characteristics. Through the operation data of a large number of different batteries, characteristic factors are extracted and the SOH under different states is calculated. The life prediction is performed by establishing a battery and user portrait algorithm relationship model. The prior art also discloses a power battery life prediction method based on big data. By analyzing the parameter big data of the battery under different discharge depths, different temperatures, different discharge rates and other conditions, the big data analysis relationship is obtained, and the battery life is predicted using the throughput method. Although the above method can obtain the cycle life of the battery, the amount of data used is large, the implementation speed is slow, and the cost is high.
[0007] Although there are many studies on life prediction, there is still a lack of intelligent and rapid testing methods for battery life based on prediction. Summary of the invention
[0008] In order to solve the problems of traditional testing methods with long cycles, high costs, and the need to disassemble batteries, the present invention proposes a method and system for intelligent and rapid testing of power battery life based on cloud-edge collaboration, combining cloud-edge collaboration technology, and "measurement-estimation combination" technology, to achieve intelligent and rapid testing of power battery life by replacing testing with prediction. The task of training the basic model for life prediction is concentrated on the cloud side, and the life prediction task is performed on the edge side. Using only partial capacity data, the degraded capacity and remaining service life of the battery to be tested can be predicted, ensuring the needs of intelligent and rapid testing of power battery life.
[0009] In some embodiments, the following technical solutions are adopted:
[0010] An intelligent and rapid testing method for power battery life based on cloud-edge collaboration, comprising:
[0011] Perform a cyclic charge and discharge test on the power battery to be tested at the edge to obtain the battery capacity test time series data within the set time;
[0012] Uploading the battery capacity test time series data to the cloud as training data to retrain the initially trained power battery life prediction basic model;
[0013] Based on the battery capacity test time series data within the set time and the re-trained power battery life prediction basic model, capacity prediction is performed at the edge to obtain the overall capacity degradation trend and remaining life of the power battery to be tested.
[0014] In other embodiments, the following technical solutions are adopted:
[0015] An intelligent rapid testing system for power battery life based on cloud-edge collaboration, including:
[0016] The data acquisition module is used to perform a cycle charge and discharge test on the power battery to be tested at the edge end and obtain the battery capacity test time series data within a set time;
[0017] A model training module is used to upload the battery capacity test time series data to the cloud as training data, and to retrain the initially trained power battery life prediction basic model;
[0018] The capacity degradation trend prediction module is used to perform capacity prediction at the edge based on the battery capacity test time series data within the set time and the retrained power battery life prediction basic model to obtain the overall capacity degradation trend and remaining life of the power battery to be tested.
[0019] In other embodiments, the following technical solutions are adopted:
[0020] An intelligent rapid testing system for power battery life based on cloud-edge collaboration, including:
[0021] The edge device is used to perform a cycle charge and discharge test on the power battery to be tested at the edge, and obtain the battery capacity test time series data within a set time;
[0022] In addition, based on the battery capacity test time series data within the set time and the retrained power battery life prediction basic model, capacity prediction is performed at the edge to obtain the overall capacity degradation trend and remaining life of the power battery to be tested;
[0023] The cloud device is used to train the basic model for predicting the life of the power battery based on the existing battery capacity data and the battery capacity test time series data received from the edge.
[0024] Compared with the prior art, the present invention has the following beneficial effects:
[0025] (1) The present invention realizes intelligent and rapid testing of the life of power batteries based on a prediction method, replaces testing with prediction, combines measurement and estimation, saves a lot of testing time, and can achieve efficient and accurate evaluation of the battery cycle life.
[0026] (2) Based on a large amount of existing battery data, the present invention concentrates the task of training the basic life prediction model on the cloud and performs the life prediction task on the edge side; it rationally combines cloud computing and edge computing to reduce data transmission delay and enhance data security.
[0027] (3) The present invention only tests partial data at the initial stage. By using the partial test data, the remaining service life of the battery can be predicted without disassembling the power battery, thereby shortening the test time and realizing intelligent and rapid detection of the power battery.
