A battery temperature prediction method and system

Abstract

A battery temperature prediction method and system, the method comprising: obtaining charging data of the battery at regular intervals, the charging data including historical charging data and real-time charging data (S101); determining historical derivative data according to the historical charging data (S102); determining eigenvector data according to the historical charging data and the historical derivative data (S103); determining model input samples under various charging conditions according to the eigenvector data (S104); establishing a deep autoregressive model according to the model input samples (S105); determining the predicted battery temperature according to the real-time charging data and the deep autoregressive model (S106). The adopted temperature prediction method based on deep autoregression can significantly improve the accuracy of temperature prediction by mining the features and hidden information of the data sequence and performing real-time prediction on the maximum temperature of the battery cell during the charging process.

Description

Technical Field

[0001] The present invention relates to the field of batteries, and particularly to a battery temperature prediction method and system. Background Art

[0002] The management and diagnosis of batteries are one of the core technologies in battery applications. Timely and accurate diagnosis of battery faults can detect the adverse effects of faults on the battery early, extend the service life of the battery, and avoid catastrophic accidents in extreme cases. Currently, the research on battery temperature mainly focuses on calibrating the heat dissipation parameters between the battery and the coolant and between the battery and the air through exploring the internal reaction mechanism and external characteristics of the battery in an experimental environment for diagnosis. However, in the real vehicle environment, the battery temperature is affected by various factors, and the existing methods are difficult to apply to the real vehicle environment, resulting in low temperature prediction accuracy.

[0003] In addition, generally, the warning threshold is directly provided based on business experience, or N times the standard deviation is directly set to determine abnormal battery temperature. These methods often lead to false alarms due to accidental deviations in the temperature sequence in practice.

[0004] Therefore, it is necessary to provide a solution to solve the technical problem of low temperature prediction accuracy in the prior art. Summary of the Invention

[0005] In order to solve the technical problem of low temperature prediction accuracy in the prior art, the present invention proposes a battery temperature prediction method and system, and the present invention is specifically implemented by the following technical solutions.

[0006] A battery temperature prediction method provided by the present invention includes:

[0007] Acquiring the charging data of the battery at regular intervals, where the charging data includes historical charging data and real-time charging data;

[0008] Determining historical derivative data according to the historical charging data;

[0009] Determining eigenvector data according to the historical charging data and the historical derivative data;

[0010] Determining model input samples under various charging conditions according to the eigenvector data;

[0011] Establishing a deep autoregressive model according to the model input samples;

[0012] Determining the predicted battery temperature according to the real-time charging data and the deep autoregressive model.

[0013] The further improvement of the battery temperature prediction method provided by the present invention lies in that the determining historical derivative data according to the historical charging data includes:

[0014] Determine the average temperature of the battery cell, the temperature range of the battery cell, the average voltage of the battery cell, and the voltage range of the battery cell according to the historical charging data;

[0015] Determine the periodic characteristics of the charging time according to the historical charging data, sine function, and cosine function;

[0016] According to the historical charging time data, standardize the month of the charging time, the week number of the charging within a year, and the hour number of the charging within a day to determine the charging frequency characteristics;

[0017] Determine the temperature lag term characteristics according to the historical battery cell charging temperature data and the preset lag term parameter.

[0018] A further improvement of the battery temperature prediction method provided by the present invention is that the determining of the historical derivative data according to the historical charging data further includes:

[0019] Determine the highest temperature prediction trend according to the historical charging data and the linear interpolation method.

[0020] A further improvement of the battery temperature prediction method provided by the present invention is that the determining of the historical derivative data according to the historical charging data further includes:

[0021] Determine the longitude and latitude information during charging according to the historical charging data, and standardize the longitude and latitude information to determine the longitude and latitude interval.

[0022] A further improvement of the battery temperature prediction method provided by the present invention is that the determining of the model input samples under various charging conditions according to the feature vector data includes:

[0023] Determine the training data under various charging conditions according to the feature vector data, the preset observation time window length, the preset prediction window length, and the preset maximum lag term;

[0024] Perform uniform sampling on the training data according to the preset random number seed to determine the model input samples.

[0025] A further improvement of the battery temperature prediction method provided by the present invention is to establish the deep autoregressive model according to the model input samples and using the chain rule of conditional probability distribution.

