A battery multi-state joint estimation method for cloud-edge collaboration

Abstract

The application relates to a battery multi-state joint estimation method for cloud-edge cooperation, and belongs to the technical field of batteries. The method is composed of an original data part, a cloud part and an edge part. By processing experimental data in the original data, a battery SOH attenuation model, a battery maximum available energy two-dimensional response surface model and a battery charging and discharging SOP response surface model are established, and the battery state is estimated at the edge end based on the models. Meanwhile, the original data is processed through feature extraction and transfer learning, a cloud battery state estimation model is obtained, the battery health state estimation of the edge end is calibrated, and the multi-state joint estimation of the battery is completed.

Description

Technical Field

[0001] This invention belongs to the field of battery technology and relates to a multi-state joint estimation method for batteries for cloud-edge collaboration. Background Technology

[0002] With the advancement of social productivity, the battery system industry has developed rapidly, finding widespread application in electric vehicles, aerospace, energy storage systems, mobile devices, and medical devices. The state of the battery is crucial to the safety and reliability of the battery system. By monitoring the battery state in real time, it is possible to predict battery failure risks and take corresponding safety protection measures to prevent overcharging, over-discharging, over-temperature, and other problems, thereby reducing the risk of system accidents and ensuring the safety of users and equipment. To optimize energy utilization efficiency and extend battery life, maximizing battery energy utilization and improving energy efficiency, understanding the battery's health status, state of charge, state of energy, and state of power is essential for optimizing battery charging and discharging strategies. However, these states are interconnected and cannot be directly measured. Summary of the Invention

[0003] To address the issue of inter-state interactions and the inability to directly measure these states, this application provides a multi-state joint estimation method for batteries using cloud-edge collaboration. This method can more accurately estimate the battery's state of health, state of charge, state of energy, and state of power. It can be applied to battery systems such as energy storage systems and electric vehicle power battery systems, and has significant practical implications for alleviating energy supply and demand imbalances, improving energy efficiency, and promoting the large-scale application of clean energy.

[0004] According to the first part of this application, a battery health state estimation method based on charging features and transfer learning is provided, including:

[0005] Obtain the charging data of the battery under test during the most recent charging process;

[0006] Based on the charging data, determine the charging characteristic data of the battery under test;

[0007] The charging feature data is input into a preset battery health status estimation model to obtain the current battery health status estimate of the battery under test; wherein, the battery health status estimation model is a neural network model that has learned the mapping relationship between the charging feature data and the battery health status of the battery under test based on transfer learning.

[0008] As one possible implementation, the charging data includes: the cumulative number of charge-discharge cycles, the charging time during which the voltage value of the battery under test is within a preset voltage range, and the voltage and temperature values ​​of the battery under test at at least one moment during the charging time; based on the charging data, the charging characteristic data of the battery under test are determined, including:

[0009] Based on the charging time, determine the charging duration during which the voltage value of the battery under test remains within a preset voltage range during the charging process;

[0010] Based on the voltage and temperature values ​​of the battery under test at at least one moment during the charging time, the voltage and temperature characteristic values ​​of the battery under test are obtained.

[0011] The cumulative number of charge-discharge cycles, charging time, voltage characteristic value, and temperature characteristic value are used as charging characteristic data.

[0012] Among them, voltage characteristic values ​​include voltage standard deviation and voltage distribution skewness.

[0013] In some embodiments of this application, the battery health state estimation model is trained in the following manner:

[0014] Construct a multi-layer deep network LSTM model;

[0015] Source domain data is determined based on experimental charging data of the sample battery under full charge and discharge conditions; wherein, the experimental charging data includes: the cumulative number of charge and discharge cycles of the sample battery during each full charge, the charging time of the voltage value within the preset voltage range, and the voltage value and temperature value at at least one moment during the charging time.

[0016] The multi-layer deep network LSTM model is pre-trained based on the source domain data to obtain the pre-trained LSTM model.

[0017] Based on semi-supervised learning, target domain data is determined according to the actual charging data of the sample battery under actual application conditions; wherein, the actual charging data includes: the cumulative number of charge-discharge cycles of the sample battery during each historical charge, the charging time of the voltage value within the preset voltage range, and the voltage value and temperature value at at least one moment during the charging time.

[0018] Based on the target domain data, the pre-trained LSTM model is fine-tuned, and the fine-tuned LSTM model is used as the battery health state estimation model.

[0019] As one possible implementation, source domain data is determined based on experimental charging data of the sample battery under full charge and discharge conditions, including:

[0020] Based on the experimental charging data, determine the experimental charging characteristic data;

[0021] Obtain the full discharge capacity value of the sample battery under full charge and discharge conditions for each full charge;

[0022] The battery health status value is defined as the ratio of the current capacity to the initial capacity. Therefore, the battery health status value corresponding to each full charge of the sample battery is the ratio of the full discharge capacity corresponding to each full charge to the initial capacity.

[0023] The battery health status value is used as the label value of the experimental charging feature data, and the experimental charging feature data and its label value are used as the source domain data.

[0024] As one possible implementation method, based on semi-supervised learning, the target domain data is determined according to the actual charging data of the sample batteries under real-world application conditions, including:

[0025] Based on the actual charging data, determine the actual charging characteristic data;

[0026] Based on semi-supervised learning, the battery health status value corresponding to the actual charging feature data is determined according to the actual charging feature data.

[0027] The battery health status value is used as the pseudo-label value of the actual charging feature data, and the actual charging feature data and its pseudo-label value are used as the target domain data.

