Method and system for monitoring state of lithium ion battery of underwater robot based on hybrid model

Abstract

The invention provides an underwater robot lithium ion battery state monitoring method and system based on a hybrid model, and the method comprises the steps: firstly obtaining the physical parameter data of a lithium ion battery, carrying out the preprocessing, inputting a lithium ion battery life prediction model, and obtaining a lithium ion battery life prediction model; according to the model, a multi-stress-factor model is used for processing voltage, temperature, current, capacitance and external pressure parameters of the lithium ion battery, and the influence of the environment on battery aging is simulated, so that the prediction accuracy is improved; the feature extraction capability is improved by enhancing the random forest model, and the multi-dimensional parameter processing capability is improved; capturing a long-term dependency relationship in the time sequence through a TCN model; and finally, learning an optimal weight combination through an XGBoost weighting model, reducing the deviation and variance of a single model, and further optimizing the overall prediction result. The method overcomes the defects of a traditional lithium ion battery monitoring method in a high-voltage complex environment, is suitable for underwater robots of different models, is easy to integrate and deploy, and has good universality and flexibility.

Description

Technical Field

[0001] The present invention relates to the technical field of battery status monitoring, and in particular to a hybrid model-based method and system for monitoring the status of lithium-ion batteries of underwater robots. Background Art

[0002] With the advancement of industrial technology, underwater robots (AUVs), as advanced marine equipment, are playing an irreplaceable role in several areas, such as marine resource exploration, seafloor topography mapping, and marine scientific research. Due to their high energy density, long cycle life, and low self-discharge rate, AUVs typically rely on lithium-ion batteries as their primary power source, providing stable and long-lasting power. Therefore, real-time monitoring and accurate estimation of the operating status of lithium-ion batteries are crucial to ensuring the safe operation of AUVs.

[0003] Traditional battery state estimation methods can be roughly divided into three different categories: physical model-based methods, data-driven methods, and hybrid model methods. For example, existing studies have proposed using an extended Kalman filter (EKF) combined with a regularized extreme learning machine (RELM) to estimate SOC under temperature changes; however, this method mainly focuses on SOC estimation on land and does not consider other factors such as pressure under the deep sea. Among the data-driven methods, long short-term memory networks (LSTM) and convolutional neural networks (CNN) are widely used. For example, Chinese patent CN119125905A discloses a lithium-ion battery state prediction method based on a neural network model with convolution and bidirectional long short-term memory. This method is highly dependent on a large amount of high-quality training data, and the decision-making process of the neural network model used lacks transparency, which becomes a problem in applications that require explanation and verification of prediction results. At the same time, these studies are not set up for the special operating environment of underwater robots under the deep sea.

[0004] Therefore, existing research still cannot accurately reflect the remaining operating life of underwater robot lithium-ion batteries, especially under extreme operating conditions such as deep-sea high pressure and sudden temperature changes. For example, Chinese patent CN117707211A discloses an underwater robot state estimation method based on multi-sensor data fusion. This method improves the accuracy of state estimation through multi-sensor data fusion, but it also does not consider the influence of water pressure. Summary of the Invention

[0005] Purpose of the invention: The first purpose of the present invention is to provide a method for monitoring the status of lithium-ion batteries of underwater robots based on a hybrid model and in combination with physical parameters of batteries including external pressure; the second purpose is to provide a system for monitoring the status of lithium-ion batteries of underwater robots based on a hybrid model.

[0006] Technical solution: A hybrid model-based method for monitoring the lithium-ion battery status of an underwater robot includes the following steps:

[0007] S1. Collecting physical parameters of the lithium-ion battery of the underwater robot to be tested, the physical parameters including voltage, temperature, capacitance, current of the lithium-ion battery and the pressure value outside the lithium-ion battery of the underwater robot to be tested;

[0008] S2. Preprocessing physical parameters;

[0009] S3. Construct a lithium-ion battery life prediction model and use a standard lithium-ion battery parameter data set to train the lithium-ion battery life prediction model. The lithium-ion battery life prediction model includes a multi-stress factor model, an enhanced random forest model, a TCN model, and an extreme gradient boosting weighted model. The extreme gradient boosting weighted model performs weighted optimization on the outputs of the multi-stress factor model, the enhanced random forest model, and the TCN model. The preprocessed physical parameters are input into the trained lithium-ion battery life prediction model to obtain the lithium-ion battery life prediction result.