[0028] Other features and advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1 This is a flow chart of a method for intelligent and rapid testing of power battery life based on cloud-edge collaboration in an embodiment of the present invention;
[0030] Figure 2 It is the intelligent and rapid testing system architecture of power battery life based on cloud-edge collaboration in an embodiment of the present invention;
[0031] Figure 3 Schematic diagram of the LSTM structure in an embodiment of the present invention;
[0032] Figure 4is the capacity degradation curve of four batteries in the embodiment of the present invention;
[0033] Figure 5 A schematic diagram of generating a Sobol sequence sample in an embodiment of the present invention;
[0034] Figure 6 is a degradation capacity curve expanded in an embodiment of the present invention;
[0035] Figure 7 This is the prediction result of using 8 groups of battery data to train the basic model in the embodiment of the present invention. DETAILED DESCRIPTION
[0036] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.
[0037] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.
[0038] Embodiment 1
[0039] The essence of cloud computing is to provide super strong and efficient computing power based on a large number of concentrated computing resources, implement parallel computing, distributed computing and load balancing, etc. Its high degree of concentration can reduce the unit computing power cost. The cloud servers used in cloud computing have a large number of CPUs and GPUs, which can accept battery test data collected by local users, extract massive cycle life information and evaluate the continuity of its use. However, as the curtain of the era of power battery big data is opened, the data transmission bandwidth and power, the storage space and computing resources of cloud servers will increase dramatically. The data that cloud servers can store and process is limited, and cannot support long-term large-capacity, high-speed data upload.
[0040] Edge computing is a distributed open platform that integrates network, computing, storage and other capabilities at the edge of the network. In edge computing, battery life test data can be directly stored and calculated on the user side instead of having to be uploaded to the cloud. This greatly reduces data transmission time and cost, and provides better support for high-bandwidth and low-latency services compared to the centralized deployment model of traditional cloud computing.
[0041] Cloud computing and edge computing have their own strengths. Cloud computing is good at long-term, non-real-time and global big data processing and analysis, while edge computing can play its strengths in the near-real-time intelligent processing of user-side services. Reasonable combination of cloud computing and edge computing can realize intelligent and rapid testing of power battery life.
[0042] Based on this, in one or more embodiments, a method for intelligent rapid testing of power battery life based on cloud-edge collaboration is disclosed, combined with Figure 1 , specifically including the following process:
[0043] (1) Perform a cyclic charge and discharge test on the power battery to be tested at the edge to obtain the battery capacity test time series data within a set time;
[0044] The method of this embodiment only needs to collect partial data of the tested power battery (approximately the first 10% of the battery capacity data) to complete the overall battery capacity degradation trend and remaining life prediction.
[0045] (2) uploading the battery capacity test time series data to the cloud as training data to retrain the initially trained power battery life prediction basic model;
[0046] In this embodiment, the basic model for predicting the life of a power battery adopts an LSTM model. LSTM introduces a "memory" control gate to record historical information and control the degree of influence of current information on historical information, effectively solving the problems of long-term dependence and gradient disappearance of traditional RNN neural networks. LSTM structurally consists of three parts: input gate, forget gate, and output gate. "Gate" is a unique feature of LSTM, which has selective permeability to information. The three "gates" work together to efficiently extract and use historical information for prediction. The basic structure of the network is as follows: Figure 3 As shown; where x t 、h t-1 and C t-1 are the input at time t, the output at time t-1, and the cell state output at time t-1 respectively; σ is the sigmoid activation function, and tanh is the hyperbolic tangent activation function.
[0047] With t as the cycle period, the forget gate of LSTM can be expressed by Formula 1.
[0048] f t =sigmoid(W f ·[h t-1 ,x t ]+b f ) (1)
[0049] Among them, W f and b fare the weight matrix and bias vector of the forget gate respectively, [h t-1 ,x t ] represents the input vector. W f The dimension of h t-1 and x t It can be expressed by formula (2).
[0050]
[0051] Similarly, the input gate can be expressed as formula (3).
[0052] i t =sigmoid(W i ·[h t-1 ,x t ]+b i ) (3)
[0053] Among them, W i and b i are the weight matrix and bias vector of the input gate respectively. The unit state of the current input is calculated by the output of the previous moment and the input of this moment, as shown in formula (4).
[0054] C′ t =tanh(W c ·[h t-1 ,x t ]+b c ) (4)
[0055] Among them, W c and b c are the weight matrix and bias vector of the current input unit state. Further, through the unit state C t-1 and the current input cell state to calculate the current cell state. Formula (5) gives C t The calculation formula for .
[0056] C t =f t ⊙C t-1 +i t ⊙C′ t (5)
[0057] New cell status C t By the current unit C' t and long-term memory unit C t-1 The two units are controlled by the forget gate and the input gate. The forget gate has the ability to save previous information, and the input gate can prevent the current irrelevant content from affecting the memory process. The output of LSTM is determined by the output gate and the update state unit.
