Automatic construction method of hybrid power battery data feature library and application method thereof

Abstract

The invention relates to the technical field of new energy vehicle battery data processing, in particular to an automatic construction method and application method of a hybrid power battery data feature library, and the method comprises the steps: obtaining vehicle data and environment data of a hybrid power vehicle, and extracting features; according to the features, the working condition of the hybrid power automobile is recognized through a constructed hybrid power working condition recognition model; extracting corresponding characteristic parameters according to the working conditions; setting a reference threshold value and a correction factor corresponding to each feature parameter, and constructing a feature library in combination with the working condition, the feature parameters and the reference threshold value and the correction factor corresponding to the working condition, the feature parameters and the reference threshold value. According to the scheme, the matching degree of the characteristic parameters and the actual working condition can be improved according to the working condition of the hybrid power automobile, the state change of the hybrid power battery is accurately reflected, and the high-precision requirement of battery management of the hybrid power automobile is met.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of new energy automobile battery data processing technology, in particular to an automatic construction method of a hybrid battery data feature library and an application method thereof. BACKGROUND

[0002] With the continuous development of the hybrid vehicle market, the demand for performance monitoring, fault diagnosis and energy management of hybrid batteries is increasingly urgent. The power system of a hybrid vehicle works cooperatively with an engine and a battery, and its working conditions are complex and diverse, with unique working states such as engine start-stop, energy recovery interruption, etc. However, current battery data analysis mainly focuses on battery data analysis for pure electric vehicles, such as identifying charging abnormalities through voltage and current curves, but does not involve hybrid-specific working conditions. Some technologies are aimed at energy flow analysis of hybrid systems to monitor power distribution between the engine and the motor, but lack automated feature extraction based on large-scale data. Therefore, the existing technology cannot meet the needs of hybrid vehicles for fine management of battery data, and it is difficult to accurately evaluate and effectively manage hybrid batteries under complex working conditions.

[0003] In summary, in the prior art, the construction of the battery feature library is mainly for pure electric vehicles or general three-electric systems, and the unique working conditions of the engine-battery cooperative work in hybrid vehicles (such as engine start-stop, energy recovery and engine driving switching, etc.) are not fully considered, resulting in low matching degree of feature parameters and actual working conditions, which cannot accurately reflect the state change of hybrid batteries and cannot meet the high-precision requirements of hybrid vehicle battery management.

[0004] Therefore, there is an urgent need for an automatic construction method of a hybrid battery data feature library and an application method thereof, which can improve the matching degree of feature parameters and actual working conditions for the working conditions of hybrid vehicles, accurately reflect the state change of hybrid batteries, and meet the high-precision requirements of hybrid vehicle battery management. SUMMARY

[0005] One of the purposes of the present application is to provide an automatic construction and application method of a hybrid battery data feature library, which can improve the matching degree of feature parameters and actual working conditions for the working conditions of hybrid vehicles, accurately reflect the state change of hybrid batteries, and meet the high-precision requirements of hybrid vehicle battery management.

[0006] The present application provides a basic scheme one: an automatic construction method of a hybrid battery data feature library, including the following contents: Obtain vehicle data and environmental data of a hybrid vehicle and extract features; wherein the features include: battery dynamic features, engine cooperative features, environmental and vehicle state features; According to the features, identify the working conditions of the hybrid vehicle through the constructed hybrid working condition recognition model; According to the working condition, the corresponding characteristic parameters are extracted; The reference threshold and the correction factor corresponding to each characteristic parameter are set, and the feature library is constructed in combination with the working condition, the characteristic parameters and the reference threshold and the correction factor corresponding thereto.

