Expressway electromechanical equipment lithium battery life estimation and maintenance state evaluation method

Abstract

The invention discloses a highway electromechanical equipment lithium battery life estimation and maintenance state evaluation method, which comprises the following steps of: firstly, calculating a capacity increment IC curve corresponding to different circulation turns of a lithium battery according to voltage and current data in the circulation of the lithium battery; for IC curves of different rounds, calculating a probability density curve after order of magnitude is removed, and extracting statistical state parameters of the probability density curve by using an EM algorithm; then extracting physical characteristics related to lithium battery life attenuation; processing the extracted statistical state parameters and physical state parameters, inputting into a lithium battery life prediction network, and outputting a next cycle battery life estimation value; and finally, aiming at the current battery life, providing a corresponding highway electromechanical equipment maintenance method. The method is simple and high in operation efficiency, the service life of the lithium battery is estimated in time according to the cycle condition of the lithium battery, maintenance guidance is provided, and the probability of emergency situations of the lithium battery of the highway electromechanical equipment is effectively reduced.

Description

Technical Field

[0001] The present invention belongs to the field of transportation technology, and in particular relates to a method for estimating the life span and maintenance status of lithium batteries of electromechanical equipment of highways based on an IC curve. Background Art

[0002] With the rapid development of modern highway construction, electromechanical equipment plays a vital role in traffic management and operation. This equipment includes surveillance cameras, electronic toll collection systems, road lighting, traffic lights, and communication facilities. Lithium batteries, as a highly efficient energy storage solution, are widely used in these devices due to their high energy density, long lifespan, and high charge and discharge efficiency. However, the performance and lifespan of lithium batteries are affected by various factors, including temperature fluctuations, the number of charge and discharge cycles, and battery aging. Without effective monitoring and maintenance, equipment failures can occur, impacting normal highway operations and traffic safety.

[0003] Accurately predicting the lifespan of lithium batteries and conducting preventive maintenance is crucial. This predictive approach allows for the early identification of potential problems and the implementation of appropriate measures, avoiding equipment downtime and safety hazards caused by battery failure. Maintenance is crucial not only for extending battery life but also for reducing operating costs and improving equipment reliability.

[0004] Current lithium battery life prediction methods primarily include model-based and data-driven prediction. Model-based prediction methods rely on physical and electrochemical models, predicting battery life by simulating the internal mechanisms of the battery. However, these battery life prediction models are mostly based on the physical and chemical properties exhibited during the lithium battery cycle. While these physical variables have a direct or indirect impact on lifespan, they are significantly affected by the external environment. Data-driven prediction methods utilize large amounts of historical data and use machine learning algorithms to predict lifespan. However, data-driven prediction methods focus solely on the time series nature of historical life curves, detaching from the physical and chemical properties of the battery itself, resulting in limited predictive capabilities.

[0005] In summary, the life prediction and maintenance of lithium batteries for highway electromechanical equipment is a key link in ensuring the stable operation of the equipment and an important guarantee for improving the efficiency and safety of the overall transportation system. Summary of the Invention

[0006] Purpose of the invention: In response to the above problems, the present invention proposes a method for estimating the life span and evaluating the maintenance status of lithium batteries for highway electromechanical equipment based on IC curves. Based on the IC curve data of the discharge cycle process of lithium batteries for highway electromechanical equipment, its statistical state parameters are calculated. Combined with physical state parameters such as temperature, the method realizes lithium battery life prediction and recommends maintenance methods through learning from historical data.

[0007] Technical Solution: To achieve the purpose of the present invention, the technical solution adopted by the present invention is: a method for estimating the life span and maintenance status of lithium batteries of highway electromechanical equipment based on IC curve, comprising the following steps:

[0008] S1. Calculate the capacity increment curve of lithium batteries for highway electromechanical equipment at different discharge cycles;

[0009] S2. Remove the order of magnitude of the capacity increment value sequence in different discharge cycles of the lithium battery, retain its sequence distribution shape, and then calculate the probability density distribution curve of the capacity increment value sequence with the order of magnitude removed;

[0010] S3. Use the maximum expectation algorithm to perform parameter estimation on the probability density distribution curve of the capacity increment value sequence to obtain the statistical state parameters of the lithium battery;

[0011] S4. Obtaining the changes in the physical quantity values ​​of the battery analysis at different cycles to obtain the physical state parameters of the lithium battery;

[0012] S5. Process the statistical state parameters and physical state parameters of the lithium battery, input the processed data into the feature coding LSTM-CNN network model, and obtain the estimated life of the lithium battery in the next cycle;

[0013] S6. Based on the life status of lithium batteries of electromechanical equipment in highways, the maintenance status intervals are divided, and the statistical status parameters and physical status parameters of lithium batteries are used to construct a maintenance evaluation model for lithium batteries of electromechanical equipment in highways.