[0058]
[0059] Among them, W o and b o are the weight matrix and bias vector of the output gate respectively.
[0060] In this embodiment, the input of the LSTM neural network model is the power battery capacity data of the previous N moments, and the output is the power battery capacity data of the N+1th moment.
[0061] The training process of the LSTM neural network model is carried out in the cloud. First, a large amount of existing capacity data of the same type of power batteries at different times is used to build a model training data set to conduct preliminary training on the basic model for power battery life prediction. Then, the battery capacity test time series data within a set time uploaded by the edge end is used to further train the preliminarily trained model.
[0062] (3) Based on the battery capacity test time series data within the set time and the retrained power battery life prediction basic model, capacity prediction is performed at the edge to obtain the overall capacity degradation trend and remaining life of the power battery to be tested.
[0063] Specifically, the edge obtains the LSTM neural network model trained in the cloud, and inputs the battery capacity test time series data within a set time into the LSTM neural network model, so as to obtain the overall capacity degradation trend of the power battery to be tested.
[0064] Combination Figure 2 The overall architecture for implementing the method of this embodiment consists of the edge and the cloud. The edge mainly includes battery testing and edge computing equipment, which are distributed in power battery stacks of different models and materials. They are responsible for collecting, processing, and uploading a small amount of battery capacity degradation and cycle life data, and sending the battery cycle life and capacity degradation results, battery testing equipment status information, etc. to the cloud. The cloud is mainly responsible for receiving historical data from the edge and storing a large amount of known battery cycle life data, training the basic model for battery life prediction, and visualizing battery life prediction results.
[0065] One cloud can deploy multiple power battery stack edge terminals. In this article, due to limited conditions, the cloud is simulated by a Dell tower workstation, and its basic configuration is as follows: Intel Xeon W-2245 8C 3.9GHz DDR4-2933 CPU; NVIDIA RTX6000 graphics card. The edge terminal is simulated by Raspberry Pi, which includes expansion modules such as Bluetooth and WIFI, and a 64GB memory card is attached for storing programs and backing up data. Its operating system is a customized system based on Linux, which is a newer version and has more complete support for the new Raspberry Pi. Raspberry Pi obtains battery test information collected by sensors through the serial port. In order to ensure the possible Bluetooth expansion requirements of the system, the serial port is remapped to mini UART.
[0066] The cloud and edge devices communicate based on the FTP protocol. The multi-threaded processing capability of the cloud workstation enables it to connect to multiple edge devices simultaneously. Within the coverage of the communication network, after the edge device sends an application to the cloud, it is assigned to the designated port of the workstation, and the communication process is supervised by a unique independent thread. The communication content is written to the workstation storage in .csv format. The cloud preprocessing and capacity feature extraction program will synchronously read the above files in sequence to train the basic model for power battery life prediction. The file size generated by the cloud in communication with edge devices is strictly limited. When the file size exceeds the upper limit of the workstation memory, the early files will be cleared to ensure sufficient available space on the cloud.
[0067] The architecture of this embodiment can realize the interactive communication between the uplink data of the edge and the downlink data of the cloud. First, at the edge device, use the Arbin battery testing device to perform the power battery cycle charge and discharge test, and upload the battery capacity test history data to the cloud. Secondly, on the cloud workstation, pre-process the battery test data (data cleaning, data elimination, etc.) and the existing massive battery capacity and life data of different models to build and train the basic model for power battery life prediction. Finally, on the edge Raspberry Pi, download the cloud-trained model, and use the first 10% of the data to quickly predict the power battery life.
[0068] As an optional implementation, based on the existing capacity data of the same type of power batteries at different times, the model training data set is expanded, specifically:
[0069] C gen =a×C1+b×C2+...+c×C N
[0070] Among them, C1-C N are the degradation capacity data of different power batteries respectively; a, b, c are the corresponding capacity coefficients respectively; among which, the capacity coefficients are selected from the samples generated based on the Sobol sequence.
[0071] The Sobol sequence is introduced as follows:
[0072] For all algorithms that require sampling, uniformly distributed random numbers mean better sample distribution. For optimization problems with unknown solution distribution, the initial values of individuals should be distributed as evenly as possible in the data space, so as to maintain the diversity of the population at a high level. Therefore, the new generator is not required to simulate a true uniform distribution, but strives to ensure that samples of any size (especially small samples) can meet low variance.