[0007] Beneficial effects: The vehicle data and the environmental data of the hybrid vehicle are acquired, and the features are extracted, wherein the features include the battery dynamic features, the engine coordination features, and the environmental and vehicle state features, to be input into the constructed hybrid working condition recognition model to recognize the working condition of the hybrid vehicle, so that the hybrid working condition recognition model can analyze the working condition of the hybrid vehicle from multiple aspects and accurately recognize the working condition of the hybrid vehicle; then, the corresponding characteristic parameters of the hybrid vehicle are extracted according to the working condition; and then, the feature library is constructed according to the extracted characteristic parameters, and the reference threshold and the correction factor of the characteristic parameters under each working condition are set to dynamically adjust the threshold of the characteristic parameters; the feature library can be applied subsequently. The present scheme can extract specific working conditions for the hybrid vehicle, adapt to the engine-battery coordination working condition of the hybrid vehicle, and further extract the characteristic parameters that are more suitable for the actual operation scenario, so as to accurately reflect the state change of the hybrid battery, and the state evaluation accuracy is improved by more than 20% compared with the general feature library; when the feature library is applied, the dynamic threshold can be determined through the characteristic parameters and the reference threshold and the correction factor corresponding thereto, and the real-time matching of the characteristic parameters and the working condition is realized through the hybrid working condition recognition model and the dynamically adjusted threshold of the characteristic parameters, thereby solving the lag problem of the traditional static threshold when the working condition is switched, and meeting the high-precision requirement of the battery management of the hybrid vehicle.

[0008] In summary, the present scheme can improve the matching degree of the characteristic parameters and the actual working condition for the working condition of the hybrid vehicle, accurately reflect the state change of the hybrid battery, and meet the high-precision requirement of the battery management of the hybrid vehicle.

[0009] Further, the battery dynamic features include: voltage change rate, current peak value, SOC change rate, temperature gradient; The engine coordination features include: current response delay time when the engine starts, ratio of engine speed to battery current, engine start-stop frequency; The environmental and vehicle state features include: environmental temperature, altitude, vehicle load, driving speed, accelerator pedal opening degree.

[0010] Beneficial effects: Multiple types of features can more comprehensively reflect the current working condition of the hybrid vehicle, so that the working condition recognition is more accurate.

[0011] Further, the hybrid working condition recognition model includes: The input layer is used to input a plurality of features; The hidden layer adopts a double-branch structure to process the working conditions with engine participation and the working conditions without engine participation, capture high-level features of the working conditions with engine participation and high-level features of the working conditions without engine participation in the output of the input layer, and generate original feature values by fusion; The output layer is used to transform the original feature values output by the hidden layer, output probability values of each hybrid working condition, and take the working condition with the largest probability as the recognition result.

[0012] Beneficial effects: The existing working condition recognition method identifies the driving mode (city, highway, etc.) based on the GPS and acceleration sensor, but does not study the hybrid working conditions (such as frequent start-stop scenarios). Therefore, in the present scheme, a hybrid working condition recognition model is constructed to recognize the working conditions specific to hybrid vehicles. In particular, the hidden layer is provided with a double-branch structure to process the working conditions with engine participation and the working conditions without engine participation, thereby adapting to the recognition of different working conditions of hybrid vehicles, and the recognized working conditions can improve the accuracy of recognition.

[0013] Further, the double-branch structure comprises: The first branch is a working condition branch with engine participation, which contains 64 neurons and adopts a ReLU activation function, is used to process the working conditions with engine participation, and has an input weight matrix with a dimension of 12x64 and a bias vector with a dimension of 1x64; and outputs high-level features of analysis. The second branch is a pure electric driving working condition branch, which contains 32 neurons and adopts a Tanh activation function, is used to process the working conditions without engine participation, and has an input weight matrix with a dimension of 12x32 and a bias vector with a dimension of 1x32; and outputs high-level features of analysis. An attention vector is further arranged in the hidden layer to fuse the high-level features output by the first branch and the high-level features output by the second branch to generate original feature values, wherein is the weight of the first branch, is the weight of the second branch, and The weight calculation method is: ; Wherein is the similarity between the output features of the th branch and the engine state features; When the engine is in the starting state, increases, decreases; when the engine is in the stopping state, increases, decreases.

[0014] Beneficial effects: The double-branch structure is analyzed separately for engine participation and non-participation (pure electric drive), different neurons and activation functions are constructed, and the unique working conditions of engine-battery cooperation in hybrid vehicles are fully considered, thereby improving the accuracy of working condition identification.

