A battery life prediction method based on AI deep learning

Abstract

The present invention relates to the field of artificial intelligence technology, and specifically discloses a battery life prediction method based on AI deep learning. First, the battery SOH is accurately estimated through a battery-based physical parameter model to obtain an SOH estimation value, and then the SOC is accurately estimated through a battery-based hybrid model to obtain an SOC estimation value. Finally, the remaining useful life is predicted using an AI / ML deep learning architecture based on the SOH estimation value and the SOC estimation value to obtain a complete battery prediction life cycle. The RUL is estimated by estimating a reference energy value calculated using a physical model-based method, thereby improving the accuracy and precision of the RUL estimation. The second-layer AI / ML hybrid algorithm tracks parameters that change over time and applies intelligence to understand how quickly it will degrade or reach its service life, providing insights into driving patterns, battery operating time, and cruising range, and allowing wireless flash calibration updates to maximize customer satisfaction.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and specifically relates to a battery life prediction method based on AI deep learning. Background Art

[0002] Pure electric vehicles (BEV), including extended-range hybrid (REEV) / plug-in hybrid (PHEV), and hybrid (HEV), are the future of mobile transportation, and countries around the world are aiming to promote higher penetration rates of electric vehicles every year. Among them, the battery is the most expensive and important component, accounting for 30-50% of the total cost of electric vehicles. It is very obvious that the operation and operating conditions of the vehicle not only affect the mileage of the vehicle (battery), but also the service life of the battery. Due to insufficient charging infrastructure and limited battery capacity, mileage anxiety has always been a key issue that has led to the lack of widespread acceptance or hesitation of electric vehicles. At the same time, supporting long-term battery warranty, high costs for customers to replace faulty batteries, lack of charging infrastructure, and low reliability have also been major problems that have plagued the industry. Solving these problems has always been the focus of OEMs and the automotive industry. Similar concerns have also led large rental fleet management companies to abandon their plans to convert to electric vehicle fleets, or different companies have tried to launch battery replacement models.

[0003] Considering that batteries are the key to facilitating the perfect operation of electric vehicles, their performance must be meticulously measured and carefully scrutinized to increase energy density, reduce costs, and extend service life. Therefore, accurate estimation of parameters such as battery health (SOH) and state of charge (SOC) becomes crucial.

[0004] SOC is a measure of remaining battery capacity and is very important in vehicle applications such as charging control and mileage prediction. It is well known that inaccurate SOC estimation will cause user dissatisfaction, lead to malfunctions due to battery consumption, cause poor vehicle performance and premature battery failure, etc. Therefore, SOC accuracy becomes a key parameter to understand the current state of the battery. The traditional SOC estimation method is a model-based method that first establishes an electrochemical model or an equivalent physical model, and then uses the Kalman filter algorithm for estimation, that is, to find the optimal state value of the battery from the error-filled data. The advantage of this method is that it can minimize the estimation error caused by measurement noise. The disadvantage is that there is no particularly effective solution for how to best use them to extend the life of the battery and how long to solve such a problem. Therefore, we need to propose a battery life estimation method based on AI deep learning to improve the accuracy of SOC estimation. Summary of the invention

[0005] The purpose of the present invention is to provide a battery life estimation method based on AI deep learning, which estimates the RUL by adopting a reference energy value calculated by a physical model-based method, thereby improving the estimation accuracy and precision of the RUL. The second layer of the AI / ML hybrid algorithm tracks parameters that change over time and applies intelligence to understand how quickly it will degrade or reach its service life, provides insights into driving patterns, battery operating time and cruising range, and allows wireless flash calibration updates to maximize customer satisfaction to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] A battery life estimation method based on AI deep learning includes the following steps:

[0008] S1. Accurately estimate the battery SOH through the SOH module based on the physical parameter model of the battery to obtain the SOH estimation value;

[0009] S2. Accurately estimate the battery SOC based on the battery hybrid model to obtain the SOC estimation value;

[0010] S3. Use AI / ML deep learning architecture to predict the remaining useful life based on the SOH estimation value and SOC estimation value to obtain the complete battery prediction life cycle.