[0028] As an example, based on semi-supervised learning, the battery health status value corresponding to the actual charging characteristic data is determined according to the actual charging characteristic data, including:

[0029] The actual charging characteristic data is input into a preset GRU model, and the prediction result output by the GRU model is used as the battery health status value corresponding to the actual charging characteristic data; wherein, the GRU model is trained based on the source domain data.

[0030] According to Part II of this application, a model-based method for multi-state joint estimation of a battery is provided. This includes:

[0031] A battery SOH decay model was established for SOH estimation, and the ampere-hour integral method was used for SOC estimation. A two-dimensional response surface model of maximum available energy and a two-dimensional response surface model of charge and discharge SOP were used for SOE and SOP estimation.

[0032] As one possible implementation, a battery SOH decay model is established to estimate the battery SOH for batteries at the edge, including:

[0033] Based on the battery SOH decay model, the initial cumulative charge corresponding to the previous battery SOH value at the current temperature is calculated;

[0034] The amount of SOH decay of the battery is calculated based on the increase in cumulative charge since the last SOH estimate.

[0035] It was calibrated using SOH data obtained from the cloud.

[0036] As one possible implementation method, the battery SOC is calculated using the ampere-hour integration method, including:

[0037] The calculation is performed once at each data sampling time using the ampere-hour integration method. For the SOC range of 0 to 0.1 and 0.9 to 1, full charge-discharge calibration is used to obtain the open circuit voltage (OCV) data every 0.01 SOC. The SOC-OCV curves in the range of 0 to 0.1 and 0.9 to 1 are then fitted using the linear interpolation method.

[0038] As one possible implementation, a three-dimensional response surface model of the battery's maximum available energy is established to estimate the battery's SOE, including:

[0039] A three-dimensional response surface model of the battery's maximum available energy, with SOH and ambient temperature as parameters, is established to estimate the battery's SOE.

[0040] As one possible implementation, a three-dimensional response surface model set for charge / discharge SOP is established to estimate the battery SOP, including:

[0041] A three-dimensional response surface model group for charge / discharge SOP is established. Based on the current battery SOH, the group with the closest SOH is selected from the established three-dimensional response surface model group for charge / discharge SOP. Then, based on the battery SOC and ambient temperature, the current battery charge / discharge SOP is obtained through the response surface model.

[0042] The beneficial effects of this invention are as follows: First, by acquiring the charging data of the battery under test during its most recent charging process, and determining the charging characteristic data of the battery under test based on the charging data, the charging characteristic data is input into a preset battery health state estimation model to obtain the estimated value of the remaining current battery health state of the battery under test, while simultaneously acquiring the battery and state of charge (SOC). Then, a battery SOH decay model is established for SOH estimation, an ampere-hour integral method is used for SOC estimation, and a combination of a maximum available energy two-dimensional response surface model and a charge / discharge SOP two-dimensional response surface model is used for SOE and SOP estimation, and cloud data is used for calibration. This scheme achieves multi-state joint estimation of the battery based on multiple models, ensuring the efficient, safe, and stable operation of the battery system.

[0043] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0044] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0045] Figure 1 This is a schematic diagram of the overall principle of the battery multi-state joint estimation method for cloud-edge collaboration in the embodiments of this application;

[0046] Figure 2 This is a data acquisition diagram for a cloud-edge collaborative battery multi-state joint estimation method in an embodiment of this application.

[0047] Figure 3 A flowchart illustrating a multi-state joint estimation method for battery in cloud-edge collaboration, provided as an embodiment of this application;

[0048] Figure 4 This is a flowchart of a cloud-based battery health status estimation model training method in an embodiment of this application;

[0049] Figure 5 This is a flowchart illustrating a cloud-based battery health state estimation method based on charging features and transfer learning, as described in an embodiment of this application.

[0050] Figure 6 This is a flowchart of the edge-end battery SOH estimation process in an embodiment of this application;

[0051] Figure 7 This is a flowchart of the edge battery SOC estimation process in an embodiment of this application;

[0052] Figure 8 This is a flowchart of the edge-end battery SOE estimation process in the embodiments of this application;

[0053] Figure 9 This is a flowchart illustrating the SOP estimation process for the edge battery in this application embodiment. Detailed Implementation

[0054] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0055] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0056] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0057] It's important to note that the strong nonlinear characteristics of batteries stem from complex internal chemical reactions. In theoretical research, battery state estimation commonly employs methods based on equivalent battery models. However, higher accuracy requirements necessitate more complex models and greater difficulty in designing estimation algorithms. Therefore, building an equivalent battery model requires striking a good balance between model accuracy and structural complexity. Battery systems play a crucial role in production; their stability and efficiency directly impact human life and social development. Multi-state joint estimation of batteries can help optimize battery management, increase equipment uptime, improve equipment safety, and optimize energy utilization. Therefore, multi-state joint estimation of batteries has significant practical implications for various fields, including electric vehicles, energy storage systems, and aerospace.

[0058] Most current research on battery state estimation focuses on a single battery state. However, research on multi-state joint estimation is gradually increasing, with studies on two-state joint estimation (SOC and another state) being the most common. Research on joint estimation of more states is relatively rare. Most multi-state joint estimation studies are based on the battery's equivalent model. This involves establishing the equivalent model and combining it with filtering algorithms to estimate SOC and SOE; calculating the battery's peak power (SOP) based on constraints such as current, voltage, and SOC to estimate SOP; and calculating the battery capacity (SOH) based on changes in SOC and charge to estimate SOH.

[0059] To address the aforementioned issues, this application provides a multi-state joint estimation method for batteries for cloud-edge collaboration.

