Lithium battery SOC end cloud collaborative deep learning estimation method and system

Abstract

The invention discloses a lithium battery SOC end cloud collaborative deep learning estimation method and system, and relates to the technical field of battery management and control, and the method comprises the steps: obtaining battery charging cycle process data from a cloud end, correcting the SOC through a reverse ampere-hour integration method, and generating self-supervision SOC estimation training data through automatic tagging; constructing a CGRM model, combining the feature extraction capability of a CNN learning device and the long-term dependency relationship capture capability of a GRU learning device, and achieving the cloud precise estimation of the SOC through the integrated output of a meta learning device; generating a characteristic time sequence of SOC estimation by adopting a sliding window method, and inputting the characteristic time sequence into the CGRM model to obtain an SOC estimation result; and transmitting the SOC data estimated by the cloud to a user side battery management system, calibrating the estimated SOC value, and realizing accurate estimation of the battery SOC at the user side. According to the method, a large number of experimental data samples are not needed, the data acquisition cost is reduced by utilizing an end-cloud cooperation mode, the problem that the experimental environment is not matched with the real working condition is solved, and the SOC estimation precision and the battery operation safety are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of battery management and control technology, and in particular to a lithium battery SOC end-cloud collaborative deep learning estimation method and system based on automatic sample generation. Background Art

[0002] Lithium-ion batteries are widely used in different fields such as energy storage systems, battery power supplies, and electric vehicles. In order to ensure their safe use, full utilization, and intelligent operation and maintenance, it is necessary to accurately estimate the battery state of charge (SOC). To this end, domestic and foreign scholars have conducted a lot of research on SOC estimation methods, mainly including the ampere-hour integration method, the open circuit voltage method, the model-based method, and the data-driven method. Among them, the ampere-hour integration method is simple and easy, but long-term use will produce cumulative errors and require an accurate initial SOC. The open circuit voltage method requires the battery to be in a static state for a long time, which cannot be met in actual application conditions. The model-based method establishes an equivalent circuit model or electrochemical mechanism model of the battery, and then applies algorithms such as the extended Kalman filter to estimate the SOC. However, the complex internal mechanism and aging of lithium-ion batteries affect the modeling accuracy. At the same time, the measurement noise and data drift in the real environment also make the model parameters difficult to identify. Therefore, the model-based method is also difficult to meet the needs of actual applications.

[0003] With the development of machine learning and artificial intelligence, data-driven methods have developed rapidly. By using machine learning models to directly estimate the SOC value of the battery through the measurement parameters of the battery's current, voltage, temperature, etc., it not only does not consider the internal mechanism of the battery, but also avoids the difficulty of establishing a battery model in the model-based method. Among them, data samples are the problem that must be solved first. However, the laboratory cannot fully simulate the operating environment of the real-world battery system. Most of the existing SOC estimates in existing studies are carried out in a stable experimental environment. The application effect of most algorithms in the real world cannot be guaranteed. At the same time, the SOC of the user-side battery management system BMS cannot be used as a reliable sample for model training, which restricts the performance of machine learning methods in estimating battery SOC. How to accurately estimate the battery state of charge (SOC) is still an important problem that needs to be solved in the energy storage, battery power, and automotive industries.

[0004] Patent document CN118858958A discloses a lithium-ion battery SOC estimation method based on an improved transformer. The method extracts the current, voltage and temperature curves of the battery data set under different working conditions, and performs outlier judgment to obtain the optimal current, voltage and temperature curves. The SOC calculated by the ampere-hour integration method is then used as the label value. The bidirectional long short-term memory network with gated linear units is introduced into the transformer structure, and the optimal current, voltage, temperature curve, state of charge and the average value of current and voltage are input to train the transformer for predicting the state of charge. This method can make the SOC estimation output smoother and more stable, but the method still trains the model based on the data set obtained from the test under different working conditions, and therefore cannot adapt to the complexity and variability of the operating environment of the real-world battery system, which limits its practical application effect.