[0011] Preferably, in step S1, the physical parameter model is an equivalent physical model based on voltage, charge and discharge current and internal resistance parameters, and the SOH module adopts a parameter-driven method or a data-driven method to calculate the SOH when estimating the battery SOH, wherein the parameter-driven method is a feature-based prediction that utilizes the evolution of characteristic parameters exhibited during the battery aging process to establish a corresponding relationship between the characteristic quantity and the battery life for life prediction; the data-driven method utilizes the test data of battery performance to mine the law of battery performance evolution from the data for life prediction.

[0012] Preferably, the steps of the SOH module using a parameter-driven method to estimate SOH are as follows:

[0013] A1. Measure the impedance spectrum curve at different stages of the battery cycle life to obtain the battery equivalent circuit model form;

[0014] A2. Analyze the influence of solution resistance, charge transfer resistance and Warburg impedance parameters in the equivalent circuit model of the cycle number according to the battery equivalent circuit model form, and obtain the estimated parameters;

[0015] A3. Based on the estimated parameters, a fitting formula is given for the change of each parameter in the equivalent circuit model with the number of battery cycles;

[0016] A4. Calculate the battery cycle life according to the fitting formula.

[0017] Preferably, the steps of the SOH module using a data-driven method to estimate SOH are as follows:

[0018] B1. Establishment of neural network model, which includes input layer, hidden layer and output layer;

[0019] B2, SOH tag data acquisition, battery SOH is calculated through equivalent circuit model parameter changes or battery capacity changes;

[0020] B3. Training data acquisition and objective function. Battery aging factors include SOC, load, ambient temperature, humidity, internal factors of chemical reaction and external factors. List the influencing factors of internal and external factors and evaluate the remaining service life of the battery.

[0021] B4. Estimate verification results.

[0022] Preferably, in step 2, the hybrid model includes a data input module, a time interval calculation module, a charge and discharge rate and current sign calculation module, an SOC estimation module and a result output module, and the data input module, the time interval calculation module, the charge and discharge rate and current sign calculation module and the result output module are all electrically connected to the SOC estimation module.

[0023] Preferably, the data input module includes a time input unit, a voltage input unit, a current input unit, a capacity input unit and a temperature input unit, and the time input unit, voltage input unit, current input unit, capacity input unit and temperature input unit are connected in parallel, and the time input unit, voltage input unit, current input unit, capacity input unit and temperature input unit are all connected to a module for calculating a time interval, a module for calculating a charge and discharge circuit rate and a current sign, and a SOC estimation module.

[0024] Preferably, the SOC estimation module first performs offline identification and then performs online calculation to obtain the SOC estimation value, wherein the offline identification is based on the calibration of the capacity test result to obtain the battery capacity Q st Based on the HPPC test results, the OVC, R0, R1 and C1 of the battery at different SOCs and temperatures are calibrated, and the table lookup function [OCV, R0, R1, C1] = f(SOC, T) is obtained.

[0025] Preferably, the steps of online calculation of the SOC estimation module are as follows:

[0026] C1. Generate model prior values:

[0027]

[0028] Lookup table:

[0029] in:

[0030] U 1，k =U 1，k-1 α k-1 -R 1，k-1 I k-1 ·(1-α k-1 )

[0031] V mdl，k =OCV mdl，k +I k ·R 0，k -U 1，k

[0032] C2, generation error: E k =V exp，k -V mdl，k ;

[0033] C3. Generate feedback gain coefficient using Kalman filtering principle:

[0034]

[0035] A k-1 =1

[0036]

[0037]

[0038]

[0039]

[0040] C4, SOC posterior correction:

[0041]

[0042] Preferably, in step S3, a prerequisite for remaining useful life prediction is to build an end-to-end, powerful and scalable cloud-based analysis platform, which covers the complete battery prediction life cycle from data collection to remaining useful life prediction.

[0043] Preferably, the AI / ML deep learning architecture includes a RUL model based on SOH estimation values ​​and SOC estimation values, and the RUL model analyzes temperature, customer driving and battery usage patterns, charging status and other relevant parameter data collected from the BMS, and the generated analysis results can help predict battery failure, predict and diagnose SOC, SOH and RUL, and enable OEMs to perform wireless software updates.