[0014] Furthermore, S1 is specifically:

[0015] For the voltage-current data of the lithium battery discharge cycle of highway electromechanical equipment, calculate the amount of charge or discharge per unit voltage at different terminal voltages;

[0016]

[0017] Among them, dQ is the change in battery capacity, dV is the change in battery voltage, is the capacity increment of the battery, Δt is the time change value, I is the current of constant current discharge, and ΔV is the voltage change value within Δt.

[0018] Furthermore, S3 is specifically:

[0019] Use the maximum expectation algorithm to build a Gaussian mixture model for the probability density curve and obtain the parameters of the built Gaussian mixture model;

[0020] The maximum expectation algorithm first assigns initial values ​​to the parameters of the Gaussian mixture model, and then estimates the latent variable distribution in the E step to find the probability γ(ik) that the i-th data point belongs to the k-th Gaussian distribution.

[0021]

[0022] Among them, N is the standard Gaussian distribution function, K is the number of Gaussian distributions to be fitted, and x i is the i-th data point, λ k and λ j are the proportions of the kth and jth Gaussian distributions in the Gaussian mixture distribution, μ k and μ j are the means of the kth and jth Gaussian distributions, σ k and σ j are the standard deviations of the kth and jth Gaussian distributions, respectively;

[0023] In the M step, the likelihood estimation is used to estimate the parameters, and the number of data points N of the k-th Gaussian distribution is calculated by formula (3): k , Formula (4) updates the weight λ of the k-th Gaussian distribution in the entire Gaussian mixture model k , Formula (5) updates the k-th Gaussian distribution mean μ k , Equation (6) updates the k-th Gaussian distribution variance σ k ;

[0024]

[0025] Finally, the E-step and M-step are iterated repeatedly to approximate the true value of the parameter, and an estimated Gaussian mixture model is established within a certain error tolerance range. The maximum expectation algorithm is used to solve the Gaussian mixture model parameters, and finally the mean μ, standard deviation σ and proportion λ corresponding to the three Gaussian distributions represented by the three peaks of the probability density distribution curve are obtained, which are the statistical state parameters of the lithium battery.

[0026] Furthermore, in S4, the current average, voltage average, temperature maximum, and discharge duration of the lithium battery of the highway electromechanical equipment in different discharge cycles are obtained, and the inflection points of the data sequence are calculated: the time when the minimum voltage occurs and the time when the maximum temperature occurs are used as the physical state parameters of the lithium battery.

[0027] Furthermore, S5 is specifically:

[0028] The extracted statistical state parameters and physical state parameters of lithium batteries of highway electromechanical equipment are normalized, and each type of parameter data is scaled to the range of [0,1]:

[0029]

[0030] in, is the normalized data, X5 is the i-th data of a certain type of parameter extracted, X 9:； X is the largest number among the extracted parameters. 956 The smallest number among the extracted parameters of a certain type;

[0031] Then the normalized data were subjected to principal component analysis and dimensionality reduction according to the two categories of statistical state parameters and physical state parameters;

[0032] The reduced-dimensional data is divided into time windows according to the set lookback length, prediction time interval delay, sampling interval step, and label length predictlen, and then divided into training and test sets in proportion.

[0033] The training set is input into the feature encoding LSTM-CNN network for learning to obtain the model weights. After the model weights are obtained, the test set data is input into the model to output the estimated value of the lithium battery life of highway electromechanical equipment;

[0034] Furthermore, the model runs as follows:

[0035] First, the model splits the input tensor according to category labels, including historical statistical state parameters, historical physical state parameters, and historical battery health. The historical statistical state parameter tensors and historical physical state parameter tensors are further input into the Dense layer for feature extraction, and the output results are spliced.

[0036] The concatenated result is then fed into the LSTM network for time series feature learning. The output is then fed into the Conv2d layer after adding one dimension, and further feature learning is performed on the tensor data through the convolution kernel.