[0026] A further improvement of the battery temperature prediction method provided by the present invention is that the determining of the battery predicted temperature according to the real-time charging data and the deep autoregressive model includes:

[0027] Determine the real-time derivative data according to the real-time charging data;

[0028] Determine the predicted probability distribution of the battery temperature within the prediction window length according to the real-time charging data, the real-time derived data, and the deep autoregressive model.

[0029] A further improvement of the battery temperature prediction method provided by the present invention further includes:

[0030] Perform anomaly detection on the predicted battery temperature;

[0031] When the predicted battery temperature is abnormal, generate battery predicted temperature anomaly information, and perform fault handling according to the battery predicted temperature anomaly information.

[0032] A further improvement of the battery temperature prediction method provided by the present invention is that the anomaly detection of the predicted battery temperature includes:

[0033] Obtain the actual battery temperature;

[0034] Determine the residual according to the actual battery temperature and the predicted battery temperature;

[0035] Determine the residual mean and the residual standard deviation according to the residual;

[0036] Determine the anomaly threshold according to the residual, the residual mean, and the residual standard deviation;

[0037] Determine the anomaly score according to the residual mean, the residual standard deviation, and the anomaly threshold, and the anomaly score is used to characterize the probability of battery temperature anomaly.

[0038] In addition, the present invention also provides a battery temperature prediction system, which uses the above method. The system includes:

[0039] The first module is used to obtain the charging data of the battery at regular intervals, and the charging data includes historical charging data and real-time charging data;

[0040] The second module is used to determine historical derived data according to the historical charging data;

[0041] The third module is used to determine the feature vector data according to the historical charging data and the historical derived data;

[0042] The fourth module is used to determine the model input samples under various charging conditions according to the feature vector data;

[0043] The fifth module is used to establish a deep autoregressive model according to the model input samples;

[0044] The sixth module is used to determine the predicted battery temperature according to the real-time charging data and the deep autoregressive model.

[0045] The present invention adopts a method driven by big data. By mining the features and implicit information of data sequences, it can perform real-time prediction and real-time diagnosis on the highest temperature of battery cells during the charging process, significantly improving the accuracy of temperature prediction; avoiding prediction deviations caused by inconsistent experimental environments in real vehicle environments; considering various relevant time series features, predicting the probability distribution of temperature, predicting possible temperature faults in advance, and giving warnings before accidents; determining the temperature warning threshold by unsupervised learning, and outputting an anomaly score to improve the warning accuracy and reduce the false alarm rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0047] Figure 1 It is a flowchart of the battery temperature prediction method provided in Embodiment 1 of the present invention;

[0048] Figure 2 It is a principle flowchart in Embodiment 1 of the present invention;

[0049] Figure 3 It is a schematic diagram of an autoregressive recurrent neural network in Embodiment 1 of the present invention;

[0050] Figure 4 It is a block diagram of the battery temperature prediction system provided in Embodiment 2 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0052] In order to solve the technical problem of low temperature prediction accuracy in the prior art, the present invention proposes a battery temperature prediction method and system, which are specifically implemented by the following technical solutions.

[0053] Embodiment 1:

[0054] Combined with Figures 1 to 3 shown, a battery temperature prediction method provided in Embodiment 1 of the present invention includes:

[0055] Step S101: Obtain the charging data of the battery at regular intervals. The charging data includes historical charging data and real-time charging data;

[0056] Step S102: Determine the historical derivative data based on the historical charging data;

[0057] Step S103: Determine the eigenvector data based on the historical charging data and the historical derivative data;

[0058] Step S104: Determine the model input samples under various charging conditions based on the eigenvector data;

[0059] Step S105: Establish a deep autoregressive model based on the model input samples;

[0060] Step S106: Determine the predicted battery temperature based on the real-time charging data and the deep autoregressive model.

[0061] In this Embodiment 1, aiming at the problem of overheating of the single-cell temperature that may occur during the charging process of the power battery, a temperature prediction method based on deep autoregression is provided, which can significantly improve the accuracy of temperature prediction.

[0062] In this Embodiment 1, Step S101 is a data acquisition step. First, vehicle data is obtained. The vehicle data includes battery-related data when the vehicle is charging, when the vehicle is stationary, and when the vehicle is running. The charging data when the vehicle is charging is split from the vehicle data. The period is at the second level. Preferably, the period in this Embodiment 1 is 10 seconds. The data during the charging process of the power battery (one piece of data every 10s) can be collected through the TBox (vehicle box), using temperature sensors, current sensors, voltage sensors, etc., and the data is transmitted back to the big data platform after being cleaned.