[0073] All random number generation algorithms based on modern CPUs are quasi-random. They are limited to a period. When the period is exceeded, they will repeat and no longer be independent random numbers. The ultimate limit of this period is determined by the number of bits in the computer, so no built-in random number is "truly" random.
[0074] Sobol sampling uses a different approach to sampling. Compared to random numbers, Sobol sequences focus on generating a uniform distribution in the probability space. However, this is not simply filling the grid, but using an essentially random but clever method to "fill" the probability space, that is, the random numbers generated later will be distributed to areas that have not been sampled before. The following is the generation principle of Sobol.
[0075] (1) Radical Inversion Operation
[0076]
[0077] Φ b,C (i)=(b -1 ...b -M )[C(a0(i)...a M-1 (i)) T ] (8)
[0078] In the above formula, b is a positive integer. If we express any integer i as a number in base b, and then replace the obtained number by position a l (i) Arrange them into a vector, then multiply this vector with the generator matrix C to get a new vector, and finally mirror the new vector to the right of the decimal point to get another number in the range [0,1). This is called the Radical Inversion operation, denoted by Φ b,C (i).
[0079] If we simplify the above process and let C be the identity matrix, we can get the Van der Corput sequence
[0080]
[0081] (2) Generate Sobol sequence
[0082] Each dimension of the Sobol sequence is composed of a radical inversion with a base of 2, but each dimension of the radical inversion has its own generator matrix C. That is,
[0083]
[0084] Because the base is 2, the generation of Sobol sequence can directly use bit operations to implement radical inversion, which is very efficient. The distribution of Sobol sequence is not only uniform, but also when the number of samples is an integer power of 2, in [0,1) 8 There is one and only one point in each basic interval with base 2 in the interval, which means that samples with high-quality distribution can be generated. The generator matrix C can be obtained by searching on relevant websites. In this embodiment, the toolbox provided by Matlab is used to generate the Sobol sequence. Figure 5 Schematic diagram of the generated Sobol sequence sample.
[0085] This example uses the NASA dataset as an example to verify the method of this example:
[0086] The NASA data set comes from the NASA Ames Center of Excellence in Prediction. The experiment uses 18650 lithium-ion batteries and conducts charge and discharge experiments under different operating conditions. Impedance is used as the damage standard. The experiment captures a large amount of capacity degradation and life data under different operating conditions of the battery. Take batteries 5, 6, 7, and 18 in the data set as an example, and draw their capacity degradation curves, such as Figure 4 As shown. Figure 4 It can be seen that the four batteries show different lifespan distributions, but the overall trend of capacity decay is similar.
[0087] Since the data of four batteries is relatively small, under limited experimental conditions, this embodiment performs data expansion and then inputs it into the cloud to train the basic model for life prediction.
[0088] The data expansion method of this embodiment is as follows:
[0089] C gen =a×C B0005 +b×C B0006 +c×C B0007 (11)
[0090] Among them, C gen To expand the uncertainty of battery degradation capacity data, C B0005~B0007is the degradation capacity data of No. 5 to No. 7 batteries, and a, b, and c are the capacity coefficients. In order to ensure that enough battery data can be expanded and have strong randomness, and because the original batteries are from the same batch of batteries, in order to ensure that the capacity degradation data after expansion is roughly similar to the original batteries, the Sobol sequence is used to generate uniformly distributed samples within [-1,1]. Parameters a and b are randomly selected from this sample population, c = 1-(a+b), then formula (11) becomes
[0091] C gen =a×C B0005 +b×C B0006 +(1-ab)×C B0007 (12)
[0092] When a and b belong to [-1, 1], there are four possible values. Table 1 gives the range and trade-off of the parameters.
[0093] Table 1 Parameter range and trade-offs
[0094]
[0095] From the above table, we can see that when the range of a and b is [-1,1], there are four cases for the value range of c. When c∈[-1,0] or [0,1], it meets the requirements; in the other two cases, the values of c are greater than 1 and are discarded if they do not meet the requirements. Five sets of data are randomly selected from the parameter set that meets the requirements as the capacity coefficient, as shown in Table 2. The degradation capacity curve of the expanded battery is as follows: Figure 6 shown.