[0005] Patent document CN118839114A discloses a method for predicting the SOC of electric mining truck power batteries based on digital twins. Through the battery Thevenin equivalent circuit model and the vehicle simulink model, a digital twin model of the electric mining truck power battery pack is established to simulate the operating state of the power battery system, and then a deep residual shrinkage network is used to learn and train the simulation data to achieve the prediction of the battery SOC. This method can dynamically adjust the training optimization model according to the real-time acquired data and external conditions, but it essentially still relies on the battery Thevenin equivalent circuit model. Therefore, it is limited by the accuracy of this type of model, and the estimation accuracy of the battery SOC is limited. Summary of the invention

[0006] The present invention provides a lithium battery SOC end-cloud collaborative deep learning estimation method and system based on automatic sample generation, aiming to automatically generate data samples for battery SOC estimation by mining real-world battery operation data, thereby solving the problems of high cost and long cycle for sample data acquisition, especially the problem that it is difficult to fully simulate the real-world battery working environment conditions under laboratory conditions. At the same time, a convolutional gated recurrent unit meta-learning model is proposed to achieve accurate end-cloud collaborative deep learning estimation of battery SOC.

[0007] In order to solve the above problems, the technical solution adopted by the present invention is as follows:

[0008] In a first aspect, the present invention provides a lithium battery SOC end-cloud collaborative deep learning estimation method, comprising the following steps:

[0009] Step 1: Clean the real-world user-end lithium battery operation data collected in the cloud to obtain effective charging cycle process data, including battery charging voltage, current and temperature data;

[0010] Step 2, according to the effective charging cycle process data, correcting the SOC using a reverse ampere-hour integration method in the cloud, and generating self-supervised SOC estimation training data by automatically labeling the SOC;

[0011] Step 3, construct a SOC estimation model based on a convolutional gated recurrent unit meta-learning network;

[0012] Step 4: Using the SOC estimation training data, a sliding window method is used to generate the feature time series required for SOC estimation, and the time series and its corresponding automatic label SOC value are input into the SOC estimation model to obtain the cloud-based estimated SOC data;

[0013] Step 5: Transmit the cloud-estimated SOC data to the user-side lithium battery management system, calibrate the battery SOC estimated by the user-side lithium battery management system, and achieve accurate estimation of the lithium battery SOC at the user side.

[0014] As a further description of the above technical solution, step 2 includes:

[0015] Step 2.1, automatically searching for a fully charged single battery that meets the cut-off voltage in the effective charging cycle process data, and setting the SOC of the fully charged single battery to 100%;

[0016] Step 2.2, calculate the SOC value at other times in the charging cycle of the searched fully charged single battery by the reverse ampere-hour integration method, realize automatic labeling of SOC, and obtain SOC estimation training data after the automatic labeling is completed;

[0017] The SOC value calculation formula is:

[0018]

[0019] Among them, SOC r (k) is the SOC value corrected by the reverse ampere-hour integration method at time step k, P is the total number of charging cycle steps, I is the current, ΔT represents the sampling interval, C is the nominal capacity of the battery, SOH is the health state of the battery, and η is the coulombic efficiency.

[0020] As a further description of the above technical solution, the SOC estimation model includes a convolutional neural network independent learner, a gated regression unit independent learner and a perceptron meta-learner; the convolutional neural network independent learner is used to extract battery operation time series data features from the input data; the gated regression unit independent learner is used to automatically capture the long-term dependency between battery operation data and SOC from the input data; the perceptron meta-learner is used to obtain cloud-based estimated SOC data based on the battery operation time series data features and the long-term dependency between battery operation data and SOC.

[0021] As a further description of the above technical solution, the convolutional neural network independent learner uses a one-dimensional CNN layer to perform a convolution operation on the input data to extract the first feature data, and then uses a pooling layer to filter out the second feature data from the first feature data, and then uses a convolution layer to extract the third feature data from the second feature data; finally, the third feature data is nonlinearly transformed and dimensionally reduced through a multi-layer fully connected layer, and the battery operating sequence data characteristics are output.

[0022] As a further description of the above technical solution, the gated regression unit independent learner includes several GRU layers and several fully connected layers.

[0023] As a further description of the above technical solution, the perceptron meta-learner includes multiple fully connected layers.