[0063] The historical charging data includes historical single-cell battery temperature, historical charging current, historical charging voltage, historical charge quantity, historical battery resistance, historical state of charge, historical charging time data, and historical single-cell battery charging temperature data, etc. The real-time charging data includes real-time single-cell battery temperature, real-time charging current, real-time charging voltage, real-time charge quantity, real-time battery resistance, real-time state of charge, real-time charging time data, and real-time single-cell battery charging temperature data, etc.

[0064] The historical derivative data includes the highest temperature prediction trend, the average value of the single-cell battery temperature, the range of the single-cell battery temperature, the average value of the single-cell battery voltage, the range of the single-cell battery voltage, the longitude and latitude interval during charging, the periodic characteristics of the charging time, the charging frequency characteristics, and the temperature lag term characteristics, etc.

[0065] Further, Step S102 includes:

[0066] Determine the predicted trend of the highest temperature based on historical charging data and the linear interpolation method. First, use the linear interpolation method to fill in the missing values of the historical battery cell temperature to obtain the highest temperature trend of the battery cell over a period of time in the future.

[0067] Determine the average temperature of the battery cell, the temperature range of the battery cell, the average voltage of the battery cell, and the voltage range of the battery cell according to the historical charging data.

[0068] Specifically, extract the temperature sequence, voltage sequence, resistance sequence, SOC sequence, etc. of each battery cell from the historical charging data, and calculate indicators such as the average temperature of the battery cell, the temperature range of the battery cell, the average voltage of the battery cell, and the voltage range of the battery cell. In this Example 1, for the data with a frequency of one data point per 10s, depending on the number of sensors collected, for example, there are 30 temperature sensors and 60 voltage sensors, then 30 temperature values and 60 voltage values will be recorded. Calculate indicators such as the average value and range of temperature and voltage based on these 30 temperature values and 60 voltage values; obtain the maximum value of these 30 temperature values, and the data with a 10s frequency forms a sequence, that is, the highest temperature trend.

[0069] Determine the longitude and latitude information during charging according to the historical charging data, and perform standardization processing on the longitude and latitude information to determine the longitude and latitude interval. Specifically, extract the longitude and latitude information during charging from the historical charging data and perform standardization processing.

[0070] Determine the periodic characteristics of the charging time according to the historical charging data, sine function, and cosine function.

[0071] According to the historical charging time data, perform standardization processing on the month of the charging time, the week number of charging within a year, and the hour number of charging within a day to determine the charging frequency characteristics; for example, if the charging month data is January, it will be standardized to -0.5; if it is December, it will be standardized to 0.5. In this way, for the 12 months of a year, after standardization, it is in the interval [-0.5, 0.5]; similarly, perform similar processing on the week number of charging within a year, the hour number of charging within a day, etc.

[0072] Determine the temperature lag term characteristics according to the historical battery cell charging temperature data and the preset lag term parameters. Specifically, extract the lag term characteristics of the temperature sequence, such as lag 1, lag 5, lag 6, lag 10, etc.; assume that the current time is t, and the time corresponding to lag 1 is t - 10; the specific selection of the lag term can be set by the user according to the model prediction effect.

[0073] In step S103, parameters that are crucial for predicting the battery temperature are selected, and the multivariate time series composed of arrays of multiple features is divided to form feature vectors. The historical charging data is set in the form of an array, and the historical derivative data is added to the columns of the array. The data collected in one data acquisition cycle forms one row of the array, thus forming feature vector data.

[0074] The historical derivative data may also include the cumulative deviation of individual battery voltages and the number of times of cumulative deviation of individual battery voltages, etc. The cumulative deviation of individual battery voltages is obtained by subtracting the median voltage of individual batteries from the battery monomer voltage and taking the absolute value.

[0075] Further, step S104 includes: determining training data under various charging conditions according to the feature vector data, the preset observation time window length, the preset prediction window length, and the preset maximum lag term; performing uniform sampling on the training data according to the preset random number seed to determine the model input samples. In this embodiment 1, given the observation time window length, the prediction window length, and the maximum lag term, the feature vector data is sliced to form model training data under different charging conditions; a random number seed is set, and uniform sampling is performed on the training data. Uniform sampling of the data for a single charging process can obtain the sampling samples for a single charging process; sampling the data for multiple charging processes can obtain the model input samples. The obtained model input samples can represent the charging data under different charging conditions during multiple charging processes and are used for model training input.