[0096] Table 2 Capacity coefficients of 5 groups
[0097]
[0098] In this embodiment, the full capacity data of 8 batteries are selected for training the cloud-based model, of which 3 are batteries B0005, B0006 and B0007, and the full data of the other 5 batteries are the 5 groups of battery data C obtained by expanding formula (11) gen (i), i = 1, 2, ... 5, the time step is set to 3, that is, the capacity value of the previous 3 moments is used to predict the next moment. The data of these 8 batteries are input into the cloud to train the LSTM life prediction basic model. After the model training is completed, it is sent to the edge Raspberry Pi, and the life prediction is performed using about the first 10% of the data of B0018.
[0099] The prediction results are as follows Figure 7 As shown. Figure 7 It can be seen that after using 8 sets of data for basic model training, accurate and fast prediction of battery life can be achieved with only about 10% of the data, and the testing time is greatly reduced.
[0100] Table 3 Comparison of prediction indicators
[0101]
[0102] Table 3 gives a comparison of prediction indicators. As can be seen from Table 3, when using 1 set of data to train the basic model, due to the small amount of data, the relevant information of battery capacity degradation cannot be obtained by LSTM network training, so the error is large, and the root mean square error (RMSE) and mean absolute error (MAE) are 0.062 and 0.049 respectively. At the same time, the RUL prediction value is 85, which is 12 cycles different from the actual value, and the error is about 12%. When using 3 sets of data to train the basic model, the amount of data has been expanded to a certain extent, so the error has decreased to a certain extent, RMSE and MAE dropped to 0.039 and 0.030 respectively, and the RUL prediction error dropped to 4.1%. When using 8 sets of data to train the model, the prediction error is further reduced. RMSE and MAE dropped to 0.034 and 0.021, compared with the training with 1 set of data, a decrease of 45% and 57.1%. The RUL prediction error dropped to 2.0%, a decrease of 83.7%.
[0103] Table 4 Comparison of battery life test time
[0104]
[0105] Table 4 is a comparison of life test time. It can be seen from Table 4 that the battery test time of traditional technology is more than 300 days. This technology uses cloud-based training models and only uses about 10% of the data to achieve fast and accurate predictions. The test time is less than 15 days, which greatly shortens the time.
[0106] Embodiment 2
[0107] In one or more embodiments, a power battery life intelligent rapid testing system based on cloud-edge collaboration is disclosed, specifically including:
[0108] The data acquisition module is used to perform a cycle charge and discharge test on the power battery to be tested at the edge end and obtain the battery capacity test time series data within a set time;
[0109] A model training module is used to upload the battery capacity test time series data to the cloud as training data, and to retrain the initially trained power battery life prediction basic model;
[0110] The capacity degradation trend prediction module is used to perform capacity prediction at the edge based on the battery capacity test time series data within the set time and the retrained power battery life prediction basic model to obtain the overall capacity degradation trend and remaining life of the power battery to be tested.
[0111] The specific implementation methods of the above modules have been described in detail in Example 1 and will not be described in detail here.
[0112] Embodiment 3
[0113] An intelligent rapid testing system for power battery life based on cloud-edge collaboration, characterized by comprising:
[0114] The edge device is used to perform a cycle charge and discharge test on the power battery to be tested at the edge, and obtain the battery capacity test time series data within a set time;
[0115] In addition, based on the battery capacity test time series data within the set time and the retrained power battery life prediction basic model, capacity prediction is performed at the edge to obtain the overall capacity degradation trend and remaining life of the power battery to be tested;
[0116] The cloud device is used to train the basic model for predicting the life of the power battery based on the existing battery capacity data and the battery capacity test time series data received from the edge.
[0117] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.