[0024] As a further description of the above technical solution, step 4 includes:

[0025] Step 4.1, sampling the time series of current, voltage, temperature and accumulated ampere-hour in the sliding time window during the charging process of the fully charged single battery, wherein the accumulated ampere-hour is calculated as follows:

[0026]

[0027] Wherein, AC(k) is the ampere-hour of the charging process from charging time t1 to the kth time step, and charging time t1 is the moment after a period of time since the start of charging.

[0028] Step 4.2, standardizing the time series of the current, voltage, temperature and accumulated ampere-hours using a Z-Score method to obtain a standardized time series;

[0029] Step 4.3, based on the SOC estimation model, predict the SOC value of the last time step of the standardized time series and use it as the SOC label of the standardized time series:

[0030]

[0031] Among them, SOC(k) represents the SOC estimated by the SOC estimation model at the kth time step, U(k-m+1, k-m+2, …, k), I(k-m+1, k-m+2, …, k), T(k-m+1, k-m+2, …, k) and AC(k-m+1, k-m+2, …, k) represent the time series of voltage, current, temperature and ampere-hour of the charging process from the k-m+1th to the kth step, m represents the length of the sliding window, and the data is sampled through a window of fixed step size sliding along the time direction. The SOC corresponding to the last moment of the sample is the model output label, and N is the length of the valid data;

[0032] Step 4.4, calculate the SOC of the battery discharge process, the calculation formula is:

[0033]

[0034] Wherein, SOC(k+p) is the battery SOC of the p-time step of the discharge process.

[0035] As a further description of the above technical solution, step 5 includes:

[0036] Step 5.1, using remote communication to transmit the cloud estimated SOC data to the user-side battery management system;

[0037] Step 5.2, the user-side lithium battery management system calibrates its estimated battery SOC using the cloud-based estimated SOC data;

[0038] Step 5.3: The user-side lithium battery management system calculates its SOC using the ampere-hour integration method, and also automatically calibrates its SOC when the full charge or full discharge conditions are met.

[0039] In the second aspect, the present invention provides a lithium battery SOC end-cloud collaborative deep learning estimation system. When the system is working, the cloud is connected to the user-end battery management system through wireless communication. The user-end battery management system sends the battery voltage, current, and temperature data to the cloud, and executes the lithium battery SOC end-cloud collaborative deep learning estimation method described in the first aspect to achieve accurate battery SOC estimation.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] 1) The method of the present invention realizes automatic labeling of SOC by automatically mining real-world battery operation data. There is no need to conduct a large number of battery-specific experiments to generate data samples. It can effectively solve the problems of high cost and long cycle of sample data acquisition, especially the problem of difficulty in fully simulating the real-world battery working environment conditions under laboratory conditions.

[0042] 2) The present invention introduces a convolutional gated recurrent unit meta-learning (CGRM) model, which can automatically and fully extract the battery operation timing data features through a multi-layer CNN learner, and automatically capture the long-term dependency between battery operation data and SOC through a multi-layer GRU learner. Finally, the outputs of the two learners are integrated through a meta-learner to achieve accurate cloud-based estimation of the SOC of lithium-ion batteries.

[0043] 3) The present invention proposes a new feature description method for characterizing battery SOC. It introduces cumulative power characteristics based on traditional current, voltage and temperature characteristics, which can effectively improve the accuracy of SOC estimation.

[0044] 4) Through end-cloud collaboration, the present invention can not only realize accurate estimation of battery SOC on the cloud side, but also realize accurate estimation of battery SOC on the user side, effectively supporting intelligent cloud-based operation and maintenance of batteries, while ensuring the efficiency and safety of batteries on the user's local side.

[0045] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the embodiments of the present invention are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0047] Figure 1 It is an implementation flow chart of the method described in the embodiment of the present invention.

[0048] Figure 2 It is a schematic diagram of the CGRM model framework and its main components according to an embodiment of the present invention.

[0049] Figure 3 4 is a diagram showing the SOC estimation error distribution result of an embodiment of the present invention.

[0050] Figure 4 It is a comparison chart of SOC estimation error indicators of the embodiment of the present invention and three existing methods. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0052] Reference Figure 1 As shown, an embodiment of the present invention provides a lithium battery SOC end-cloud collaborative deep learning estimation method, comprising the following steps:

[0053] Step 1: Clean the real-world user-end lithium battery operation data collected in the cloud to obtain effective charging cycle process data, including battery charging voltage, current and temperature data.