[0076] Further, in step S105, a deep autoregressive model is established according to the model input samples and using the chain rule of conditional probability distribution. In this embodiment 1, an autoregressive model based on deep learning is used for model training based on the feature vector data.

[0077] Let z i，t represent the value of the i-th sequence at time step t, x i，t represent the feature, and t0 represent the prediction start time. Based on the autoregressive recurrent neural network, the probability distribution of z i，t is predicted, and is represented by the likelihood function l(z i，t |θ i，t ), where θ i，t represents the parameter space to be learned. The model is as Figure 3 shown, with the training process on the left and the prediction process on the right.

[0078] During training, at each time step t, the input to the network includes the feature x i，t , the value z i，t-1 at the previous time step, and the state at the previous time step . First, calculate the current state Subsequently, calculate the parameters of the likelihood l(z|θ). Finally, learn the network parameters by maximizing the log-likelihood ). In this Example 1, a network structure with 2 hidden layers and 100 units in each layer is adopted, and the neuron units use LSTM (Long Short-Term Memory Network).

[0079] During prediction, the historical data with t < t0 is fed into the network to obtain the initial state Then, sampling is used to obtain the prediction results: for t0, t0 + 1,..., T, randomly sample at each time step to get This sampled value is used as the input for the next step. Repeating this process, a series of sampled values from t0 to T can be obtained, and the target values required, such as quantiles, expectations, etc., can be calculated using these sampled values. In this way, the prediction results form a probability distribution rather than a single-point estimate. The specific form of depends on the likelihood function l(z|θ). Since the temperature prediction is a continuous real number, we choose the Gaussian distribution for the likelihood function. Then θ = (μ, σ), where μ and σ represent the mean and standard deviation parameters of the Gaussian distribution. In the following formula, represents the state at the current time step, b μ represents the slope and intercept terms of μ, b σ represents the slope and intercept terms of σ.

[0080]

[0081]

[0082]

[0083] The output target of the network is the parameters of the probability distribution.

[0084] Furthermore, step S106 includes: determining real-time derived data according to real-time charging data; determining the predicted probability distribution of the battery temperature within the prediction window length according to the real-time charging data, real-time derived data, and the deep autoregressive model. Clean the real-time charging data to determine the real-time derived data, and input the real-time charging data and real-time derived data into the deep autoregressive model to obtain the predicted probability distribution of the battery temperature within the prediction window length.

[0085] Further, the method further includes: performing anomaly detection on the predicted battery temperature; when there is an anomaly in the predicted battery temperature, generating battery predicted temperature anomaly information, and performing fault handling based on the battery predicted temperature anomaly information. Specifically, further processing can be performed according to the abnormal predicted battery temperature, and fault alarm can be performed according to the processing result. The alarm information can be displayed on a display screen or played through a speaker.

[0086] Furthermore, performing anomaly detection on the predicted battery temperature includes:

[0087] Obtaining the actual battery temperature;

[0088] Determining a residual based on the actual battery temperature and the predicted battery temperature;

[0089] Determining the mean residual and the standard deviation of the residual based on the residual;

[0090] Determining an anomaly threshold based on the residual, the mean residual, and the standard deviation of the residual;

[0091] Determining an anomaly score based on the mean residual, the standard deviation of the residual, and the anomaly threshold, where the anomaly score is used to characterize the probability of battery temperature anomaly.

[0092] Specifically, calculate the residual sequence e = [e (t-h) ,...e (t-1) , e(t)], where the residual y (t) is the actual temperature value collected by the sensor in real time, is the prediction value of the deep autoregressive model, h refers to the prediction window length, and t corresponds to the current time point.

[0093] In this Embodiment 1, an unsupervised learning method is used to determine the anomaly threshold, specifically as follows:

[0094] Assume that the anomaly threshold is generated in the following way: ε = μ(e) + zσ(e), where μ(e) and σ(e) are the mean and standard deviation of the residual respectively, and z > 0 is the coefficient of the standard deviation. Then:

[0095]

[0096] Among them:

[0097] Δμ(e) = μ(e) - μ({e ∈ e|e < ε})

[0098] Δσ(e) = σ(e) - σ({e ∈ e|e < ε})

[0099] e a = {e ∈ e|e > ∈}

[0100] Eseq = |e a ∈ e a |

[0101] The determination method of the outlier threshold is that if the large residuals are removed, the mean and standard deviation of the original residual sequence should decrease significantly. In addition, penalties are imposed on the magnitude and quantity of the residual values outside the range to obtain an adaptive outlier threshold.