[0015] Further, the corresponding feature parameters are extracted according to the working conditions, including: If the working condition is cold start-battery assistance, the extracted feature parameters are the starting voltage drop amplitude And the starting current peak duration ; If the working condition is high-speed cruising-energy recovery, the extracted feature parameters are the recovery current and the speed synchronization coefficient And the recovery efficiency decay rate ; If the working condition is low-speed congestion-frequent start-stop, the extracted feature parameters are the start-stop cycle voltage fluctuation standard deviation And the ratio of start-stop frequency to SOC change rate per unit time ; If the working condition is rapid acceleration-engine dominated, the extracted feature parameters are the battery current mutation rate when the engine intervenes And the battery temperature rise rate during acceleration ; If the working condition is deceleration braking-energy recovery, the extracted feature parameters are the recovery voltage overshoot And the ratio of recovery energy to braking intensity ; If the working condition is idling-battery maintenance, the extracted feature parameters are the battery self-discharge rate during idling And the consistency coefficient of single battery voltage during idling .

[0016] Beneficial effects: The feature parameters extracted under each working condition are different, but the corresponding feature parameters can accurately reflect the state change of the hybrid battery under the current working condition, thereby better meeting the needs of hybrid vehicle battery management.

[0017] Further, the starting voltage drop amplitude Reflects the instantaneous power supply capability of the battery when the engine is cold started; the operation steps are to collect the battery voltage before starting for a first predetermined time And the lowest voltage after starting for a second predetermined time , the calculation formula is ; The starting current peak duration Reflects the stability of the battery during high-current discharge; the operation steps are to record the current reaching the peak value during the starting process the moment when the current drops to a preset percentage of the peak value the moment when the current drops to a preset percentage of the peak value , the calculation formula is ; the recovery current and vehicle speed synchronization coefficient , reflecting the matching degree of the energy recovery system and the vehicle driving state at high-speed cruising; the operation steps are to collect the vehicle speed and the recovery current for a continuous third preset time , and the calculation formula is: ; wherein is the average vehicle speed, is the average recovery current; the recovery efficiency decay rate , reflecting the change of energy recovery efficiency during long-time high-speed cruising; the operation steps are to divide the high-speed cruising period into multiple first preset time interval, calculate the energy recovery efficiency of each interval, and the calculation formula is: ; wherein is the number of intervals; the start-stop cycle voltage fluctuation standard deviation , reflecting the stability of the battery in the case of frequent start-stop leading to frequent fluctuation of battery voltage; the operation steps are to record the battery voltage data within a continuous start-stop cycle, calculate the voltage average value of each cycle, and then calculate the standard deviation of these average values, and the calculation formula is: ; wherein is the voltage average value of the nth cycle, is the average voltage of the start-stop cycle; the ratio of the number of starts and stops per unit time to the SOC change rate , reflecting the degree of influence of frequent start-stop on the battery SOC; the operation steps are to count the number of starts and stops within a second preset time period , calculate the SOC change rate within the second preset time period , and the calculation formula is ; ; the battery current mutation rate when the engine intervenes , reflecting the smoothness of the switching between the battery and the engine power at the moment when the engine intervenes during sudden acceleration; the operation steps are to collect the current for a fourth preset time before the engine interventionThe battery temperature rising rate in the acceleration process The current , the calculation formula is: ; The battery temperature rising rate in the acceleration process , reflecting the heating of the battery when the acceleration is urgent; the operation steps are to record the temperature at the beginning of the acceleration and the temperature at the end of the acceleration , and the duration of the acceleration , the calculation formula is: ; The recovery voltage overshoot , reflecting the acceptance ability of the battery to the recovery energy at the beginning of the deceleration brake; the operation steps are to collect the voltage before the recovery and the highest voltage in the recovery process , the calculation formula is: ; The ratio of the recovery energy to the braking intensity , reflecting the utilization efficiency of the energy recovery system to the braking energy; the operation steps are to calculate the recovery energy and the braking intensity in a single braking process, the calculation formula is: ; The battery self-discharge rate at idle speed , reflecting the leakage of the battery under idle state; the operation steps are to record the SOC at the beginning of the idle speed and the SOC after the idle speed lasts for hours , the calculation formula is: ; The consistency coefficient of the single battery voltage at idle speed , reflecting the voltage balance of each single battery under idle state; the operation steps are to collect the voltage of all single batteries , calculate the maximum voltage and the minimum voltage , the calculation formula is: ; Wherein is the average single voltage.

[0018] Beneficial effects: different calculations are carried out for corresponding characteristic parameters under different working conditions, so as to guarantee the accuracy of the extraction of the characteristic parameters, and the extracted characteristic parameters can more accurately reflect the condition of the hybrid battery under the current working condition.