[0060] Figure 1 This is a schematic diagram of the cloud-edge collaborative battery multi-state joint estimation method provided in the embodiments of this application.

[0061] The overall solution consists of three parts: raw data, cloud, and edge. Raw data provides the data foundation for the cloud and edge. The cloud uses raw data to predict battery status and provides calibration services for edge status estimation. The edge uses raw data to estimate battery status and improves accuracy and stability with cloud-based correction.

[0062] The raw data section includes field data and experimental data. Field data consists of real-time battery data detected during state estimation, including battery current, voltage, and temperature. By processing battery experimental data, a battery SOH decay model with accumulated charge and capacity as parameters, a two-dimensional response surface model of the battery's maximum usable energy with SOH and temperature as parameters, and a set of response surface models for charge / discharge SOPs with temperature and SOC as parameters under different SOH conditions are established.

[0063] The cloud-based component processes cloud data and estimates battery health status through feature extraction and transfer learning. First, based on semi-supervised learning and experimental data, a multi-layer deep LSTM model is constructed. The source domain data is determined using experimental data from sample batteries under full charge-discharge conditions, and pre-trained using this source domain data to obtain the pre-trained LSTM model. Then, the trained LSTM model is fine-tuned based on field data, and this fine-tuned LSTM model is used as the battery health status estimation model. The battery health status is then estimated using the estimation model based on field data.

[0064] The edge-end component includes four modules: SOH estimation, SOC estimation, SOE estimation, and SOP estimation. First, based on the battery SOH decay model, the initial cumulative charge corresponding to the previous battery SOH value at the current temperature is calculated. Then, based on the increase in cumulative charge since the last SOH estimation, the SOH decay is calculated. Because the ambient temperature changes little in a short period, the current ambient temperature can be considered a constant value over a previous period, thus obtaining the current SOH estimate. Finally, the SOH value obtained from the cloud is used for calibration. SOC estimation is performed once at each data sampling time using the ampere-hour integration method. The SOC estimate is then calibrated for full charge-discharge. For every 0.01 SOC, the open-circuit voltage (OCV) data is used to fit the SOC-OCV curves for SOC ranges of 0 to 0.1 and 0.9 to 1 using linear interpolation. A three-dimensional response surface model of the battery's maximum available energy (SOE) is established using SOH and ambient temperature as parameters to estimate the battery's SOE. Based on the current battery SOH, the group with the closest SOH values ​​is selected from the established three-dimensional response surface model set for charge and discharge SOP. Then, based on the battery SOC and ambient temperature, the current battery charging SOP and discharging SOP are obtained through the response surface model, and calibrated using cloud data.

[0065] Figure 2 This diagram illustrates the data acquisition process for a cloud-edge collaborative battery multi-state joint estimation method provided in this application. As part of the embodiment, the raw data includes experimental data and field data. The field data consists of battery data detected in real-time during state estimation, including battery current, voltage, and temperature. By processing battery experimental data, a battery SOH decay model using accumulated charge and capacity as parameters, a two-dimensional response surface model of the battery's maximum usable energy using SOH and temperature as parameters, and a set of charge / discharge SOPs using temperature and SOC as parameters under different SOH conditions are established. The experimental data and field data provide data support for state estimation at the edge and cloud. The state estimation results from the cloud provide correction services for the edge.

[0066] Figure 3 This is a flowchart illustrating a battery multi-state joint estimation method for cloud-edge collaboration provided in an embodiment of this application. This battery multi-state joint estimation method for cloud-edge collaboration, as described in this embodiment, can be used for batteries, such as... Figure 1 As shown, the method may include the following steps:

[0067] Step 101: Obtain the charging data of the battery under test in the most recent charging process;

[0068] As one implementation method, if the electronic device performing the method has an interactive display screen, the charging data of the battery under test in the most recent charging process can be obtained by relevant personnel monitoring the charging process, inputting the monitored charging data through the interactive display screen of the electronic device, and submitting it so that the electronic device can obtain the corresponding charging data.

[0069] As another implementation, if the battery's charging data is stored in a server, the electronic device can send a request to the server to obtain the charging data of the most recent charging process and receive the charging data returned by the server based on the request.

[0070] In some embodiments of this application, the charging data may include: cumulative charge-discharge cycle count, cumulative charge, SOC, the charging time during which the voltage value of the battery under test is within a preset voltage range, and the voltage and temperature values ​​of the battery under test at at least one moment during the charging time.

[0071] The cumulative charge-discharge cycle count refers to the current number of times the battery under test has been charged. This cumulative charge-discharge cycle count can be a directly obtained value. The cumulative capacity refers to the current capacity of the battery under test. This cumulative capacity can also be a directly obtained value. State of Charge (SOC) is calculated once at each data sampling time using the ampere-hour integration method. The charging time during which the voltage value of the battery under test is within a preset voltage range refers to the start and end times of the voltage value within that preset voltage range. For example, if the preset voltage range is 3.8V-4.0V, then the charging time during this preset time period refers to the time during which the voltage value of the battery under test is 3.8V and the time during which the voltage value is 4.0V during the charging process.

[0072] Furthermore, the voltage and temperature values ​​of the battery under test at at least one moment during the charging time can be obtained by sampling the voltage and temperature values ​​at corresponding moments within the charging time at preset time intervals. For example, if the charging time is 8:30:02-8:40:00 and the sampling interval is 10 seconds, this step can be to sequentially obtain the voltage and temperature values ​​at times 8:30:02, 8:30:12, 8:30:22, 8:30:32, 8:30:42, and 8:30:52.