[0037] The output of the convolutional layer is transformed in shape and then input into the Dense layer for dimension transformation, and finally the predicted lifespan label corresponding to each time window is output.

[0038] Furthermore, S6 is specifically:

[0039] If the remaining battery life capacity is higher than the threshold, it is set to the maintenance-required state, corresponding to label 0; if the remaining battery life capacity is lower than the threshold, it is set to the replacement-required state, corresponding to label 1;

[0040] Based on the extracted statistical and physical state parameters of lithium batteries for highway electromechanical equipment during discharge cycles, a logistic model for maintenance state estimation was established, and its parameters were calibrated using maximum likelihood estimation.

[0041] Based on the obtained Logistic model, for the newly input battery status information of unknown battery life, the maintenance strategy is output by selecting the maintenance threshold, thereby realizing the maintenance status evaluation;

[0042]

[0043] Where S is the maintenance strategy, which takes values ​​of 0 and 1, x is the concatenated vector of statistical state parameters and physical state parameters of the current discharge cycle of the lithium battery, ts is the maintenance threshold, w and b are the model parameters to be calibrated, and the maximum likelihood estimation is used to estimate their parameters.

[0044] A computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any of the methods described above.

[0045] An electronic device comprises one or more processors, one or more memories, and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing any of the methods described.

[0046] Beneficial effects: Compared with the prior art, the technical solution of the present invention has the following beneficial technical effects:

[0047] This paper extracts the IC curve from the current and voltage data of lithium batteries used in highway electromechanical equipment during charge and discharge cycles. Based on the statistical state parameters of the IC curve and the physical state parameters of the battery during charge and discharge, a lithium battery life prediction model and a lithium battery maintenance assessment model are established. Both models share a common set of easily accessible basic parameter inputs, run rapidly, and, combining physical and statistical properties, achieve high accuracy, making them suitable for lithium battery life prediction and maintenance assessment under various operating conditions.

[0048] A lifespan estimation and maintenance method for lithium batteries in electromechanical equipment on highways based on IC curves extracts statistical state parameters of lithium battery life with the help of IC curves, and extracts key physical state parameters in combination with observational data. By integrating model-based and data-driven methods, the accuracy and reliability of predictions can be further improved, which has important practical significance for the operation and management of modern highways. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 Flow chart of the method of the present invention.

[0050] Figure 2 The IC curves corresponding to the discharge process of B0005 battery in different cycles.

[0051] Figure 3 The probability density curve of dQ / dV of B0005 battery in different discharge cycles after discarding the order of magnitude and its estimated GMM curve.

[0052] Figure 4 The battery voltage and current changes of B0005 battery in different discharge cycles.

[0053] Figure 5 This is the temperature change diagram of B0005 battery in different discharge cycles.

[0054] Figure 6 Schematic diagram of time window division.

[0055] Figure 7 This is the feature encoding LSTM-CNN network architecture diagram.

[0056] Figure 8 This is the lithium battery life prediction result diagram.

[0057] Figure 9 Confusion matrix plot for the lithium battery maintenance evaluation model.

[0058] Figure 10 ROC curve of the lithium battery maintenance method evaluation model. DETAILED DESCRIPTION

[0059] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0060] This embodiment uses the charge and discharge data of batteries B0005, B0006, B0007, and B0018 from the NASA lithium battery data set. These four batteries are subjected to repeated cycles of charge-discharge-impedance at room temperature until the end of their life (EOL), that is, the battery capacity decays by 30%, from 2Ahr to 1.4Ahr. The charging process is charged to 4.2V at a constant current of 1.5A, and then the constant voltage is maintained until the charging current is less than 20mA; the discharging process is discharged at a constant current of 2A for the four batteries until the voltage is 2.7V, 2.5V, 2.3V, and 2.5V respectively. The selected lithium batteries are all 18650 cylindrical lithium-ion batteries, which are commonly used as UPS power supplies for highway electromechanical equipment. The method for estimating the life and maintenance status of lithium batteries for highway electromechanical equipment based on IC curves described in the present invention is specifically implemented as follows:

[0061] 1. Feature data extraction.