[0010] Specifically, in step S1, collecting the physical parameters of the lithium-ion battery of the underwater robot to be tested includes:

[0011] A multi-channel voltage acquisition unit is used to monitor the voltage status of the lithium-ion battery; a distributed optical fiber temperature measurement system is used to monitor the external temperature changes of the lithium-ion battery; a Hall effect sensor ring array is used to monitor the magnetic field strength of the lithium-ion battery and calculate the lithium-ion battery current; a supercapacitor management module is used to detect the capacitance value and capacity of the lithium-ion battery; and a pressure sensor array is used to detect the external pressure value of the lithium-ion battery.

[0012] Specifically, in step S2, the preprocessing includes: performing data cleaning on the physical parameters, deleting noise features and incomplete data, then normalizing the data, and finally performing data enhancement to obtain the preprocessed physical parameters.

[0013] Specifically, in step S3, the construction of the multi-stress factor model includes:

[0014] S31. Construct a mathematical model of the state of health (SOH) and remaining life (RUL) of lithium-ion batteries;

[0015] S32. Based on the mathematical model, a multi-stress factor model is constructed based on the Arrhenius formula:

[0016]

[0017] Where: S(V) is the capacity decay rate, V is the stress factor, A represents the upper limit of the rate of capacity decay of lithium-ion batteries under given conditions, E ais the activation energy, R is the gas constant, P is the pressure value of the external environment, and n is the voltage index, which is used to describe the influence of voltage on capacity decay;

[0018] S33. Input the lithium-ion battery parameter standard data set into the multi-stress factor model for training.

[0019] Specifically, in step S3, the construction of the enhanced random forest model includes: constructing a decision tree based on a sample subset randomly selected from the lithium-ion battery parameter standard data set, each decision tree is trained by randomly selecting features, and weights are assigned according to the accuracy of each decision tree during the verification process. The formula is:

[0020]

[0021] Where: is the output of the predicted i-th input sample, N is the total number of decision trees in the forest, and f n (X i ) is the nth decision tree for input X i The prediction result, ω n is the weight of the nth tree.

[0022] Specifically, in step S3, the TCN model includes the following formula:

[0023]

[0024] h t+1 =εh t v t

[0025] Where: is the convolution value on the a-th row vector and the b-th convolution kernel, E is the convolution kernel, l is the length of the time series, W is the width of the convolution kernel, v t is the new timing information, is the output of the convolutional layer, α a is the weight of the a-th row vector, h t+1 is the final output, h t is the current input, and ε is the output coefficient, which is used to adjust the scale of the final output.

[0026] Specifically, in step S3, the extreme gradient boosting weighted model includes:

[0027] The output results of the input multi-stress factor model, enhanced random forest model, and TCN model are optimized using XGBoost weighted optimization. The formula is as follows:

[0028]

[0029] Where: Obj is the model output, n is the number of samples, y i is the true value of the i-th sample, is the predicted value of the i-th sample, is the loss function, which is used to measure the difference between the predicted value and the true value, K is the number of trees, and f k is the kth tree, Ω(f k ) is a regularization term used to control the complexity of the model.

[0030] Preferably, the above method further comprises the following steps:

[0031] S4. Estimating the state of the lithium-ion battery according to the estimated result of the lithium-ion battery life, and outputting the estimated result of the lithium-ion battery state.

[0032] Specifically, step S4 includes:

[0033] S41. Calculate the time required to execute the task using the underwater robot positioning system. The formula is as follows:

[0034]

[0035] Where: T mission is the time required for task execution; L is the length of the task path; k path is the path correction coefficient, v is the speed of the underwater robot; v current is the water flow velocity; θ is the angle between the water flow and the direction of movement;

[0036] S42. Calculate the remaining mission time of the underwater robot using the estimated result of the lithium-ion battery life. The formula is:

[0037]

[0038] Where: T safe is the remaining mission time of the underwater robot, RUL is the estimated life of the lithium-ion battery, α is the safety factor, 0.7<α<1;

[0039] S43: Output the remaining mission time of the underwater robot as the lithium-ion battery state estimation result.