[0054] Step 2: Search for fully charged cells in the cloud, and use the reverse ampere-hour integration method to correct the SOC in the cloud based on the effective charging cycle process data to obtain self-supervised SOC estimation training data, and realize automatic SOC labeling of battery charging process data, as follows:

[0055] Step 2.1, automatically searching for a fully charged single battery that meets the cut-off voltage in the effective charging cycle process data, and setting the SOC of the fully charged single battery to 100%;

[0056] Step 2.2, calculate the SOC value at other times in the charging cycle of the searched fully charged single battery by the reverse ampere-hour integration method, realize automatic labeling of SOC, and obtain SOC estimation training data after the automatic labeling is completed;

[0057] The SOC value calculation formula is:

[0058]

[0059] Among them, SOC r (k) is the SOC value corrected by the reverse ampere-hour integration method at k time steps, P is the total number of charging cycle steps, I is the current (positive for charging and negative for discharging), ΔT represents the sampling interval, C is the nominal capacity of the battery, SOH is the health state of the battery, and η is the coulombic efficiency. Since the current of the battery charging process is stable and the SOC at the full charge point is accurate, the cumulative error of the SOC calculated by the ampere-hour integration method in the current charging cycle can be ignored, and the corrected SOC can be used as the real SOC of the charging process, thereby realizing the automatic labeling of the SOC of the charging process data.

[0060] Step 3: Construct a SOC estimation model based on a convolutional gated recurrent unit meta-learning network (CGRM).

[0061] like Figure 2 As shown, the SOC estimation model includes a convolutional neural network independent learner, a gated regression unit (GRU) independent learner, and a perceptron meta-learner.

[0062] The convolutional neural network independent learner is used to extract battery operation sequence data features from the input data. The convolutional neural network independent learner uses a one-dimensional CNN layer to perform convolution operations on the input data to extract features from the input data layer; then uses a pooling layer to filter out important features; then uses a convolution layer to further extract important features; finally, through two layers of fully connected layers, further nonlinear transformation and dimensionality reduction are performed to output the final SOC prediction result of the learner, that is, the battery operation sequence data features.

[0063] The GRU independent learner is used to automatically capture the long-term dependency between battery operation data and SOC from the input data. The GRU independent learner includes two layers of GRU layers and two layers of fully connected neural networks. Among them, the latter layer of GRU uses the prediction results of the previous layer of GRU to further predict the trend, thereby capturing more complex and deep time domain features.

[0064] The perceptron meta-learner is used to obtain the cloud-based estimated SOC data based on the battery operation sequence data characteristics and the long-term dependency between battery operation data and SOC. The perceptron meta-learner consists of two layers of fully connected neural networks to integrate the outputs of two independent learners.

[0065] Step 4: Using the SOC estimation training data, a sliding window method is used to generate the feature time series required for SOC estimation, and the time series and its corresponding automatic label SOC value are input into the SOC estimation model to obtain the cloud-based estimated SOC data, as follows:

[0066] Step 4.1, sampling the time series of current, voltage, temperature and accumulated ampere-hour in the sliding time window during the charging process of the fully charged single battery, wherein the accumulated ampere-hour is calculated as follows:

[0067]

[0068] Among them, AC(k) is the ampere-hour of the charging process from charging time t1 to the kth time step. In order to eliminate the influence of SOC difference and battery polarization problem at the beginning of charging and improve the regularity of characteristics, the charging time t1 is defined as the moment after a period of time (such as 10 minutes) from the start of charging.

[0069] Step 4.2, the time series of current, voltage, temperature and accumulated ampere-hours are standardized using the Z-Score method to obtain standardized time series.