[0073] Then, based on the charging data, the charging characteristic data of the battery under test are determined.

[0074] In other words, feature extraction is performed on charging data to obtain charging feature data. This charging feature data can include charging duration, cumulative charge / discharge cycles, voltage characteristic values, and temperature characteristic values.

[0075] In some embodiments of this application, step 102 may include the following steps: determining the charging duration during which the voltage value of the battery under test is within a preset voltage range based on the charging time; obtaining the voltage characteristic value and temperature characteristic value of the battery under test based on the voltage value and temperature value of the battery under test at least at one moment during the charging time; and using the cumulative charge-discharge cycle count, charging duration, voltage characteristic value, and temperature characteristic value as charging characteristic data.

[0076] It is understandable that, for a selected charging duration, the current during the constant current charging phase is constant, and the charging duration within the preset voltage range can characterize the charging capacity, amplifying the differences in voltage curves under different aging cycles. Therefore, these differences can be captured using two voltage-related statistical features (voltage standard deviation and voltage distribution skewness). For the preset voltage range, the voltage standard deviation reflects the voltage dispersion, while the voltage distribution skewness reflects the skewness of the voltage distribution. Simultaneously, the current battery temperature characteristics and cycle count are extracted during this stage.

[0077] As one possible implementation, voltage characteristics can include, but are not limited to, voltage standard deviation and voltage distribution skewness. Considering the convenience of practical applications, and to avoid error amplification and information loss caused by secondary processing of the charging curve, the charging time, voltage standard deviation, and voltage skewness of the preset voltage segments in the charging curve are directly extracted.

[0078] The charging time can be calculated using the following formula (1):

[0079] ΔV charge_time =t end -t start (1)

[0080] ΔV charge_time For charging time; t end The time t represents the end of the preset voltage range during the charging process of the battery under test; start This refers to the start time when the voltage value of the battery under test reaches the preset voltage range during the charging process.

[0081] As one possible implementation, the voltage characteristic value may include the voltage standard deviation and the voltage distribution skewness. As an example, if the voltage characteristic value includes the voltage standard deviation and the voltage distribution skewness, the voltage characteristic value of the battery under test can be obtained based on the voltage value at at least one moment during the charging time, and can be calculated using the following formulas (2) and (3):

[0082]

[0083]

[0084] Where, ΔV std For voltage standard deviation, V i Let i be the voltage value at time i. The average voltage value at at least one moment; n is the number of sampling points; ΔV skew This represents the skewness of the voltage distribution.

[0085] In addition, the temperature characteristic value of the battery under test can be obtained based on the temperature value of the battery under test at least at one moment during the charging time. This can be achieved by calculating the average temperature value of the temperature value of the battery under test at at least one moment during the charging time and using the obtained average temperature value as the temperature characteristic value.

[0086] Step 102: Input the charging feature data into the preset battery health state estimation model to obtain the current battery health state estimate of the battery under test; wherein, the battery health state estimation model is a neural network model that has learned the mapping relationship between the charging feature data of the battery under test and the battery health state based on transfer learning.

[0087] In some embodiments of this application, the battery health state estimation model may be trained based on an LSTM model, or it may be trained based on other neural network models according to actual needs. This application does not limit this.

[0088] As one implementation method, an initial neural network model is pre-trained based on the charging characteristic data of the battery under test in an ideal full-charge and discharge state and the battery health status value, so that the neural network model learns the mapping relationship between charging characteristics and battery health status value in a full-charge and discharge scenario. Based on the charging characteristic data of the battery under test in a real application scenario, the capacity pseudo-label of the corresponding charging characteristic data is determined, and the pre-trained neural network model is transferred and fine-tuned based on the charging characteristic data of the battery under test in a real application scenario and its corresponding battery health status value pseudo-label, so as to learn the characteristic information of the battery under test in a real application scenario, and finally obtain the battery health status estimation model.

[0089] Step 103: Establish a battery SOH decay model with accumulated charge and capacity as parameters, calibrate it using cloud-predicted SOH values, and perform SOC estimation. Then, perform multi-state joint estimation of the battery using a two-dimensional response surface model of the battery's maximum usable energy with SOH and temperature as parameters, and a two-dimensional response surface model of charge and discharge SOP with temperature and SOC as parameters under different SOH conditions.

[0090] The State of Health (SOH) estimation process requires determining whether to perform SOH estimation at each data sampling time based on the SOH estimation cycle. If SOH estimation is not performed, the battery SOH remains the same as the previous time. If SOH estimation is performed, firstly, based on the battery SOH decay model, the initial accumulated capacity corresponding to the previous battery SOH value at the current temperature is calculated. Then, based on the increase in accumulated capacity since the last SOH estimation, the SOH decay is calculated. Because the ambient temperature changes little in a short period, the current ambient temperature can be considered a constant value over a previous period, thus obtaining the current SOH estimate. Finally, the current battery capacity is calculated based on the current SOH and the initial capacity corresponding to the current ambient temperature. The calculation formula is as follows:

[0091] C a (t)=C0(t)×SOH(t)

[0092] In the formula, C a C0(t) represents the battery capacity at the current time, C0(t) represents the initial battery capacity corresponding to the current temperature, and SOH(t) represents the battery SOH at the current time.

[0093] The SOC estimation is performed once at each data sampling time and is calculated using the ampere-hour integration method. The calculation formula is as follows:

[0094]

[0095] In the formula, SOC(t) and SOC(t-1) represent the SOC of the battery at the current time and the previous time, respectively. The change in SOC is calculated based on the change in battery charge and battery capacity. Then, the current SOC can be obtained based on the SOC of the battery at the previous time, and full charge-discharge calibration is used.