[0102] The outlier score is calculated according to the following formula:

[0103]

[0104] where represents the maximum value of the residual sequence e a at the i-th prediction. That is, the residual sequence is standardized to output the outlier score. Among them, the higher the outlier score, the greater the possibility of abnormality of the battery temperature.

[0105] In addition, the determination of the outlier threshold can also be based on the business method, setting standards, and allowing users to set the outlier threshold by themselves.

[0106] Traditional prediction methods are single-sequence time series prediction. In these methods, the model parameters at each given time are independently estimated from past observations, and the models are usually manually selected to explain different factors, such as autocorrelation structure, trend, seasonality, etc. Embodiment 1 of the present invention is based on a deep autoregressive model; considering that the temperature change of the power battery is affected by multiple related time series, such as current, charging resistance, charge amount, and ambient temperature during the charging process, Embodiment 1 of the present invention incorporates these related time series attributes to fit a more complex and accurate model. At the same time, Embodiment 1 of the present invention also reduces the workload brought by manual feature engineering and model selection. In addition, by writing a prediction framework, the model supports time series training and prediction with a frequency of seconds, which is different from the traditional model that only supports model training above the minute level. Embodiment 1 of the present invention can be better applied to the temperature prediction and early warning of power batteries.

[0107] Traditional time series prediction can only give a single-point estimate of the temperature. Embodiment 1 of the present invention can not only provide a specific single-point estimate, but also provide the probability distribution of the battery temperature within a certain period of time in the future; by providing the entire probability prediction distribution of the temperature, Embodiment 1 of the present invention can better assist in decision-making.

[0108] In this Embodiment 1, the model input samples are obtained by uniform sampling. According to the observation time window length, the prediction window length, and the maximum lag term of the temperature sequence, time window sliding is performed on all observed time series. For all the samples formed after time window sliding, the model input samples are obtained by uniform sampling. This processing method can, on the one hand, improve the model training speed, and on the other hand, obtain the characteristics under different charging conditions, which is superior to the traditional processing methods of using all samples or randomly splitting samples.

[0109] In the prior art, the warning threshold is directly provided according to business experience, or N times the standard deviation is directly set to identify anomalies. These traditional methods often lead to false alarms due to accidental deviations in the temperature sequence in practice. In this Embodiment 1, an unsupervised learning method is used to determine the temperature warning threshold and output the anomaly score. This Embodiment 1 uses an unsupervised learning method to provide the warning threshold, identify and eliminate accidental deviations, and on this basis, provide the anomaly score value, improving the warning accuracy and reducing the false alarm rate.

[0110] The present invention adopts a method based on a deep autoregressive model and big data drive. By mining the characteristics and implicit information of the data sequence, it can perform real-time prediction and real-time diagnosis on the highest temperature of the battery cell during the charging process, avoiding prediction deviations caused by inconsistent experimental environments in the real vehicle environment. Considering various relevant time series characteristics, it predicts the probability distribution of the temperature, predicts possible temperature faults in advance, and gives an alarm before an accident. An unsupervised learning method is used to determine the temperature warning threshold, identify and eliminate accidental deviations, and on this basis, provide the anomaly score value, improving the warning accuracy and reducing the false alarm rate.

[0111] Embodiment 2:

[0112] Combined with Figure 4 As shown, this Embodiment 2 provides a battery temperature prediction system 100. Using the method in Embodiment 1, the battery temperature prediction system 100 includes:

[0113] The first module 11 is used to obtain the charging data of the battery at regular intervals, and the charging data includes historical charging data and real-time charging data;

[0114] The second module 12 is used to determine historical derivative data according to the historical charging data;

[0115] The third module 13 is used to determine the feature vector data according to the historical charging data and the historical derivative data;

[0116] The fourth module 14 is used to determine the model input samples under various charging conditions according to the feature vector data;

[0117] The fifth module 15 is used to establish a deep autoregressive model according to the model input samples;

[0118] The sixth module 16 is used to determine the predicted battery temperature according to the real-time charging data and the deep autoregressive model.