[0044] The battery life prediction method based on AI deep learning proposed in the present invention has the following advantages compared with the prior art:

[0045] 1. The present invention firstly accurately estimates the battery SOH through a battery-based physical parameter model to obtain an estimated SOH value, then accurately estimates the SOC through a battery-based hybrid model to obtain an estimated SOC value, and finally uses an AI / ML deep learning architecture to predict the remaining useful life according to the estimated SOH value and the estimated SOC value to obtain a complete battery predicted life cycle. The RUL is estimated by using a reference energy value calculated using a physical model-based method, thereby improving the accuracy and precision of the RUL estimate. The second-layer AI / ML hybrid algorithm tracks parameters that change over time and applies intelligence to understand how quickly it will degrade or reach its service life, providing insights into driving patterns, battery operating time, and range, and allowing wireless flash calibration updates to maximize customer satisfaction.

[0046] 2. The present invention accurately estimates the battery SOH through the SOH module and the battery SOC based on the battery hybrid model, receives input parameter data from the battery monitoring system through the physical parameter model and the hybrid model, and estimates the RUL value by combining the physical model ML. These RUL values ​​are finally shared with the user through the dashboard in the form of relevant inputs and parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a flowchart of the present invention;

[0048] Figure 2 The present invention is a system flow chart for calculating the remaining useful life (RUL) of a vehicle (battery);

[0049] Figure 3 A conceptual diagram of the architecture of the RUL estimation model of the present invention. DETAILED DESCRIPTION

[0050] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. The specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0051] The present invention provides Figure 1 A battery life estimation method based on AI deep learning is shown, comprising the following steps:

[0052] S1. Accurately estimate the battery SOH through the SOH module based on the physical parameter model of the battery to obtain the SOH estimation value;

[0053] The physical parameter model is an equivalent physical model based on voltage, charge and discharge current and internal resistance parameters. The SOH module adopts a parameter-driven method or a data-driven method to calculate the SOH when estimating the battery SOH. The parameter-driven method is a feature-based prediction that utilizes the evolution of characteristic parameters exhibited during the battery aging process to establish a corresponding relationship between the characteristic quantity and the battery life for life prediction; the data-driven method utilizes the test data of battery performance to mine the law of battery performance evolution from the data for life prediction.

[0054] The steps of the SOH module using the parameter-driven method to estimate SOH are as follows:

[0055] A1. Measure the impedance spectrum curve at different stages of the battery cycle life to obtain the battery equivalent circuit model form; the impedance spectrum curve is a small-amplitude sinusoidal voltage signal with a frequency of w1 applied to the battery system, and the system generates a sinusoidal current response with a frequency of w2. The change in the ratio of the excitation voltage to the response current is the impedance spectrum of the electrochemical system. The impedance spectrum includes an ultra-high frequency part (the intersection of the impedance curve and the horizontal axis, the ohmic impedance Rb), a high frequency part (semicircle, the impedance of lithium ions passing through the solid electrolyte Rsei), a medium frequency part (semicircle, charge transfer impedance, also known as electrode polarization impedance Rct) and a low frequency part (45° straight line, lithium ion diffusion impedance, also known as concentration polarization impedance W).

[0056] A2. Analyze the influence of solution resistance, charge transfer resistance and Warburg impedance parameters in the equivalent circuit model of the cycle number according to the form of the battery equivalent circuit model to obtain the estimated parameters. The equivalent circuit model simplifies the battery into a circuit system, corresponding to the impedance components of each frequency in the impedance spectrum. The circuit system contains an ohmic resistor, SEI membrane resistor, SEI membrane capacitor, charge transfer resistor and double-layer capacitor. The SEI membrane resistor is connected in parallel with the SEI membrane capacitor, the charge transfer resistor and the double-layer capacitor are connected in parallel, and the ohmic resistor is connected with the parallel SEI membrane resistor and SEI membrane capacitor as well as the charge transfer resistor and the double-layer capacitor. The equivalent circuit model is established in the form of a curve attached to the electrochemical impedance spectrum. The ohmic resistor is less affected by factors such as SOC, temperature, and rate, while the charge transfer resistor and lithium ion diffusion resistor are significantly affected by these factors.

[0057] A3. According to the estimated parameters, a fitting formula for the change of each parameter in the equivalent circuit model with the number of battery cycles is given. According to the number of battery cycles and the corresponding capacity change, a linear function relationship is obtained by fitting: y = -0.1x + 667.55, where x represents the number of battery cycles and y represents the capacity change;

[0058] A4. Calculate the battery cycle life according to the fitting formula.