[0096] The State of Energy (SOE) of a battery is crucial for energy management and optimization of a battery system. A battery has maximum usable energy when fully charged; when fully discharged, its remaining energy is zero. The method for estimating the battery SOE is as follows:

[0097] SOE(t)=E a (t)×SOC(t)

[0098] In the formula, SOE(t) represents the current battery SOE, and E... a (t) represents the maximum available energy of the battery at the current moment. The maximum available energy of the battery is calculated using a three-dimensional response surface model of the maximum available energy, as shown in the following formula:

[0099]

[0100] SOP estimation is based on the current battery SOH. From the established three-dimensional response surface model set for charge / discharge SOP, the model with the closest SOH values ​​is selected. Then, based on the battery SOC and ambient temperature, the current battery charge / discharge SOP is obtained through the response surface model. The calculation formula is as follows:

[0101] SOP(t) = f SOP (SOC(t), T) env (t))

[0102] In the formula, SOP(t) is the battery charge / discharge SOP at the current moment, and the right side of the equation represents the three-dimensional response surface model of the battery charge / discharge SOP with SOC and ambient temperature as parameters.

[0103] This application presents a multi-state joint estimation method for batteries used in cloud-edge collaboration. First, a battery health state estimation method based on charging characteristics and transfer learning is employed. This method acquires charging data from the most recent charging process of the battery under test and determines its charging characteristic data based on this data. This charging characteristic data is then input into a preset battery health state estimation model to obtain the current estimated battery health state. Next, a battery SOH degradation model is established for SOH estimation, and an ampere-hour integral method is used for SOC estimation. A two-dimensional response surface model of maximum available energy and a two-dimensional response surface model of charge / discharge SOP are used for SOE and SOP estimation. Ultimately, this multi-state joint estimation not only avoids equipment failures or data loss caused by battery degradation but also optimizes battery usage, ensuring efficient, safe, and stable operation of the battery system.

[0104] Next, another embodiment of the training process for the battery health state estimation model is proposed.

[0105] Figure 4 This is a flowchart of a battery health state estimation model training method in an embodiment of this application.

[0106] like Figure 4 As shown, the training method for this battery health state estimation model may include the following steps:

[0107] Step 201: Construct a multi-layer deep network LSTM model.

[0108] In some embodiments of this application, the single-layer LSTM model includes a forget gate, an input gate, cell state units, and finally an output gate and an output unit, the parameters of which are calculated as follows:

[0109]

[0110] Among them, f t i t Ct o t h t These represent the forget gate, input gate, cell state unit, and finally the output gate and output unit in a single-layer network, respectively. σ and tanh represent the sigmoid function and tanh function, respectively, with values ​​limited to 0~1 and -1~1 for each parameter. t-1 h t-1 x t These represent the cell state unit from the previous time step, the output unit from the previous time step, and the input unit from the current time step, respectively. W and b are the weights and biases of each parameter. f W i W C W o b represents the weight matrices for the forget gate, input gate, cell state unit, and output gate, respectively. f b i b C b o The biases represent the forget gate, input gate, cell state unit, and output gate.

[0111] Based on a single-layer LSTM model, add another identical single-layer LSTM model and stack them together. The input x of the second layer... t It is the h of the previous level. t This process continues through the third layer, and so on, up to the nth layer in a multi-layered deep network. In stacked LSTMs, this hierarchical structure allows us to represent time-series data in more complex ways, capturing information at different scales. Simultaneously, we define parameters such as the input, output, and hidden layer sizes of the LSTM model, the number of LSTM layers, the learning rate, and the number of iterations. Specific parameter settings need to be adjusted based on the prediction results, as shown in Table 1 below.

[0112] Table 1 LSTM Model Parameter Settings

[0113] Input feature dimension 5 Number of neurons in hidden layer 128 Output Dimension (RUL) 1 LSTM layers 4 Learning rate 0.01 Magnitude of learning rate change 0.9 Number of iterations 3000

[0114] Step 202: Determine source domain data based on experimental charging data of the sample battery under full charge and discharge conditions; wherein, experimental charging data includes: the cumulative number of charge and discharge cycles of the sample battery during each full charge, the charging time of the voltage value within the preset voltage range, and the voltage value and temperature value at at least one moment during the charging time.

[0115] It should be noted that the sample battery can be the same battery as the battery under test, or it can be the same type of battery but not the same battery.

[0116] In some embodiments of this application, step 202 specifically includes: determining experimental charging characteristic data based on experimental charging data; obtaining the full discharge capacity value corresponding to each full charge under full charge and discharge conditions for the sample battery; using the battery health status value (i.e., the ratio of current capacity to initial capacity) as the label value of the experimental charging characteristic data, and using the experimental charging characteristic data and its label value as source domain data. The full discharge capacity value of the sample battery under full charge and discharge conditions can be determined based on the integral of current and time during the discharge process.

[0117] It is understandable that the full discharge capacity of the sample battery can be calculated under full charge and discharge conditions, that is, the health status of the sample battery can be determined. Therefore, the charging characteristic data under full charge and discharge conditions and its corresponding battery health status value can be used as source domain data to pre-train the constructed multi-layer deep network LSTM model, so that the model can learn the mapping relationship between charging characteristics and remaining life under full charge and discharge scenarios.

[0118] Furthermore, the specific implementation process for determining the experimental charging characteristic data based on the experimental charging data can be consistent with the implementation method in the above embodiments, and will not be repeated here.