[0062] For the selected battery discharge cycle data, when the discharge current is not 2A, the battery voltage will rise again, so the discharge moments where the current is not 2A at the beginning and end need to be discarded. After removing the abnormal values, the IC curves of the discharge process of the four batteries are calculated according to formula (1): Figure 2 The left figure shows the discharge IC curves for the B0005 battery at different cycle times. It can be seen that for this type of lithium battery commonly used in highway electromechanical equipment, the IC curve exhibits two prominent peaks and a non-prominent peak sandwiched between them. As the number of cycles increases, and thus the battery life decreases, the dQ / dV values ​​corresponding to the peak values ​​of the three peaks decrease, and the corresponding V values ​​also decrease.

[0063] Raw IC curve data contains numerical noise caused by external environmental influences, making it difficult to plot its probability density distribution curve. Therefore, the raw data must be denoised and smoothed to maximize the representation of IC data variations. Locally weighted regression (Loess Regression) uses regional weighting for smoothing, effectively fitting the curve trend while offering faster computational speed and minimal information loss. Figure 2 The right figure in the figure shows the result after smoothing using Loess regression.

[0064] The probability density distribution of dQ / dV can be calculated based on the smoothed IC curve. Since the dQ / dV of different batteries varies in the order of 1e4, its order of magnitude is discarded to reduce the cost of subsequent calculations. This operation does not change the shape of the various curves and is beneficial to the performance of the subsequent prediction model. Figure 5 The middle curve is the probability density distribution of the B0005 battery dQ / dV after discarding the order of magnitude. The shape of its probability density distribution is similar to that of the IC curve and is multimodal. At the same time, since the probability density is a dimensionless positive number, it is more generalizable and reliable than GMM parameter estimation based on the IC curve.

[0065] The maximum expectation (EM) algorithm is now used to estimate the GMM parameters of the probability density curve of the dQ / dV of different battery discharge cycles after the order of magnitude is discarded. Since the EM algorithm is sensitive to initial values, different initial parameter settings have a significant impact on the final estimation results. Therefore, a grid search is used to find the most appropriate initial values. After finding the appropriate initial values, the EM algorithm is used to estimate the GMM parameters. Figure 3 The middle curve is the GMM curve corresponding to the parameters estimated by the EM algorithm. The error between the estimated curve and the original curve is within 5%, which can be considered that the parameter estimation effect is good.

[0066] In addition to the statistical results obtained from the IC curve, the current, voltage and temperature changes during the battery discharge cycle are also worthy of attention. Figure 4 The curves of battery voltage and current changing with time corresponding to different discharge cycles of B0005 battery show that as the number of cycles increases, the time it takes for the battery discharge voltage to reach the lowest value continues to shorten, and the duration of the battery discharge current also continues to shorten. This is consistent with common sense. The lower the battery life, the shorter the discharge time and the smaller the maximum voltage. Figure 5The temperature change curve of the B0005 battery over time for different discharge cycles shows that as the number of cycles increases, the time when the battery temperature reaches the highest value continues to shorten, and the maximum temperature value during battery discharge also increases. That is, the shorter the battery life, the higher the maximum discharge temperature and the faster it heats up.

[0067] Therefore, the final prediction model input parameters selected are the nine GMM estimated parameters representing the probability density curve of dQ / dV after discarding the order of magnitude, and the physical variables representing the physical performance of the battery discharge process. Since the sum of the three Gaussian distributions is 1, the proportion of the third Gaussian distribution can be inferred from the first two proportions, indicating that the three proportions are collinear. Therefore, the third Gaussian distribution proportion is discarded. All 12 variables are shown in Table 1:

[0068] Table 1 Prediction model input parameters

[0069]

[0070] 2. Data normalization and dimensionality reduction.

[0071] The feature data in Table 1 have different dimensions and orders of magnitude. When training the life prediction model, data with uncertain orders of magnitude will lead to slower gradient descent and may also cause gradient explosion. Therefore, it is necessary to scale the feature data. Since the feature data are all positive and their distribution and change trends are relatively stable, we choose to use Equation (7) to normalize the data and scale it to the range [0, 1] to ensure better gradient descent and learning performance during life prediction model training.

[0072] Although the distribution and dimensionality of the normalized feature data are uniform, the data still contains a large number of probability density variables, and some of these variables may be correlated, which is detrimental to the performance of the subsequent lifespan prediction model. Therefore, dimensionality reduction is necessary to reduce the number of variables while minimizing the correlation between them. Therefore, principal component analysis (PCA) is performed on the probability density variables in the feature data to reduce the dimensionality, and the two categorical variables after processing are used for model training.