[0059] The steps of the SOH module using a data-driven approach to estimate SOH are as follows:

[0060] B1. Establishment of a neural network model. The neural network model includes an input layer, a hidden layer and an output layer, where p = (p1, ..., pi, ...) is the input, w = (w1, ..., wi, ...) is the weight, b is the offset, the internal formula is q = wpT + b, a = fa (q) is the scalar output function, and the neural network model uses a single hidden layer. The number of nodes in the hidden layer is the same as the number of nodes in the output layer, otherwise the neural network model is too complex and will lead to overfitting;

[0061] B2, SOH tag data acquisition, battery SOH is calculated through equivalent circuit model parameter changes or battery capacity changes;

[0062] B3. Training data acquisition and objective function. Battery aging factors include SOC, load, ambient temperature, humidity, internal factors of chemical reaction and external factors. List the influencing factors of internal factors and external factors and evaluate the remaining service life of the battery.

[0063] B4. Estimate verification results.

[0064] S2. Accurately estimate the battery SOC based on the battery hybrid model to obtain the SOC estimation value;

[0065] The hybrid model includes a data input module, a time interval calculation module, a charge and discharge rate calculation and current sign calculation module, an SOC estimation module and a result output module. The data input module, the time interval calculation module, the charge and discharge rate calculation and current sign calculation module and the result output module are all electrically connected to the SOC estimation module.

[0066] The data input module includes a time input unit, a voltage input unit, a current input unit, a capacity input unit and a temperature input unit. The time input unit, the voltage input unit, the current input unit, the capacity input unit and the temperature input unit are connected in parallel. The time input unit, the voltage input unit, the current input unit, the capacity input unit and the temperature input unit are all connected to a time interval calculation module, a charge and discharge circuit rate calculation and current sign calculation module and an SOC estimation module. The time input unit senses time through a time sensor, the voltage input unit senses voltage through a voltage monitoring sensor, the current input unit senses voltage through a current sensor, the capacity input unit senses voltage through a capacitive sensor, and the temperature input unit senses voltage through a temperature sensor.

[0067] For example, the module for calculating the time interval first calculates the discharge current of the battery at a 10-hour rate, that is, the capacity divided by 10. For a group of 500AH batteries, the 10-hour rate discharge current is 50A, and for two groups of 500AH, the 10-hour rate discharge current is 100A; then the actual discharge current is divided by the 10-hour rate discharge current to calculate a ratio, and the discharge rate in the "Battery Discharge Rate and Discharge Capacity" table is queried based on this ratio. From this discharge rate, the closest value is selected, and the corresponding discharge rate and effective discharge capacity resolution are seen. The data in the table is recorded, and the discharge environment temperature at that time is checked. Finally, the discharge duration is calculated, and the charge and discharge time interval is derived;

[0068] The calculation formula of the charge and discharge rate and current sign module is: charge and discharge rate = charge and discharge current / rated capacity. The discharge rate is a measure of the discharge speed. The capacity used is discharged in 1 hour is called 1C discharge. If it is discharged in 5 hours, the second is 1 / 5 = 0.2C discharge. Generally, the battery capacity can be detected by different discharge currents.

[0069] The SOC estimation module first performs offline identification and then performs online calculation to obtain the SOC estimation value, wherein the offline identification is based on the capacity test result to calibrate the battery capacity Q st Based on the HPPC test results, the OVC, R0, R1 and C1 of the battery at different SOCs and temperatures are calibrated, and the table lookup function [OCV, R0, R1, C1] = f(SOC, T) is obtained.

[0070] The steps of online calculation of the SOC estimation module are as follows:

[0071] C1. Generate model prior values:

[0072]

[0073] Lookup table:

[0074] in:

[0075] U 1，k =U 1，k-1 α k-1 -R 1，k-1 I k-1 ·(1-α k-1 )

[0076] V mdl，k =OCV mdl，k +I k ·R 0，k -U 1，k

[0077] C2, generation error: E k =V exp，k -V mdl，k ;

[0078] C3. Generate feedback gain coefficient using Kalman filtering principle:

[0079]

[0080] A k-1 =1

[0081]

[0082]

[0083]

[0084]

[0085] C4, SOC posterior correction:

[0086]

[0087] The result output module includes SOC output, model terminal voltage output and measured terminal voltage output.