[0119] Step 203: Pre-train the multi-layer deep network LSTM model based on the source domain data to obtain the pre-trained LSTM model.

[0120] In some embodiments of this application, in order to unify the units of measurement and speed up network training, the source domain data can be normalized before the multilayer deep network LSTM model is pre-trained based on the normalized data.

[0121] As one possible approach, to verify the fit and prediction accuracy of the pre-trained model, the data extracted from the laboratory can be divided into training and test sets for pre-training, for example, the training and test sets can be divided in a ratio of 8:2 or 9:1.

[0122] As an example, the training set can be input into a multi-layer deep network LSTM model, and the loss after each iteration can be calculated using the mean squared error (MSE) loss function. The calculation process is shown in formula (4):

[0123]

[0124] Among them, y i It is the normalized value of the health state value of the i-th battery in the sample battery. This is the i-th predicted value of the sample battery obtained through the LSTM model. Then, the Adaptive Moment Estimation (Adam) optimizer is used, and the learning rate variation rule is defined using the StePLearning Rate (StepLR) function. Backpropagation and optimization updates are performed on the multi-layer deep LSTM network model. During training, parameters such as the input, output, and hidden layer sizes of the LSTM model, the number of LSTM layers, the learning rate, and the number of iterations are repeatedly adjusted until the ideal prediction result is achieved, resulting in the pre-trained LSTM model.

[0125] Step 204: Based on semi-supervised learning, determine the target domain data according to the actual charging data of the sample battery under actual application conditions; wherein, the actual charging data includes: the cumulative number of charge-discharge cycles of the sample battery during each historical charge, the charging time of the voltage value within the preset voltage range, and the voltage value and temperature value at least at one moment during the charging time.

[0126] It's understandable that batteries in real-world applications cannot be fully charged and discharged. For example, in actual vehicles, users might charge the battery when its charge is low, rather than waiting until it's completely depleted, to ensure normal vehicle operation. Therefore, to ensure the trained battery remaining life prediction model accurately predicts remaining life and is more adaptable to real-world battery application scenarios, the pre-trained LSTM model can be fine-tuned using charging data from actual battery usage scenarios.

[0127] In some embodiments of this application, the specific implementation process of step 204 may include: determining actual charging feature data based on actual charging data; determining the battery health status value corresponding to the actual charging feature data based on the actual charging feature data using a semi-supervised learning approach; using the battery health status value as the pseudo-label value of the actual charging feature data, and using the actual charging feature data and its pseudo-label value as target domain data.

[0128] The specific implementation method for determining the actual charging feature data based on actual charging data is consistent with the implementation method in the above embodiments, and will not be repeated here. Since in practical applications, battery charging strategies are generally multi-stage constant current charging, and not full charging and discharging, only the features of the preset voltage range can be extracted, and the battery's full discharge capacity value cannot be obtained. A semi-supervised learning approach can be used to determine the battery health state value corresponding to the actual charging feature data based on the actual charging feature data, and the battery health state value can be used as a pseudo-label for the actual charging feature data.

[0129] As one possible implementation method, based on semi-supervised learning, the method for determining the discharge capacity value corresponding to the actual charging characteristic data can be as follows: the actual charging characteristic data is input into a preset GRU model, and the prediction result output by the GRU model is used as the battery health state value corresponding to the actual charging characteristic data; wherein, the GRU model is trained based on source domain data. Figure 5 As shown, the initial GRU model is iteratively learned based on experimental charging feature data. The resulting GRU model can predict the corresponding battery health status value based on actual charging feature data, and use the prediction result as the pseudo label corresponding to the actual charging feature data.

[0130] Step 205: Based on the target domain data, fine-tune the pre-trained LSTM model and use the fine-tuned LSTM model as the battery health state estimation model.

[0131] In some embodiments of this application, the fine-tuning process first freezes the architecture and parameters of the n-1 layers of the pre-trained LSTM model, and then retrains the n-layer network and the fully connected layer using target domain data, changing the parameters and weights of these two layers.

[0132] As an example, the fine-tuning process uses MSE to calculate the loss after each iteration, and leverages the Adam optimizer and StepLR function to define the rules for changing the learning rate, thus fine-tuning the pre-trained LSTM model.

[0133] like Figure 3 The diagram shows a flowchart of a multi-state joint estimation method for batteries used in cloud-edge collaboration. It can be divided into the following parts: the original data part, the cloud part, and the edge part.

[0134] The raw data provides the data foundation for the cloud and edge. The cloud uses the raw data to predict the state and provides calibration services for the state estimation of the edge. The edge uses the raw data to estimate the battery state and improves accuracy and stability with the correction in the cloud.

[0135] The raw data section includes field data and experimental data. Field data consists of real-time battery data detected during state estimation, including battery current, voltage, and temperature. By processing battery experimental data, a battery SOH decay model with accumulated charge and capacity as parameters, a two-dimensional response surface model of the battery's maximum usable energy with SOH and temperature as parameters, and a set of response surface models for charge / discharge SOPs with temperature and SOC as parameters under different SOH conditions are established.

[0136] The cloud-based component processes cloud data and estimates battery health status through feature extraction and transfer learning. First, based on semi-supervised learning and experimental data, a multi-layer deep LSTM model is constructed. The source domain data is determined using experimental data from sample batteries under full charge-discharge conditions, and pre-trained using this source domain data to obtain the pre-trained LSTM model. Then, the trained LSTM model is fine-tuned based on field data, and this fine-tuned LSTM model is used as the battery health status estimation model. The battery health status is then estimated using the estimation model based on field data.