[0073] 3. Time window division.

[0074] Time windowing is an important method in time series forecasting. A time window is an m×n tensor, where m is the timestamp length and n is the feature embedding dimension. Its purpose is to sample a finite time series to obtain a continuous subsequence that still maintains temporal correlation, thereby expanding the finite sample without destroying the original temporal order.

[0075] The time window involves several important parameters: lookback, delay, step, and predictlen. Lookback is the lookback length, indicating that the data for the lookback timestamp before the current timestamp is selected. Step is the sampling interval, indicating that data is sampled at the step interval within the sequence of the previous lookback length. The resulting data set is part of the training set. Delay is the prediction interval, indicating that the data sequence obtained through the above process is predicted to be the data sequence after delay timestamps. Predictlen is the label length, which is also the length of the predicted sequence, without sampling.

[0076] like Figure 6 The following is the basic principle of time window division. In sequence X, from X=x ％ We start sampling with the lookback as the sampling length and the step as the sampling interval. The resulting data set is the area within the lookback range shown in the figure. The corresponding label is X = x / =x ％ The data set with a length of predictlen after +delay is the area within the range indicated by predictlen in the figure.

[0077] For the feature data that has completed preprocessing, the time window division parameters shown in Table 2 are used to create the data set, and the divided time window data corresponding to B0005, B0006, and B0007 are used as the training data of the life prediction model, and the time window data corresponding to B0018 is used as the test data of the model.

[0078] Table 2 Time window division parameters

[0079]

[0080] 4. Comparison of model input and prediction results.

[0081] The feature encoding LSTM-CNN network architecture proposed in this invention is as follows Figure 7 As shown in Figure 2. The model input is a concatenated time window tensor of historical lifespan labels, historical statistical features, and historical physical features. The input data batch size is bs, the time window length is ws, the historical lifespan label dimension is 1, the historical statistical feature dimension is stat, and the historical physical feature dimension is para. The shape of the model input tensor is (bs, ws, (stat+para+1)).

[0082] The model first splits the input tensor by category. For high-dimensional historical features, it further extracts features through a Dense layer, concatenating the outputs. The concatenated results are then fed into an LSTM network for time series feature learning. The output is then fed into a Conv2d layer, where convolution kernels are used to further refine the tensor data. The convolution layer output is reshaped and then fed into a Dense layer for dimensionality transformation. Finally, the predicted lifespan label for each time window is output in the shape of (bs, 1).

[0083] The present invention selects LSTM network and MLP network as comparison models of feature encoding LSTM-CNN network, and the three models share the training parameters shown in Table 3. All three models use the divided time window data corresponding to B0005, B0006, and B0007 as the training data of the life prediction model, and the time window data corresponding to B0018 as the test data of the model. Considering that the model training results are affected by the random seed, in order to minimize the randomness of the model prediction results, the present invention selects the average value of B0018 life prediction for ten consecutive rounds as the final prediction result. The prediction results of the feature encoding LSTM-CNN network are shown in Figure 3. Figure 8 shown.

[0084] Table 3 Model training parameters

[0085]

[0086] The average evaluation index of the life prediction results of B0018 for ten consecutive rounds of each model is shown in Table 4:

[0087] Table 4 Average evaluation indicators of different models

[0088]

[0089] The evaluation metrics in Table 4 show that the feature-encoding LSTM-CNN network proposed in this paper performs best in predicting the lifespan of lithium batteries for complete highway electromechanical equipment. Both the feature-encoding LSTM-CNN network and the LSTM network provide good predictions of overall battery life trends. However, the MLP network exhibits an overall overestimation of the predicted lifespan in later lifespan predictions. This overestimation at the end of a battery's lifespan is highly detrimental to battery maintenance and repair. Regarding localized lifespan recovery during battery cycling, both the feature-encoding LSTM-CNN network and the LSTM model outperform the MLP network. The LSTM network exhibits some minor, overall overestimations of the lifespan in the early and middle stages of the prediction, while the feature-encoding LSTM-CNN network better captures local variations in the curve. The feature-encoding LSTM-CNN network performs best among the three models in terms of evaluation metrics across ten consecutive prediction rounds, achieving improvements of 9.52%, 4.00%, and 14.29% in MAE, RMSE, and MAPE, respectively, over the LSTM model.