[0088] S3. Use AI / ML deep learning architecture to predict the remaining useful life based on the estimated SOH and SOC values ​​to obtain the complete battery prediction life cycle. In short, estimating the SOC and SOH of the battery can provide insight into the current state of battery health and energy. Although this is important in itself, it also enables us to go a step further and predict the remaining useful life (RUL) of the battery based on the number of cycles remaining in the battery life. Traditional RUL prediction methods are quite limited. This application proposes an AI-based method to calculate the RUL of the battery based on the estimation of current state parameters such as SOC and SOH. This method takes into account the electrical characteristics of the battery when estimating the SOC, rather than simply relegating it to the status of a black box.

[0089] like Figure 2 As shown, the collection of data from different running vehicles and the real-time analysis of these data are crucial in the method proposed by this invention patent. Therefore, the data is collected, pre-processed, isolated and provided to the AI / ML-based algorithm for calculation, so as to predict the RUL of different vehicles. The prerequisite for the remaining useful life prediction is to build an end-to-end, powerful and scalable cloud-based analysis platform that covers the complete battery prediction life cycle from data collection to remaining useful life prediction.

[0090] The patented method then uses a custom AI / ML-based model to estimate and compare the performance and behavior of the battery and compare it with the results of traditional methods. The model can also be used on embedded controller platforms (chips with sufficient computing power) and has been proven to be much better than traditional models.

[0091] The AI / ML deep learning architecture includes a RUL model based on SOH estimation and SOC estimation. The RUL model analyzes temperature, customer driving and battery usage patterns, charging status and other relevant parameter data collected from the BMS. The generated analysis results can help predict battery failure, predict and diagnose SOC, SOH and RUL, and enable OEMs to perform wireless software updates.

[0092] The application of AI and ML opens up a variety of possibilities for electric vehicle batteries. The patented method has a proven ability to develop integrated cloud-based AI / ML models that analyze temperature, customer driving and battery usage patterns, charge status and other relevant parameter data collected from the BMS. The resulting analysis can help predict battery failures, predictive diagnose SOC, SOH and RUL, and enable OEMs to address these issues through wireless software upgrades (OTA) updates.

[0093] In addition to temperature, voltage, and current, the RUL model also uses some estimates generated by traditional tools. At the same time, some estimates are also generated using AI / ML and then input into the RUL model.

[0094] like Figure 3 As shown, the RUL is estimated by using a reference energy value calculated using a physical model-based approach. While this approach helps improve RUL estimation, the hybrid approach greatly improves the accuracy and precision of RUL estimation. The second layer of AI / ML hybrid algorithms tracks parameters that change over time and applies intelligence to understand how quickly it will degrade or reach the end of its service life. The SOC and SOH modules receive input parameter data from the battery monitoring module and estimate the RUL value by combining it with a physical model-based ML. These RULs are ultimately shared with the user through the dashboard in the form of relevant inputs / parameters.

[0095] In summary, the present invention can optimize the fleet management of large rental fleets and form an operating mechanism that improves cost-effectiveness, while OEMs and insurance companies can benefit from reduced warranty claims - and can solve battery size design and potential warranty issues as early as possible before production, or just at the beginning and design stage. Accurate RUL prediction can also help customers understand the total cost of ownership of electric vehicles - in short, the longer the RUL time, the higher the cost / benefit ratio for customers, and the higher the benefits to customers.

[0096] The complexity of the RUL processing algorithms limits its use in the embedded controllers of conventional BEV / PHEVs. Therefore, the present invention proposes a cloud-based solution where vehicle data is analyzed, filtered and grouped to enable seamless algorithmic operations. These algorithms further provide insights into driving patterns, battery runtime and range, and allow over-the-air (OTA) calibration updates to maximize customer satisfaction, such as improved fast charging and improved range.

[0097] This application uses AI / ML-based solutions to accurately calculate RUL, and the benefits are reflected in strengthening warranty costs, reducing vehicle recalls, and providing methods and foresight to extend battery life, thereby optimizing the return on investment of users, OEMs, insurers and other stakeholders. It is also worth mentioning that in this patented method, the model with a self-learning neural network can regularly retrain the model using larger data to improve its applicability.