[0137] The edge-end component includes four modules: SOH estimation, SOC estimation, SOE estimation, and SOP estimation. First, based on the battery SOH decay model, the initial cumulative charge corresponding to the previous battery SOH value at the current temperature is calculated. Then, based on the increase in cumulative charge since the last SOH estimation, the SOH decay is calculated. Because the ambient temperature changes little in a short period, the current ambient temperature can be considered a constant value over a previous period, thus obtaining the current SOH estimate. Finally, the SOH value obtained from the cloud is used for calibration. SOC estimation is performed once at each data sampling time using the ampere-hour integration method. The SOC estimate is then calibrated for full charge-discharge. For every 0.01 SOC, the open-circuit voltage (OCV) data is used to fit the SOC-OCV curves for SOC ranges of 0 to 0.1 and 0.9 to 1 using linear interpolation. A three-dimensional response surface model of the battery's maximum available energy (SOE) is established using SOH and ambient temperature as parameters to estimate the battery's SOE. Based on the current battery SOH, the group with the closest SOH values ​​is selected from the established three-dimensional response surface model set for charge and discharge SOP. Then, based on the battery SOC and ambient temperature, the current battery charging SOP and discharging SOP are obtained through the response surface model, and calibrated using cloud data.

[0138] 1. To clearly express the battery SOH estimation of the cloud data in this application, it will be expressed as follows: Figure 4 For example, the sample battery and the battery under test are the same battery, and the battery under test is used in an actual battery. The battery health state estimation method based on charging features and transfer learning is introduced in the form of an example.

[0139] (1) Obtain experimental charging data of the battery under experimental conditions and actual charging data of the battery in reality;

[0140] (2) Feature extraction was performed on the experimental charging data and the actual charging data respectively to extract experimental charging feature data and actual charging feature data;

[0141] (3) Normalize the data;

[0142] (4) The multilayer deep network LSTM model is trained based on the normalized experimental charging data, and the GRU model is trained based on the normalized experimental charging data.

[0143] (5) Based on the normalized actual charging feature data, the pseudo-labels corresponding to the actual charging feature data are obtained by semi-supervised learning using the trained GRU model.

[0144] (6) Freeze the n-1 layers of the LSTM model, use the normalized actual charging feature data and the corresponding pseudo-labels to perform transfer learning on the last LSTM layer and the fully connected layer, fine-tune the LSTM model, and obtain the battery health estimation model.

[0145] (7) In subsequent applications, the charging data of the actual battery in the most recent charging process is obtained and the features are extracted to obtain charging feature data.

[0146] (8) The normalized charging characteristic data is input into the battery remaining life prediction model, and the output of the model is reverse normalized to obtain the current health status estimate of the battery under test.

[0147] 2. To clearly express the battery SOH estimation of the edge data in this application, it will be based on... Figure 7 For example, the basic method is as follows:

[0148] (1) Obtain experimental charging data of the battery under experimental conditions and actual charging data of the battery in reality;

[0149] (2) Update the battery's cumulative charge in real time at each data sampling moment;

[0150] (3) Calibrate using the SOH value estimated in the cloud;

[0151] (4) Perform full charge-discharge calibration regularly;

[0152] (5) Obtain the final battery health status value.

[0153] 3. To clearly express the battery SOC estimation of the edge data in this application, it will be based on... Figure 8 For example, the basic method is as follows:

[0154] (1) Obtain experimental charging data of the battery under experimental conditions and actual charging data of the battery in reality;

[0155] (2) The ampere-hour integration method is used for calculation at each data sampling time;

[0156] (3) Full charge-discharge calibration is adopted;

[0157] (4) Obtain the final battery SOC.

[0158] 4. To clearly express the battery SOE estimation of the edge data in this application, it will be based on... Figure 9 For example, the basic method is as follows:

[0159] (1) Obtain experimental charging data of the battery under experimental conditions and actual charging data of the battery in reality;

[0160] (2) Update the battery's cumulative charge in real time at each data sampling moment;

[0161] (3) Obtain the final battery SOE.

[0162] 5. To clearly express the battery SOP estimation of the edge data in this application, the basic method is as follows:

[0163] (1) Obtain experimental charging data of the battery under experimental conditions and actual charging data of the battery in reality;

[0164] (2) Based on the current battery SOH, select the group with the closest SOH from the established charge / discharge SOP three-dimensional response surface model group;

[0165] (3) The battery charge / discharge SOP at the current moment is obtained by using the response surface model based on the battery SOC and ambient temperature.

[0166] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0167] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0168] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0169] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0170] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0171] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A multi-state joint estimation method for batteries for cloud-edge collaboration, characterized in that: The method includes the following steps: Three models were built using raw data to process edge data and estimate battery status. The cloud data is processed and battery health is estimated by extracting features from the raw data and using transfer learning. The edge data is calibrated using cloud-based prediction results. Then, a set of battery SOH decay models, battery maximum available energy two-dimensional response surface models, and charge / discharge power two-dimensional response surface models are established to perform multi-state joint estimation of the battery. Specifically, the processing of edge data includes: By processing experimental data related to battery experiments, we established a battery SOH decay model with cumulative charge and capacity as parameters to estimate SOH, used the ampere-hour integration method to estimate SOC, a two-dimensional response surface model of the battery maximum usable energy with SOH and temperature as parameters, and a set of two-dimensional response surface models of charge and discharge SOP with temperature and SOC as parameters under different SOH conditions to estimate SOE and SOP. The cloud data is processed in the following ways: Obtain the charging data of the battery under test during the most recent charging process; Based on the charging data, determine the charging characteristic data of the battery under test; The charging feature data is input into a preset battery health status estimation model to obtain the current battery health status estimate of the battery under test; wherein, the battery health status estimation model is a neural network model that has learned the mapping relationship between the charging feature data and the battery health status of the battery under test based on transfer learning. The cloud-based prediction results are then calibrated against the edge data using the following method: The cloud prediction results and edge prediction results are output in a weighted manner.