[0070] Step 4.3, based on the SOC estimation model, predict the SOC value of the last time step of the standardized time series and use it as the SOC label of the standardized time series:

[0071]

[0072] Wherein, SOC(k) represents the SOC estimated by the SOC estimation model at the kth time step, U(k-m+1, k-m+2, …, k), I(k-m+1, k-m+2, …, k), T(k-m+1, k-m+2, …, k) and AC(k-m+1, k-m+2, …, k) represent the time series of voltage, current, temperature and ampere-hour of the charging process from the k-m+1th to the kth step, m=90 represents the length of the sliding window, and the data is sampled through a window of fixed step size sliding along the time direction. The SOC corresponding to the last moment of the sample is the model output label, and N is the length of the valid data;

[0073] Step 4.4, the SOC of the battery discharge process takes the estimated SOC of the charging process as the starting point, and the SOC of the battery discharge process is calculated by the ampere-hour integration method. The calculation formula is:

[0074]

[0075] Wherein, SOC(k+p) is the battery SOC of the p time step of the discharge process, and SOC(k) is the SOC of the charging process estimated by the CGRM.

[0076] Step 5: transmit the cloud-estimated SOC data to the user-side lithium battery management system, calibrate the battery SOC estimated by the user-side lithium battery management system, and achieve accurate estimation of the lithium battery SOC at the user side, as follows:

[0077] Step 5.1, using remote communication to transmit the cloud-estimated SOC to the user-side battery management system BMS;

[0078] Step 5.2, the user-side BMS calibrates its estimated battery SOC using the SOC estimated by the cloud;

[0079] In step 5.3, the user-side BMS calculates its SOC using the ampere-hour integration method. At the same time, it also automatically calibrates its SOC when the full charge or full discharge conditions are met.

[0080] The present invention realizes the end-cloud collaborative deep learning estimation of lithium battery SOC with automatic sample generation. The method has shown excellent lithium-ion battery SOC effect in the test. The estimated SOC under different conditions is highly consistent with the actual SOC. Figure 3 Further tests show that the lithium-ion battery SOC estimation results of the method and system of the present invention are significantly better than other currently most advanced methods (for example, methods using convolutional neural networks (CNNs), methods using gated regression units (GRUs), and even methods using CNN and GRU cascade networks (CNN-GRU)). Detailed performance comparisons are shown in the attached Figure 4This embodiment uses deep learning to achieve accurate SOC estimation of lithium-ion batteries in the real world both in the cloud and on the user side. It does not require specialized experiments on a large number of batteries to generate data samples, and can effectively solve the problems of high cost and long cycle for obtaining sample data, especially the problem of difficulty in fully simulating the working environment of real-world batteries under laboratory conditions.

[0081] In view of the key role of battery SOC estimation in energy storage systems, battery power supplies, electric vehicles, etc., the present invention has important practical value, and is particularly suitable for ensuring the safety of use and improving the energy utilization rate of lithium-ion battery systems, supporting the safe and efficient application and intelligent operation and maintenance of lithium-ion batteries.

[0082] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A lithium battery SOC end-cloud collaborative deep learning estimation method, characterized in that: The following steps are involved: Step 1: Clean the real-world user-end lithium battery operation data collected in the cloud to obtain effective charging cycle process data, including battery charging voltage, current and temperature data; Step 2, according to the effective charging cycle process data, correcting the SOC using a reverse ampere-hour integration method in the cloud, and generating self-supervised SOC estimation training data by automatically labeling the SOC; Step 3, construct a SOC estimation model based on a convolutional gated recurrent unit meta-learning network; Step 4: Using the SOC estimation training data, a sliding window method is used to generate the feature time series required for SOC estimation, and the time series and its corresponding automatic label SOC value are input into the SOC estimation model to obtain the cloud-based estimated SOC data; Step 5: Transmit the cloud-estimated SOC data to the user-side lithium battery management system, calibrate the battery SOC estimated by the user-side lithium battery management system, and achieve accurate estimation of the lithium battery SOC at the user side.

2. The lithium battery SOC end-cloud collaborative deep learning estimation method according to claim 1 is characterized in that: The step 2 comprises: Step 2.1, automatically searching for a fully charged single battery that meets the cut-off voltage in the effective charging cycle process data, and setting the SOC of the fully charged single battery to 100%; Step 2.2, calculate the SOC value at other times in the charging cycle of the searched fully charged single battery by the reverse ampere-hour integration method, realize automatic labeling of SOC, and obtain SOC estimation training data after the automatic labeling is completed; The SOC value calculation formula is: Among them, SOC r (k) is the SOC value corrected by the reverse ampere-hour integration method at time step k, P is the total number of charging cycle steps, I is the current, ΔT represents the sampling interval, C is the nominal capacity of the battery, SOH is the health state of the battery, and η is the coulombic efficiency.