[0119] Based on the deep autoregressive model and using the RNN (Recurrent Neural Network) architecture for probabilistic prediction, the present invention can perform temperature anomaly detection based on the residual sequence. Aiming at the problem of overheating of individual cells that may occur during the charging process of power batteries, the present invention provides a temperature prediction method based on deep autoregression, which can significantly improve the accuracy of temperature prediction and the effect of temperature anomaly diagnosis.

[0120] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A battery temperature prediction method, characterized in that, Including: Obtain the charging data of the battery periodically, where the charging data includes historical charging data and real-time charging data; Determine historical derivative data according to the historical charging data; Determine eigenvector data according to the historical charging data and the historical derivative data; Determine model input samples under various charging conditions according to the eigenvector data; Establish a deep autoregressive model according to the model input samples; Determine the predicted battery temperature according to the real-time charging data and the deep autoregressive model; Among them, the determining of the historical derivative data according to the historical charging data includes: determining the temperature lag term feature according to the historical battery cell charging temperature data and the preset lag term parameter.

2. The battery temperature prediction method according to claim 1, wherein The determining of the historical derivative data according to the historical charging data further includes: Determine the average temperature of the battery cell, the temperature range of the battery cell, the average voltage of the battery cell, and the voltage range of the battery cell according to the historical charging data; Determine the periodic feature of the charging time according to the historical charging data, sine function, and cosine function; According to the historical charging time data, standardize the month of the charging time, the week number within a year when charging, and the hour number within a day when charging to determine the charging frequency feature.

3. The battery temperature prediction method according to claim 2, wherein The determining of the historical derivative data according to the historical charging data further includes: Determine the highest temperature prediction trend according to the historical charging data and linear interpolation method.

4. The battery temperature prediction method according to claim 3, wherein, The determining of the historical derivative data according to the historical charging data further includes: Determine the longitude and latitude information during charging according to the historical charging data, and standardize the longitude and latitude information to determine the longitude and latitude interval.

5. The battery temperature prediction method according to claim 1, wherein The determining of the model input samples under various charging conditions according to the eigenvector data includes: Determine training data under various charging conditions according to the eigenvector data, the preset observation time window length, the preset prediction window length, and the preset maximum lag term; Perform uniform sampling on the training data according to the preset random number seed to determine the model input samples.

6. The battery temperature prediction method according to claim 1, wherein Establish the deep autoregressive model according to the model input samples and using the chain rule of conditional probability distribution.

7. The battery temperature prediction method according to claim 1, wherein The determining of the predicted battery temperature according to the real-time charging data and the deep autoregressive model includes: Determine real-time derivative data according to the real-time charging data; Determine the probability distribution of the predicted battery temperature within the prediction window length according to the real-time charging data, the real-time derivative data, and the deep autoregressive model.

8. The battery temperature prediction method according to claim 1, wherein Also included: Perform anomaly detection on the predicted battery temperature; When there is an anomaly in the predicted battery temperature, generate battery predicted temperature anomaly information, and perform fault handling according to the battery predicted temperature anomaly information.

9. The battery temperature prediction method according to claim 8, wherein, The performing of the anomaly detection on the predicted battery temperature includes: Obtain the actual battery temperature; Determine the residual according to the actual battery temperature and the predicted battery temperature; Determine the mean residual and the standard deviation of the residual according to the residual; Determine the anomaly threshold according to the residual, the mean residual, and the standard deviation of the residual; Determine the anomaly score according to the mean residual, the standard deviation of the residual, and the anomaly threshold, and the anomaly score is used to characterize the probability of battery temperature anomaly.

10. A battery temperature prediction system that uses the method according to any one of claims 1 to 9, characterized in that, The system includes: The first module is used to obtain the charging data of the battery periodically, and the charging data includes historical charging data and real-time charging data; The second module is used to determine historical derivative data according to the historical charging data; The third module is used to determine eigenvector data according to the historical charging data and the historical derivative data; The fourth module is used to determine model input samples under various charging conditions according to the eigenvector data; The fifth module is used to establish a deep autoregressive model according to the model input samples; The sixth module is used to determine the predicted battery temperature according to the real-time charging data and the deep autoregressive model; Among them, the determination of historical derivative data according to the historical charging data includes: determining the temperature lag term feature according to the historical single-cell charging temperature data of the battery and the preset lag term parameter.

Images