[0040] The present invention also provides an underwater robot lithium-ion battery status monitoring system based on a hybrid model, comprising:

[0041] Multi-physics field data acquisition module: used to collect the physical parameters of the lithium-ion battery of the underwater robot to be tested, including the voltage, temperature, capacitance, current of the lithium-ion battery and the pressure value outside the lithium-ion battery of the underwater robot to be tested;

[0042] Data preprocessing module: used to preprocess physical parameters;

[0043] Lithium-ion battery life prediction module: used to build a lithium-ion battery life prediction model and train the lithium-ion battery life prediction model using a standard lithium-ion battery parameter dataset. The lithium-ion battery life prediction model includes a multi-stress factor model, an enhanced random forest model, a TCN model, and an extreme gradient boosting weighted model. The extreme gradient boosting weighted model performs weighted optimization on the outputs of the multi-stress factor model, the enhanced random forest model, and the TCN model, and inputs the preprocessed physical parameters into the trained lithium-ion battery life prediction model to obtain the lithium-ion battery life prediction results.

[0044] Beneficial effects: Compared with the existing technology, the significant effects of the present invention are: the present invention aims at the problem of real-time status monitoring of lithium-ion batteries of underwater robots in the special operating environment of the deep sea, and proposes a lithium-ion battery life prediction model. The model is a hybrid model, which obtains the physical parameter data of the lithium-ion battery, uses a multi-stress factor model to process the lithium-ion battery voltage, temperature, current, capacitance and external pressure parameters, processes the influence of environmental factors on battery aging, and corrects the battery aging equation, thereby significantly improving the accuracy of lithium-ion battery life status prediction; enhances the feature extraction capability and the ability to process multi-dimensional parameters by enhancing the random forest model; captures long-term dependencies in time series by using the TCN model; finally, learns the best weight combination by the XGBoost weighted model, reduces the bias and variance of a single model, and further optimizes the overall prediction results. The present invention overcomes the shortcomings of traditional lithium-ion battery monitoring methods in high-pressure and complex environments, and is applicable to underwater robots of different models. It is easy to integrate and deploy, and is convenient for users to upgrade and apply on the basis of existing equipment. It has good versatility and flexibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a flow chart of the method of the present invention.

[0046] Figure 2 It is a flow chart of the lithium ion life prediction model of the present invention. DETAILED DESCRIPTION

[0047] The present invention will be further described below with reference to the accompanying drawings.

[0048] Example 1

[0049] See also Figure 1 As shown, this embodiment provides a method for monitoring the status of lithium-ion batteries of underwater robots based on a hybrid model, comprising the following steps:

[0050] S1. Collect physical parameters of the lithium-ion battery of the underwater robot to be tested, where the physical parameters include voltage, temperature, capacitance, current of the lithium-ion battery and pressure value outside the lithium-ion battery of the underwater robot to be tested.

[0051] In this embodiment, physical parameters are collected by setting an embedded sensor array on the underwater robot. In the embedded sensor array, a multi-channel voltage acquisition unit is used to monitor the voltage status of the lithium-ion battery in real time; a distributed optical fiber temperature measurement system is used to monitor the external temperature changes of the lithium-ion battery in real time and calculate the ambient temperature of the lithium-ion battery; a Hall effect sensor ring array is used to monitor the magnetic field strength around the lithium-ion battery and calculate the lithium-ion battery current; a supercapacitor management module is used to detect the capacitance value and capacity of the lithium-ion battery; and a pressure sensor array is used to detect the external pressure value of the lithium-ion battery in real time.

[0052] S2. Preprocess the physical parameters collected in real time by the embedded sensor array.

[0053] First, the physical parameters are cleaned to eliminate noise features and incomplete data. Then, data conversion is performed to integrate, transform, and normalize the data. In this embodiment, z-score normalization is used:

[0054]

[0055] Where: Z is the normalized data, G is the original data, μ is the mean of the original data, and σ is the standard deviation of the original data.

[0056] Then, in order to weaken the impact of missing or sparse original data, data enhancement is performed on the normalized physical parameter data to obtain a new data set, namely the preprocessed physical parameters.

[0057] S3. Construct a lithium-ion battery life prediction model and use a standard lithium-ion battery parameter data set to train the lithium-ion battery life prediction model. The lithium-ion battery life prediction model includes a multi-stress factor model, an enhanced random forest model, a TCN model, and an extreme gradient boosting (XGBoost) weighted model. The XGBoost weighted model performs weighted optimization on the outputs of the multi-stress factor model, the enhanced random forest model, and the TCN model. The preprocessed physical parameters are input into the trained lithium-ion battery life prediction model to obtain the lithium-ion battery life prediction result.