[0090] 5. Lithium battery maintenance status assessment.

[0091] For lithium batteries used in highway electromechanical equipment, if the remaining life capacity is above 90%, the battery is considered healthy and requires no maintenance. If the remaining life capacity is below 90% but above 80%, appropriate equipment maintenance is required. If the remaining life capacity is below 70%, the battery is considered to have reached its expected life cycle and will be replaced. In actual highway environments, it is difficult to directly observe the lifespan and actual capacity of lithium batteries. Therefore, a battery maintenance model is constructed based on data that can be measured by other sensors.

[0092] To simplify the problem, batteries with a remaining life capacity above 80% are classified as requiring maintenance, corresponding to label 0; batteries with a remaining life capacity below 80% are classified as requiring replacement, corresponding to label 1. A logistic model is developed based on the statistical and physical state variables obtained from the cycles of lithium batteries used in highway electromechanical equipment. The lithium battery maintenance assessment model is built using training and test data for the service life prediction model. The maintenance state estimation process does not utilize battery life data, but only the statistical and physical state parameters of the lithium battery over different cycles.

[0093] The prediction results of the maintenance status assessment model on the test set are shown in Table 5, and the confusion matrix is ​​shown in Figure 9 As shown, the ROC curve is as Figure 10 shown.

[0094] Table 5 Prediction results of maintenance status assessment model

[0095]

[0096] The prediction indicators show that the maintenance status assessment model correctly predicted a small number of objects that did not require maintenance as requiring maintenance, while predicting all objects that required maintenance as requiring maintenance. The predictions for all objects requiring maintenance were accurate, and even if a few objects were predicted as requiring maintenance, this did not affect the operation of the entire highway's electromechanical equipment and could even benefit the lithium batteries. The model performed well.

Claims

1. A method for estimating the life of lithium batteries of electromechanical equipment on highways and evaluating the maintenance status, characterized in that: The steps include: S1. Calculate the capacity increment curve of lithium batteries for highway electromechanical equipment at different discharge cycles; S2, removing the order of magnitude of the capacity increment value sequence in different discharge cycles of the lithium battery, retaining its sequence distribution shape, and then calculating the probability density distribution curve of the capacity increment value sequence with the order of magnitude removed; S3. Use the maximum expectation algorithm to estimate the parameters of the probability density distribution curve of the capacity increment value sequence to obtain the statistical state parameters of the lithium battery; S4. Obtain the numerical changes of physical quantities of different cycles of battery analysis to obtain physical state parameters of the lithium battery; S5. Process the statistical state parameters and physical state parameters of the lithium battery, input the processed data into the feature coding LSTM-CNN network model, and obtain the estimated life of the lithium battery in the next cycle; S6. According to the life status of lithium batteries of electromechanical equipment in highways, the maintenance status intervals are divided, and the statistical status parameters and physical status parameters of lithium batteries are used to build a maintenance evaluation model for lithium batteries of electromechanical equipment in highways.

2. The method for estimating the life span and maintaining the status of lithium batteries of electromechanical equipment of highways based on IC curves according to claim 1 is characterized in that: S1 is specifically: For the voltage-current data of the lithium battery discharge cycle of highway electromechanical equipment, calculate the amount of charge or discharge per unit voltage at different terminal voltages; Among them, dQ is the change in battery capacity, dV is the change in battery voltage, is the capacity increment of the battery, Δt is the time change value, I is the current of constant current discharge, and ΔV is the voltage change value within Δt.

3. The method for estimating the life span and maintaining the status of lithium batteries of electromechanical equipment of highways based on IC curve according to claim 1 is characterized in that: S3 is specifically: Use the maximum expectation algorithm to build a Gaussian mixture model for the probability density curve and obtain the parameters of the built Gaussian mixture model; The maximum expectation algorithm first assigns initial values ​​to the parameters of the Gaussian mixture model, and then estimates the latent variable distribution in step E to find the probability γ(ik) that the i-th data point belongs to the k-th Gaussian distribution. Among them, N is the standard Gaussian distribution function, K is the number of Gaussian distributions to be fitted, and x i is the i-th data point, λ k and λ j are the proportions of the kth and jth Gaussian distributions in the Gaussian mixture distribution, μ k and μ j are the means of the kth and jth Gaussian distributions, σ k and σ j are the standard deviations of the kth and jth Gaussian distributions respectively; In the M step, the likelihood estimation is used to estimate the parameters, and the number of data points N of the k-th Gaussian distribution is calculated by formula (3): k , Formula (4) updates the weight λ of the k-th Gaussian distribution in the entire Gaussian mixture model k , Formula (5) updates the k-th Gaussian distribution mean μ k , Formula (6) updates the k-th Gaussian distribution variance σ k ; Finally, the E and M steps are iterated repeatedly to approximate the true values ​​of the parameters, and an estimated Gaussian mixture model is established within a certain error tolerance range. The maximum expectation algorithm is used to solve the Gaussian mixture model parameters, and finally the mean μ, standard deviation σ and proportion λ corresponding to the three Gaussian distributions represented by the three peaks of the probability density distribution curve are obtained, which are the statistical state parameters of the lithium battery.