[0019] Further, the feature library adopts a distributed database, wherein the feature parameters are stored according to a three-level index of a hybrid vehicle model, a driving mileage segment and a working condition type.

[0020] Beneficial effect: multi-level index storage facilitates data query and traceability.

[0021] Further, the correction factor is a dynamically changed value; and the reference threshold value and the correction factor are used to dynamically adjust the threshold value of the feature parameter, including: threshold value of feature parameter = reference threshold value x (1+ correction factor).

[0022] Beneficial effect: the threshold value of the feature parameter is dynamically adjusted, and based on the set reference threshold value, the threshold value is adjusted according to the correction factor, so as to adapt to the normal change of the feature parameter caused by the external environment or long-term use of the hybrid vehicle.

[0023] The second purpose of the present application is to provide an automatic construction and application system of a hybrid battery data feature library, which can improve the matching degree of the feature parameter and the actual working condition, accurately reflect the state change of the hybrid battery, and meet the high-precision requirement of the battery management of the hybrid vehicle.

[0024] The present application provides a basic scheme two: an application method of a hybrid battery data feature library, characterized in that the feature library is constructed by using the automatic construction method of the hybrid battery data feature library. Real-time acquisition of vehicle data of the hybrid vehicle and extraction of features; wherein the features include: battery dynamic features, engine coordination features, environment and vehicle state features; According to the features, the working condition of the hybrid vehicle is identified through the constructed hybrid working condition identification model; According to the identified working condition, the corresponding feature parameters are extracted, and the reference threshold value and the correction factor of the corresponding feature parameters are extracted in the feature library; According to the extracted reference threshold value and the correction factor, the threshold value of the feature parameter is determined; If the feature parameter exceeds the corresponding threshold value, a safety warning is performed.

[0025] Beneficial effects: the feature library constructed by the automatic construction method of the hybrid battery data feature library is applied, vehicle data of the hybrid vehicle is obtained in real time, features are extracted, and according to the features, the hybrid working condition recognition model is constructed, the working condition of the hybrid vehicle is recognized, according to the recognized working condition, the corresponding feature parameters are extracted, and the basic threshold and correction factor of the corresponding feature parameters in the feature library are extracted, according to the extracted basic threshold and correction factor, the threshold of the feature parameter is determined, if the feature parameter exceeds the corresponding threshold, safety warning is carried out, through the hybrid working condition recognition model and the dynamically adjusted threshold of the feature parameter, the real-time matching of the feature parameter and the working condition is realized, the lag problem of the traditional static threshold in the working condition switching is solved, so as to meet the high precision requirement of the hybrid vehicle battery management.

[0026] In summary, the scheme can improve the matching degree of feature parameters and actual working conditions according to the working conditions of hybrid vehicles, accurately reflect the state change of hybrid batteries, and meet the high precision requirement of hybrid vehicle battery management.

[0027] Further, it also includes predicting the cycle life of the hybrid battery according to the extracted features. The prediction of the cycle life of the hybrid battery includes: establishing a hybrid battery cycle life prediction model through the change trend of the SOC jump value of the energy recovery stage with the mileage, and the prediction formula is: ; Wherein is the predicted life, is the initial life, is the attenuation coefficient, is the total of the SOC jump value.

[0028] Beneficial effects: for the application of the constructed feature library, safety warning and life prediction can be carried out, so that the hybrid vehicle battery management can be better. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 is a flowchart of an embodiment of the automatic construction and application method of the hybrid battery data feature library. DETAILED DESCRIPTION