[0058] To pre-train the model, a standard lithium-ion battery parameter dataset is required. In this embodiment, different parameter data of lithium-ion batteries during operation are collected from the dataset tested by Gulsoy et al. (Gulsoy B, Vincent T, Briggs C, Kalathingal A, Niri MF, Marco J. Dataset of accumulated internal gas pressure and temperature during lithium-ion battery operation and aging. Data Brief. 2025Feb 26; 59:111420. doi:10.1016 / j.dib.2025.111420. PMID:40124283; PMCID:PMC11930082.). The pre-processing method of step S2 is referred to to obtain a standard lithium-ion battery parameter dataset after pre-processing. At the same time, because the parameters need to be kept within the safe operating range of the underwater robot, range constraints are imposed on multiple parameters in the standard lithium-ion battery parameter dataset.

[0059] Please refer to Figure 2 As shown in the figure, the lithium-ion battery life prediction model is a hybrid model, which constructs a multi-stress factor model, an enhanced random forest model, and a TCN model respectively, and then inputs the outputs of the multi-stress factor model, the enhanced random forest model, and the TCN model into the XGBoost weighted model for weighted optimization.

[0060] The construction process of the multi-stress factor model is as follows:

[0061] First, a mathematical model of the health status SOH and remaining life RUL of lithium-ion batteries is constructed:

[0062]

[0063] RUL=f(SOC(t),SOH(t),T,I,U)

[0064] Where: SOC is the remaining capacity, t is the time, SOC(t0) is the initial charging state, Q nominal is the rated capacity of the lithium-ion battery (unit: A·H), I is the instantaneous current, where discharge is positive, charge is negative, and C currrent (t) is the current capacity of the lithium-ion battery at time t, C nominal is the rated capacity of the lithium-ion battery; f is the relationship function between the remaining life RUL of the lithium-ion battery and the health state SOH, remaining charge SOC, temperature T, current I, and voltage U.

[0065] Then, based on the established mathematical models of lithium-ion battery SOH and RUL, a multi-stress factor model was constructed based on the Arrhenius formula:

[0066]

[0067] Where: S(V) is the capacity decay rate, V is the stress factor, A represents the upper limit of the rate of capacity decay of lithium-ion batteries under given conditions, E a is the activation energy, R is the gas constant, P is the pressure value of the external environment, and n is the voltage index, which is used to describe the influence of voltage on capacity decay.

[0068] Multiple sample subsets are randomly selected from the standard lithium-ion battery parameter dataset and input into the multi-stress factor model for training to optimize the model's prediction accuracy.

[0069] The enhanced random forest model construction process is as follows:

[0070] The sample extraction process is as follows: Multiple sample subsets are randomly extracted from a standard dataset of lithium-ion battery parameters. Decision trees are constructed based on these sample subsets. Each decision tree is trained by randomly selecting features to enhance the model's generalization capabilities. Compared to the traditional random forest model, which evenly distributes weights, the enhanced random forest model distributes weights for the accuracy of individual decision trees during training. This process is performed using the following formula:

[0071]

[0072] Where: is the output of the predicted i-th input sample, N is the total number of decision trees in the forest, and f n (X i ) is the nth decision tree for input X i The prediction result, ω n is the weight of the nth tree.

[0073] The present invention introduces the TCN model to capture the long-term dependencies in the aging process of lithium-ion batteries.

[0074] The construction process of the TCN (Temporal Convolutional Network) model is as follows:

[0075]

[0076] h t+1 =εh t v t

[0077] Where: is the convolution value on the a-th row vector and the b-th convolution kernel, E is the convolution kernel, l is the length of the time series, W is the width of the convolution kernel, v t is the new timing information, is the output of the convolutional layer, α a is the weight of the a-th row vector, h t+1 is the final output, h t is the current input, and ε is the output coefficient, which is used to adjust the scale of the final output.

[0078] The three models mentioned above are trained separately using the standard dataset of lithium-ion battery parameters.

[0079] The present invention uses the XGBoost weighted model to perform weighted optimization on the outputs of the above-mentioned multi-stress factor model, enhanced random forest model, and TCN model. The formula is as follows:

[0080]

[0081] Where: Obj is the model output, n is the number of samples, y i is the true value of the i-th sample, is the predicted value of the i-th sample, is the loss function, which is used to measure the difference between the predicted value and the true value, K is the number of trees, and f k is the kth tree, Ω(f k ) is a regularization term used to control the complexity of the model.

[0082] Through weighted optimization of the XGBoost weighted model, the lithium-ion battery life estimation model outputs the final lithium-ion battery life estimation result.