[0098] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A battery life prediction method based on AI deep learning, characterized by: The steps include: S1. Accurately estimate the battery SOH through the SOH module based on the physical parameter model of the battery to obtain the SOH estimation value; The physical parameter model is an equivalent physical model based on voltage, charge and discharge current and internal resistance parameters. The SOH module uses a parameter-driven method or a data-driven method to calculate the SOH when estimating the battery SOH. The parameter-driven method is a feature-based prediction that uses the evolution of characteristic parameters exhibited during the battery aging process to establish a corresponding relationship between the characteristic quantity and the battery life for life prediction; the data-driven method uses the test data of battery performance to mine the law of battery performance evolution from the data for life prediction; S2. Accurately estimate the battery SOC based on the battery hybrid model to obtain the SOC estimation value; The hybrid model includes a data input module, a time interval calculation module, a charge and discharge rate calculation module and a current sign calculation module, an SOC estimation module and a result output module, and the data input module, the time interval calculation module, the charge and discharge rate calculation module and the current sign calculation module and the result output module are all electrically connected to the SOC estimation module; S3, use AI / ML deep learning architecture to predict the remaining useful life based on the SOH estimation value and SOC estimation value to obtain the complete battery prediction life cycle; The steps of online calculation of the SOC estimation module are as follows: C1. Generate model prior values: Lookup table: in: U 1，k =U 1，k-1 ·a k-1 -R 1，k-1 ·I k-1 ·(1-a k-1 ) V mdl，k =OCV mdl，k +I k ·R 0，k -U 1，k C2, generation error: E k =V exp，k -V mdl，k ; C3. Generate feedback gain coefficient using Kalman filtering principle: A k-1 ＝1 C4, SOC posterior correction:

2. The method for battery life estimation based on AI deep learning according to claim 1, characterized in that: The steps of the SOH module using the parameter-driven method to estimate SOH are as follows: A1. Measure the impedance spectrum curve at different stages of the battery cycle life to obtain the battery equivalent circuit model form; A2. Analyze the influence of solution resistance, charge transfer resistance and Warburg impedance parameters in the equivalent circuit model of the cycle number according to the battery equivalent circuit model form, and obtain the estimated parameters; A3. Based on the estimated parameters, a fitting formula is given for the change of each parameter in the equivalent circuit model with the number of battery cycles; A4. Calculate the battery cycle life according to the fitting formula.

3. The method for battery life prediction based on AI deep learning according to claim 2, characterized in that: The steps of the SOH module using a data-driven approach to estimate SOH are as follows: B1. Establishment of neural network model, which includes input layer, hidden layer and output layer; B2, SOH tag data acquisition, battery SOH is calculated through equivalent circuit model parameter changes or battery capacity changes; B3. Training data acquisition and objective function. Battery aging factors include SOC, load, ambient temperature, humidity, internal factors of chemical reaction and external factors. List the influencing factors of internal and external factors and evaluate the remaining service life of the battery. B4. Estimate verification results.

4. The method for battery life prediction based on AI deep learning according to claim 3 is characterized in that: The data input module includes a time input unit, a voltage input unit, a current input unit, a capacity input unit and a temperature input unit, which are connected in parallel. The time input unit, the voltage input unit, the current input unit, the capacity input unit and the temperature input unit are all connected to a time interval calculation module, a charge and discharge circuit rate calculation and current sign calculation module and an SOC estimation module.

5. The method for battery life estimation based on AI deep learning according to claim 4, characterized in that: The SOC estimation module first performs offline identification and then performs online calculation to obtain the SOC estimation value, wherein the offline identification is based on the capacity test result to calibrate the battery capacity Q st Based on the HPPC test results, the OVC, R0, R1 and C1 of the battery at different SOCs and temperatures are calibrated, and the table lookup function [OCV, R0, R1, C1] = f(SOC, T) is obtained.

6. The method for battery life prediction based on AI deep learning according to claim 5, characterized in that: In step S3, the prerequisite for remaining useful life prediction is to build an end-to-end, powerful and scalable cloud-based analysis platform, which covers the complete battery prediction life cycle from data collection to remaining useful life prediction.

7. The method for battery life prediction based on AI deep learning according to claim 6, characterized in that: The AI / ML deep learning architecture includes a RUL model based on SOH estimation and SOC estimation. The RUL model analyzes temperature, customer driving and battery usage patterns, charging status and other relevant parameter data collected from the BMS. The generated analysis results can help predict battery failure, predict and diagnose SOC, SOH and RUL, and enable OEMs to perform wireless software updates.

Images