Claims
1. An intelligent and rapid testing method for power battery life based on cloud-edge collaboration, comprising: Perform a cyclic charge and discharge test on the power battery to be tested at the edge to obtain the battery capacity test time series data within the set time; Uploading the battery capacity test time series data to the cloud as training data to retrain the initially trained power battery life prediction basic model; Based on the battery capacity test time series data within the set time and the retrained power battery life prediction basic model, capacity prediction is performed on the edge to obtain the overall capacity degradation trend and remaining life of the power battery to be tested; The power battery life prediction basic model adopts an LSTM neural network model; the input of the LSTM neural network model is the power battery capacity data of the previous N moments, and the output is the power battery capacity data of the N+1th moment; The training process of the LSTM neural network model is carried out in the cloud. First, a large amount of existing capacity data of the same type of power batteries at different times is used to construct a model training data set to preliminarily train the basic model for power battery life prediction; then, the battery capacity test time series data within a set time uploaded by the edge end is received to further train the preliminarily trained model; The LSTM neural network model includes an input gate, a forget gate and an output gate, specifically including: The expression of the forget gate is: f t =sigmoid(W f ·[h t-1 ,x t ]+b f ); Among them, x t is the input at time t, h t-1 is the output at time t-1, W f and b f are the weight matrix and bias vector of the forget gate respectively, [h t-1 ,x t ] represents the input vector, W f The dimension of h t-1 and x t The expression is: The input gate expression is: i t =sigmoid(W i ·[h t-1 ,x t ]+b i ); Among them, W i and b i are the weight matrix and bias vector of the input gate respectively; the unit state of the current input is calculated by the output of the previous moment and the input of this moment, and the expression is: C' t =tanh(W c ·[h t-1 ,x t ]+b c ); Among them, W c and b c are the weight matrix and bias vector of the current input unit state, tanh is the hyperbolic tangent activation function; through the unit state C t-1 The cell state at the current moment is calculated based on the cell state of the current input. The expression is: C t =f t ⊙C t-1 +i t ⊙C’ t ; The output of LSTM is determined by the output gate and the update state unit, and the expression is: Among them, W o and b o They are the weight matrix and bias vector of the output gate respectively; Based on the existing capacity data of multiple power batteries of the same type at different times, the model training data set is expanded. Specifically, the degradation capacity data of different power batteries are multiplied by the corresponding capacity coefficients. The capacity coefficients are selected from the samples generated based on the Sobol sequence. The Sobol sequence is used to generate uniformly distributed samples in [-1,1], and the parameters a and b are randomly selected from this sample population. The data expansion method is: C gen =a×C B0005 +b×C B0006 +c×C B0007 Among them, C gen To expand the uncertainty battery degradation capacity data, C B0005~B0007 is the degradation capacity data of batteries No. 5 to No. 7, a, b, c are capacity coefficients respectively; the Sobol sequence is used to generate uniformly distributed samples in [-1,1], parameters a and b are randomly selected in this sample population, c = 1-(a+b).
2. The intelligent and rapid testing method for power battery life based on cloud-edge collaboration as claimed in claim 1 is characterized in that: The cloud and edge devices communicate with each other based on the FTP protocol; the cloud can synchronously connect to multiple edge devices.
3. The intelligent and rapid testing method for power battery life based on cloud-edge collaboration as claimed in claim 2 is characterized in that: After the edge device sends a request to the cloud, it is assigned to the designated port of the workstation, and the communication process is supervised by a unique independent thread.
4. The intelligent and rapid testing method for power battery life based on cloud-edge collaboration as claimed in claim 1 is characterized in that: When the file size generated during communication between the cloud and edge devices exceeds the workstation memory limit, the workstation's early files are cleared.
5. An intelligent rapid testing system for power battery life based on cloud-edge collaboration, characterized in that: include: The data acquisition module is used to perform a cycle charge and discharge test on the power battery to be tested at the edge end and obtain the battery capacity test time series data within a set time; A model training module is used to upload the battery capacity test time series data to the cloud as training data, and to retrain the initially trained power battery life prediction basic model; The power battery life prediction basic model adopts an LSTM neural network model; the input of the LSTM neural network model is the power battery capacity data of the previous N moments, and the output is the power battery capacity data of the N+1th moment; The training process of the LSTM neural network model is carried out in the cloud. First, a large amount of existing capacity data of the same type of power batteries at different times is used to construct a model training data set to preliminarily train the basic model for power battery life prediction; then, the battery capacity test time series data within a set time uploaded by the edge end is received to further train the preliminarily trained model; The LSTM neural network model includes an input gate, a forget gate and an output gate, specifically