[0083] The preprocessed physical parameters are input into the trained lithium-ion battery life prediction model to obtain the lithium-ion battery life prediction result RUL.

[0084] S4. Estimating the state of the lithium-ion battery according to the estimated result of the lithium-ion battery life, and outputting the estimated result of the lithium-ion battery state.

[0085] First, the time required to execute the task is calculated using the underwater robot GPS positioning and tracking system. The formula is as follows:

[0086]

[0087] Where: T mission is the time required for task execution; L is the length of the task path; k path is the path correction coefficient, a parameter used to adjust the mission path length. Its range is 1.1 to 2 under normal circumstances, and is set to a constant greater than 2 in some extreme detours; v is the speed of the underwater robot; vcurrent is the water flow velocity; θ is the angle between the water flow and the direction of movement, ranging from 0°≤θ≤180°.

[0088] Then, the lithium-ion battery life estimation result is used to calculate the remaining mission time of the underwater robot. The formula is:

[0089]

[0090] Where: T safe is the remaining mission time of the underwater robot, RUL is the estimated life of the lithium-ion battery, and α is the safety factor, which is used to adjust the uncertainty or environmental risk of the RUL prediction to ensure the safety and reliability of mission execution. Its range is generally: 0.7<α<1.

[0091] The remaining mission time of the underwater robot is output as the lithium-ion battery status estimation result. If the remaining mission time of the underwater robot reaches a threshold, a status warning is issued to the underwater robot control end, such as sending the status estimation result and warning information to the system display module of the surface mother ship or the ground control center.

[0092] Example 2

[0093] This embodiment provides a hybrid model-based underwater robot lithium-ion battery status monitoring system, including the following modules:

[0094] Multi-physics field data acquisition module: used to collect physical parameters of the lithium-ion battery of the underwater robot to be tested, including the voltage, temperature, capacitance, current of the lithium-ion battery and the pressure value outside the lithium-ion battery of the underwater robot to be tested;

[0095] Data preprocessing module: used to preprocess physical parameters;

[0096] Lithium-ion battery life prediction module: used to build a lithium-ion battery life prediction model, and use the lithium-ion battery parameter standard data set to train the lithium-ion battery life prediction model. The lithium-ion battery life prediction model includes a multi-stress factor model, an enhanced random forest model, a TCN model and an XGBoost weighted model. The XGBoost weighted model performs weighted optimization on the outputs of the multi-stress factor model, the enhanced random forest model and the TCN model, and inputs the preprocessed physical parameters into the trained lithium-ion battery life prediction model to obtain the lithium-ion battery life prediction result.

Claims

1. A method for monitoring the status of lithium-ion batteries of underwater robots based on a hybrid model, characterized in that: The following steps are involved: S1. Collecting physical parameters of the lithium-ion battery of the underwater robot to be tested, wherein the physical parameters include voltage, temperature, capacitance, current of the lithium-ion battery and pressure value outside the lithium-ion battery of the underwater robot to be tested; S2. Preprocessing physical parameters; S3. Construct a lithium-ion battery life prediction model and train the lithium-ion battery life prediction model using a standard lithium-ion battery parameter data set. The lithium-ion battery life prediction model includes a multi-stress factor model, an enhanced random forest model, a TCN model, and an extreme gradient boosting weighted model. The extreme gradient boosting weighted model performs weighted optimization on the outputs of the multi-stress factor model, the enhanced random forest model, and the TCN model. The preprocessed physical parameters are input into the trained lithium-ion battery life prediction model to obtain a lithium-ion battery life prediction result.

2. The method for monitoring the status of an underwater robot lithium-ion battery according to claim 1, wherein: In step S1, collecting the physical parameters of the lithium-ion battery of the underwater robot to be tested includes: A multi-channel voltage acquisition unit is used to monitor the voltage status of the lithium-ion battery; a distributed optical fiber temperature measurement system is used to monitor the external temperature changes of the lithium-ion battery; a Hall effect sensor ring array is used to monitor the magnetic field strength of the lithium-ion battery and calculate the lithium-ion battery current; a supercapacitor management module is used to detect the capacitance value and capacity of the lithium-ion battery; and a pressure sensor array is used to detect the external pressure value of the lithium-ion battery.

3. The method for monitoring the status of an underwater robot lithium-ion battery according to claim 1, wherein: In step S2, the preprocessing includes: performing data cleaning on the physical parameters, deleting noise features and incomplete data, then normalizing the data, and finally performing data enhancement to obtain the preprocessed physical parameters.