2. The battery multi-state joint estimation method for cloud-edge collaboration according to claim 1, characterized in that: The method for estimating SOH is as follows: First, based on the battery SOH decay model, calculate the initial cumulative charge corresponding to the previous SOH value of the battery at the current temperature. Then, based on the increase in cumulative charge after the previous SOH estimate, calculate the decay of the battery SOH. Since the ambient temperature of the battery changes little in a short period of time, the ambient temperature of the battery at the current moment can be regarded as a constant value in the previous period of time, thus obtaining the estimated SOH value of the battery at the current moment.

3. The battery multi-state joint estimation method for cloud-edge collaboration according to claim 1, characterized in that: The method for SOC estimation is as follows: The calculation is performed once at each data sampling time using the ampere-hour integration method. Full charge-discharge calibration is used for the SOC range of 0 to 0.1 and 0.9 to 1. Open circuit voltage (OCV) data is obtained every 0.01 SOC. The SOC-OCV curves in the range of 0 to 0.1 and 0.9 to 1 are then fitted using linear interpolation.

4. The battery multi-state joint estimation method for cloud-edge collaboration according to claim 1, characterized in that: The method for SOE estimation is as follows: A three-dimensional response surface model of the battery's maximum available energy (SOE) was established using SOH and ambient temperature as parameters to estimate the battery's SOE.

5. The battery multi-state joint estimation method for cloud-edge collaboration according to claim 1, characterized in that: The method for estimating SOP is as follows: Based on the current battery SOH, select the group with the closest SOH from the established three-dimensional response surface model group for charge and discharge SOP. Then, based on the battery SOC and ambient temperature, obtain the current battery charge and discharge SOP through the response surface model.

6. The battery multi-state joint estimation method for cloud-edge collaboration according to claim 1, characterized in that: The charging data includes: The cumulative number of charge-discharge cycles, the charging time during which the voltage value of the battery under test is within a preset voltage range, and the voltage and temperature values ​​of the battery under test at at least one moment during the charging time; the step of determining the charging characteristic data of the battery under test based on the charging data includes: Based on the charging time, determine the duration for which the voltage value of the battery under test remains within a preset voltage range during the charging process; Based on the voltage and temperature values ​​of the battery under test at at least one moment during the charging time, the voltage characteristic value and temperature characteristic value of the battery under test are obtained. The cumulative number of charge-discharge cycles, the charging duration, the voltage characteristic value, and the temperature characteristic value are used as the charging characteristic data.

7. The battery multi-state joint estimation method for cloud-edge collaboration according to claim 6, characterized in that: The battery health state estimation model was trained in the following way: Construct a multi-layer deep network LSTM model; Source domain data is determined based on experimental charging data of the sample battery under full charge and discharge conditions; wherein, the experimental charging data includes: the cumulative number of charge and discharge cycles of the sample battery during each full charge, the charging time of the voltage value within the preset voltage range, and the voltage value and temperature value at at least one moment during the charging time. The multi-layer deep network LSTM model is pre-trained based on the source domain data to obtain the pre-trained LSTM model. Based on semi-supervised learning, target domain data is determined according to the actual charging data of the sample battery under actual application conditions; wherein, the actual charging data includes: the cumulative number of charge-discharge cycles of the sample battery during each historical charge, the charging time of the voltage value within the preset voltage range, and the voltage value and temperature value at at least one moment during the charging time. Based on the target domain data, the pre-trained LSTM model is fine-tuned, and the fine-tuned LSTM model is used as the battery health state estimation model.

8. The battery multi-state joint estimation method for cloud-edge collaboration according to claim 7, characterized in that: The step of determining source domain data based on experimental charging data of the sample battery under full charge and discharge conditions includes: Based on the experimental charging data, determine the experimental charging characteristic data; Obtain the full discharge capacity value of the sample battery for each full charge under the full charge and discharge conditions; The battery health status value is defined as the ratio of the current capacity to the initial capacity. Therefore, the battery health status value corresponding to each full charge of the sample battery is the ratio of the full discharge capacity corresponding to each full charge to the initial capacity. The battery health status value is used as the label value of the experimental charging feature data, and the experimental charging feature data and its label value are used as the source domain data.

9. A battery multi-state joint estimation method for cloud-edge collaboration according to claim 8, characterized in that: The semi-supervised learning-based approach determines the target domain data based on the actual charging data of the sample batteries under real-world application conditions, including: Based on the actual charging data, determine the actual charging characteristic data; Based on semi-supervised learning, the battery health status value corresponding to the actual charging feature data is determined according to the actual charging feature data. The battery health status value is used as the pseudo-label value of the actual charging feature data, and the actual charging feature data and its pseudo-label value are used as the target domain data.

10. A battery multi-state joint estimation method for cloud-edge collaboration according to claim 9, characterized in that: The semi-supervised learning-based method determines the battery health status value corresponding to the actual charging feature data based on the actual charging feature data, including: The actual charging feature data is input into a preset GRU model, and the prediction result output by the GRU model is used as the battery health status value corresponding to the actual charging feature data; wherein, the GRU model is trained based on the source domain data.

Images