3. The lithium battery SOC end-cloud collaborative deep learning estimation method according to claim 2 is characterized in that: The SOC estimation model includes a convolutional neural network independent learner, a gated regression unit independent learner and a perceptron meta-learner; the convolutional neural network independent learner is used to extract battery operation time series data features from input data; the gated regression unit independent learner is used to automatically capture the long-term dependency between battery operation data and SOC from input data; the perceptron meta-learner is used to obtain cloud-based estimated SOC data based on the battery operation time series data features and the long-term dependency between battery operation data and SOC.

4. The lithium battery SOC end-cloud collaborative deep learning estimation method according to claim 3 is characterized in that: The convolutional neural network independent learner uses a one-dimensional CNN layer to perform a convolution operation on the input data to extract the first feature data, and then uses a pooling layer to filter out the second feature data from the first feature data, and then uses a convolution layer to extract the third feature data from the second feature data; finally, the third feature data is nonlinearly transformed and dimensionally reduced through a multi-layer fully connected layer, and the battery operation sequence data characteristics are output.

5. The lithium battery SOC end-cloud collaborative deep learning estimation method according to claim 3 is characterized in that: The gated regression unit independent learner includes several GRU layers and several fully connected layers.

6. The lithium battery SOC end-cloud collaborative deep learning estimation method according to claim 3 is characterized in that: The perceptron meta-learner includes a plurality of fully connected layers.

7. The lithium battery SOC end-cloud collaborative deep learning estimation method according to claim 2 is characterized in that: The step 4 comprises: Step 4.1, sampling the time series of current, voltage, temperature and accumulated ampere-hour in the sliding time window during the charging process of the fully charged single battery, wherein the accumulated ampere-hour is calculated as follows: Wherein, AC(k) is the ampere-hour of the charging process from charging time t1 to the kth time step, and charging time t1 is the time after a period of time from the start of charging; Step 4.2, standardizing the time series of the current, voltage, temperature and accumulated ampere-hours using a Z-Score method to obtain a standardized time series; Step 4.3, based on the SOC estimation model, predict the SOC value of the last time step of the standardized time series and use it as the SOC label of the standardized time series: Among them, SOC(k) represents the SOC estimated by the SOC estimation model at the kth time step, U(k-m+1,k-m+2,…,k), I(k-m+1,k-m+2,…,k), T(k-m+1,k-m+2,…,k) and AC(k-m+1,k-m+2,…,k) represent the time series of voltage, current, temperature and ampere-hour of charging process from the k-m+1th to the kth step, m represents the length of the sliding window, and the data is sampled through a window of fixed step size sliding along the time direction. The SOC corresponding to the last moment of the sample is the model output label, and N is the length of the valid data; Step 4.4, calculate the SOC of the battery discharge process, the calculation formula is: Wherein, SOC(k+p) is the battery SOC of the p-time step of the discharge process.

8. The lithium battery SOC end-cloud collaborative deep learning estimation method according to claim 1 is characterized in that: The step 5 comprises: Step 5.1, using remote communication to transmit the cloud estimated SOC data to the user-side battery management system; Step 5.2, the user-side lithium battery management system calibrates its estimated battery SOC using the cloud-based estimated SOC data; Step 5.3: The user-side lithium battery management system calculates its SOC using the ampere-hour integration method, and also automatically calibrates its SOC when the full charge or full discharge conditions are met.

9. A lithium battery SOC end-cloud collaborative deep learning estimation system, characterized in that: When the system is working, the cloud is connected to the user-side battery management system through wireless communication. The user-side battery management system sends the battery voltage, current, and temperature data to the cloud, and executes the lithium battery SOC end-cloud collaborative deep learning estimation method described in any one of claims 1 to 8 to achieve accurate battery SOC estimation.

Images