4. The method for monitoring the status of an underwater robot lithium-ion battery according to claim 1, wherein: In step S3, the construction of the multi-stress factor model includes: S31. Construct a mathematical model of the state of health (SOH) and remaining life (RUL) of lithium-ion batteries; S32. On the basis of the mathematical model, a multi-stress factor model is constructed based on the Arrhenius formula: Where: S(V) is the capacity decay rate, V is the stress factor, A represents the upper limit of the rate of capacity decay of lithium-ion batteries under given conditions, E a is the activation energy, R is the gas constant, P is the pressure value of the external environment, and n is the voltage index, which is used to describe the influence of voltage on capacity decay.

5. The method for monitoring the status of an underwater robot lithium-ion battery according to claim 1, wherein: In step S3, the construction of the enhanced random forest model includes: constructing a decision tree based on a sample subset randomly extracted from the lithium-ion battery parameter standard data set, each decision tree is trained by randomly selecting features, and weights are assigned according to the accuracy of the individual decision trees during the verification process. The formula is: Where: is the output of the predicted i-th input sample, N is the total number of decision trees in the forest, and f n (X i ) is the nth decision tree for input X i The prediction result, ω n is the weight of the nth tree.

6. The method for monitoring the status of an underwater robot lithium-ion battery according to claim 1, wherein: In step S3, the TCN model includes the following formula: h t+1 =εh t v t Where: is the convolution value on the a-th row vector and the b-th convolution kernel, E is the convolution kernel, l is the length of the time series, W is the width of the convolution kernel, v t is the new timing information, is the output of the convolutional layer, α a is the weight of the a-th row vector, h t+1 is the final output, h t is the current input, and ε is the output coefficient, which is used to adjust the scale of the final output.

7. The method for monitoring the status of an underwater robot lithium-ion battery according to claim 1, wherein: In step S3, the extreme gradient boosting weighted model includes: The output results of the input multi-stress factor model, enhanced random forest model, and TCN model are optimized by extreme gradient boosting weighted optimization. The formula is as follows: Where: Obj is the model output, n is the number of samples, y i is the true value of the i-th sample, is the predicted value of the i-th sample, is the loss function, which is used to measure the difference between the predicted value and the true value, K is the number of trees, and f k is the kth tree, Ω(f k ) is a regularization term used to control the complexity of the model.

8. The method for monitoring the status of an underwater robot lithium-ion battery according to claim 1, wherein: The following steps are also included: S4. Estimating the state of the lithium-ion battery according to the estimated result of the lithium-ion battery life, and outputting the estimated result of the lithium-ion battery state.

9. The method for monitoring the status of an underwater robot lithium-ion battery according to claim 8, characterized in that: The step S4 comprises: S41. Calculate the time required to execute the task using the underwater robot positioning system. The formula is: Where: T mission is the time required for task execution; L is the length of the task path; k path is the path correction coefficient, v is the speed of the underwater robot; v current is the water flow velocity; θ is the angle between the water flow and the direction of movement; S42. Calculate the remaining mission time of the underwater robot using the estimated result of the lithium-ion battery life. The formula is: Where: T safe is the remaining mission time of the underwater robot, RUL is the estimated life of the lithium-ion battery, and α is the safety factor; S43: Output the remaining mission time of the underwater robot as the lithium-ion battery state estimation result.

10. A hybrid model-based underwater robot lithium-ion battery status monitoring system, characterized in that: include: Multi-physics field data acquisition module: used to collect physical parameters of the lithium-ion battery of the underwater robot to be tested, including the voltage, temperature, capacitance, current of the lithium-ion battery and the pressure value outside the lithium-ion battery of the underwater robot to be tested; Data preprocessing module: used to preprocess physical parameters; Lithium-ion battery life prediction module: used to build a lithium-ion battery life prediction model, and use the lithium-ion battery parameter standard data set to train the lithium-ion battery life prediction model. The lithium-ion battery life prediction model includes a multi-stress factor model, an enhanced random forest model, a TCN model and an extreme gradient boosting weighted model. The extreme gradient boosting weighted model performs weighted optimization on the outputs of the multi-stress factor model, the enhanced random forest model and the TCN model, and inputs the preprocessed physical parameters into the trained lithium-ion battery life prediction model to obtain the lithium-ion battery life prediction result.

Images