4. The method for estimating the life span and maintaining the status of lithium batteries of electromechanical equipment of highways based on IC curve according to claim 1 is characterized in that: In S4, the current mean, voltage mean, temperature maximum, and discharge duration of the highway electromechanical equipment lithium battery in different discharge cycles are obtained, and the inflection points of the data sequence are calculated: the time when the minimum voltage occurs and the time when the maximum temperature occurs are used as the physical state parameters of the lithium battery.

5. The method for estimating the life span and maintaining the status of lithium batteries of electromechanical equipment of highways based on IC curves according to claim 1 is characterized in that: S5 is specifically: The extracted statistical state parameters and physical state parameters of lithium batteries of highway electromechanical equipment are normalized, and each type of parameter data is scaled to the range of [0,1]: in, is the normalized data, X5 is the i-th data of a certain type of parameter extracted, X 9:； is the largest number among the extracted parameters, X 956 The smallest number among the extracted parameters of a certain type; Then, the normalized data were subjected to principal component analysis and dimensionality reduction according to the two categories of statistical state parameters and physical state parameters; The reduced-dimensional data is divided into time windows according to the set lookback length, prediction time interval, sampling interval, and label length, and then divided into training and test sets in proportion. The training set is input into the feature encoding LSTM-CNN network for learning to obtain the model weights. After the model weights are obtained, the test set data is input into the model to output the estimated value of the lithium battery life of highway electromechanical equipment.

6. The method for estimating the life span and evaluating the maintenance status of lithium batteries of electromechanical equipment of highways based on IC curves according to claim 5 is characterized in that: The model runs as follows: First, the model splits the input tensor according to category labels, including historical statistical state parameters, historical physical state parameters, and historical battery health. The historical statistical state parameter tensors and historical physical state parameter tensors are further input into the Dense layer for feature extraction, and the output results are spliced; The concatenated result is then fed into the LSTM network for time series feature learning. The output result is fed into the Conv2d layer after adding one dimension, and further feature learning is performed on the tensor data through the convolution kernel. The output of the convolutional layer is transformed in shape and then input into the Dense layer for dimension transformation, and finally the predicted life span label corresponding to each time window is output.

7. The method for estimating the life span and maintaining the status of lithium batteries of electromechanical equipment of highways based on IC curves according to claim 1 is characterized in that: S6 is specifically: When the remaining battery life capacity is higher than the threshold, it is set to a maintenance-required state, corresponding to label 0; when the remaining battery life capacity is lower than the threshold, it is set to a replacement-required state, corresponding to label 1; Based on the extracted statistical state parameters and physical state parameters of lithium batteries in highway electromechanical equipment discharge cycles, a logistic model for maintenance state estimation is established, and its parameters are calibrated by maximum likelihood estimation. Based on the obtained Logistic model, for the newly input battery status information of unknown battery life, the maintenance strategy is output by selecting the maintenance threshold, thereby realizing the maintenance status evaluation; Among them, S is the maintenance strategy, which takes values ​​of 0 and 1, x is the concatenated vector of statistical state parameters and physical state parameters of the current discharge cycle of the lithium battery, ts is the maintenance threshold, w and b are model parameters, and the maximum likelihood estimation is used to estimate their parameters.

8. A computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, characterized in that: When the instructions are executed by a computing device, the computing device is caused to perform the method according to any one of claims 1 to 7.

9. An electronic device, characterized in that: The method comprises one or more processors, one or more memories and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing the method as claimed in any one of claims 1 to 7.

Images