including: The expression of the forget gate is: f t =sigmoid(W f ·[h t-1 ,x t ]+b f ); Among them, x t is the input at time t, h t-1 is the output at time t-1, W f and b f are the weight matrix and bias vector of the forget gate respectively, [h t-1 ,x t ] represents the input vector, W f The dimension of h t-1 and x t The expression is: The input gate expression is: i t =sigmoid(W i ·[h t-1 ,x t ]+b i ); Among them, W i and b i are the weight matrix and bias vector of the input gate respectively; the unit state of the current input is calculated by the output of the previous moment and the input of this moment, and the expression is: C' t =tanh(W c ·[h t-1 ,x t ]+b c ); Among them, W c and b c are the weight matrix and bias vector of the current input unit state, tanh is the hyperbolic tangent activation function; through the unit state C t-1 The cell state at the current moment is calculated based on the cell state of the current input. The expression is: C t =f t ⊙C t-1 +i t ⊙C’ t ; The output of LSTM is determined by the output gate and the update state unit, and the expression is: Among them, W o and b o They are the weight matrix and bias vector of the output gate respectively; The capacity degradation trend prediction module is used to perform capacity prediction at the edge based on the battery capacity test time series data within the set time and the retrained power battery life prediction basic model to obtain the overall capacity degradation trend and remaining life of the power battery to be tested; Based on the existing capacity data of multiple power batteries of the same type at different times, the model training data set is expanded. Specifically, the degradation capacity data of different power batteries are multiplied by the corresponding capacity coefficients. The capacity coefficients are selected from the samples generated based on the Sobol sequence. The data expansion method is: C gen =a×C B0005 +b×C B0006 +c×C B0007 Among them, C gen To expand the uncertainty battery degradation capacity data, C B0005~B0007 is the degradation capacity data of batteries No. 5 to No. 7, a, b, c are capacity coefficients respectively; the Sobol sequence is used to generate uniformly distributed samples in [-1,1], parameters a and b are randomly selected in this sample population, c = 1-(a+b).
6. An intelligent rapid testing system for power battery life based on cloud-edge collaboration, characterized in that: include: The edge device is used to perform a cycle charge and discharge test on the power battery to be tested at the edge, and obtain the battery capacity test time series data within a set time; In addition, based on the battery capacity test time series data within the set time and the retrained power battery life prediction basic model, capacity prediction is performed at the edge to obtain the overall capacity degradation trend and remaining life of the power battery to be tested; Cloud device, used to train the basic model for predicting the life of power batteries based on the existing battery capacity data and the battery capacity test time series data received from the edge end; The power battery life prediction basic model adopts an LSTM neural network model; the input of the LSTM neural network model is the power battery capacity data of the previous N moments, and the output is the power battery capacity data of the N+1th moment; The training process of the LSTM neural network model is carried out in the cloud. First, a large amount of existing capacity data of the same type of power batteries at different times is used to construct a model training data set to preliminarily train the basic model for power battery life prediction; then, the battery capacity test time series data within a set time uploaded by the edge end is received to further train the preliminarily trained model; The LSTM neural network model includes an input gate, a forget gate and an output gate, specifically including: The expression of the forget gate is: f t =sigmoid(W f ·[h t-1 ,x t ]+b f ); Among them, x t is the input at time t, h t-1 is the output at time t-1, W f and b f are the weight matrix and bias vector of the forget gate respectively, [h t-1 ,x t ] represents the input vector, W f The dimension of h t-1 and x t The expression is: The input gate expression is: i t =sigmoid(W i ·[h t-1 ,x t ]+b i ); Among them, W i and b i are the weight matrix and bias vector of the input gate respectively; the unit state of the current input is calculated by the output of the previous moment and the input of this moment, and the expression is: C' t =tanh(W c ·[h t-1 ,x t ]+b c ); Among them, W c and b c are the weight matrix and bias vector of the current input unit state, tanh is the hyperbolic tangent activation function; through the unit state C t-1 The cell state at the current moment is calculated based on the cell state of the current input. The expression is: C t =f t ⊙C t-1 +i t ⊙C’ t ; The output of LSTM is determined by the output gate and the update state unit, and the expression is: Among them, W o and b o They are the weight matrix and bias vector of the output gate respectively; Based on the existing capacity data of multiple power batteries of the same type at different times, the model training data set is expanded. Specifically, the degradation capacity data of different power batteries are multiplied by the corresponding capacity coefficients. The capacity coefficients are selected from the samples generated based on the Sobol sequence. The data expansion method is: C gen =a×C B0005 +b×C B0006 +c×C B0007 Among them, C gen To expand the uncertainty battery degradation capacity data, C B0005~B0007 is the degradation capacity data of batteries No. 5 to No. 7, a, b, c are capacity coefficients respectively; the Sobol sequence is used to generate uniformly distributed samples in [-1,1], parameters a and b are randomly selected in this sample population, c = 1-(a+b).