[0030] The following will be further described in detail through specific embodiments: Example 1 This embodiment is basically as shown in the accompanying Figure 1 , a method for automatically constructing a hybrid battery data feature library is provided, which includes the following contents: Vehicle data and environment data of the hybrid vehicle are acquired and features are extracted; specifically, the vehicle data and environment data of the hybrid vehicle are collected in real time through a vehicle-mounted sensor and a CAM bus; wherein the vehicle data includes but is not limited to: single-cell voltage, total current, temperature, SOC (state of charge) of the hybrid battery, engine start-stop signal, energy recovery state signal; the environment data includes but is not limited to: ambient temperature, altitude (GPS); The sampling frequency in the embodiment is 10 Hz; The collected vehicle data and environment data are preprocessed, including: eliminating abnormal data frames with time disorder, voltage or current mutation exceeding ± 5%; for missing data at the engine start-stop moment, an interpolation method based on adjacent frame trend is used for supplement; if the collected vehicle data and environment data are to be constructed into a data set, a working condition label is also generated, which is manually labeled in the embodiment; the working condition label includes: cold start-battery assistance, high-speed cruising-energy recovery, low-speed congestion-frequent start-stop, sudden acceleration-engine dominated, deceleration braking-energy recovery, idling-battery maintenance; Features are extracted from the preprocessed vehicle data and environment data; in the embodiment, the features are obtained by calculating the vehicle data and environment data; wherein the features are battery dynamic features, engine coordination features and environment and vehicle state features; The battery dynamic features include: voltage change rate, current peak value, SOC change rate and temperature gradient; The engine coordination features include: current response delay time when the engine starts, ratio of engine speed to battery current, and engine start-stop frequency; The environment and vehicle state features include: ambient temperature, altitude, vehicle load, driving speed and accelerator pedal opening degree; The extracted features are standardized; The min-max standardization formula is used for continuous features: ; Normalization is performed, wherein is the original feature value, is the normalized feature value, and are the minimum and maximum values of the feature respectively; Discrete features are one-hot encoded, with the start state encoded as [1, 0] and the stop state encoded as [0, 1]; Sliding window is used to intercept time series features, and in the embodiment, the sliding window size is set to 10 seconds and the step size is 5 seconds, generating a 10x12 feature matrix as an input sample.

[0031] According to characteristics, the working condition of the hybrid vehicle is identified through the constructed hybrid working condition identification model; The construction of the hybrid working condition identification model includes; The hybrid working condition identification model is constructed, trained and optimized, and the hybrid working condition identification model meeting the preset training target is obtained; The hybrid working condition identification model includes: The input layer is used for inputting several features and setting corresponding nodes; in this embodiment, 12 features of 3 categories are inputted, so 12 nodes are set; The hidden layer adopts a double-branch structure, respectively processes the working condition of the engine participating and the working condition of the engine not participating, captures the high-level features of the working condition of the engine participating and the high-level features of the working condition of the engine not participating in the output of the input layer, and generates the original feature value by fusion; Specifically, the first branch is the engine participating working condition branch, which contains 64 neurons, adopts the ReLU activation function, is used for processing the working condition involving the engine participating such as engine starting, engine driving and energy recovery switching, and the input weight matrix The dimension is 12x64, the bias vector The dimension is 1x64; the output is the analyzed high-level feature; The second branch is the pure electric driving working condition branch, which contains 32 neurons, adopts the Tanh activation function, is used for processing the working condition of the engine not participating such as pure electric driving and energy recovery, and the input weight matrix The dimension is 12x32, the bias vector The dimension is 1x32; the output is the analyzed high-level feature; An attention vector (the dimension is 1x2) is also set in the hidden layer, which is used for fusing the high-level features output by the first branch and the high-level features output by the second branch to generate the original feature value, wherein is the weight of the first branch, is the weight of the second branch, and The weight calculation method is: ; Wherein is the similarity between the output feature of the branch and the engine state feature; When the engine is in the starting state, is increased (usually 0.6-0.8), is reduced; when the engine is in the stopping state, is increased (usually 0.7-0.9), is reduced.

[0032] The output layer is used to transform the original feature values ​​output by the hidden layer, output the probability value of each hybrid-specific working condition, and take the working condition with the highest probability as the recognition result. The output layer has 6 nodes, each corresponding to one of 6 hybrid-specific operating conditions, providing full scenario coverage. It addresses the interaction modes between the battery and engine, including: cold start - battery assistance, high-speed cruising - energy recovery, low-speed congestion - frequent start-stop, rapid acceleration - engine dominance, deceleration and braking - energy recovery, and idling - battery sustainment. A Softmax activation function is used to output the probability values ​​for each hybrid-specific operating condition, using the following formula: in For the first The input of each node, For the first The input of each node, The output probability of a node is given by the node with the highest probability, which corresponds to the hybrid-specific operating condition and is the identification result.

[0033] The training and optimization process includes: Dataset Construction: In this embodiment, 100 hybrid vehicles were selected, covering different models, including plug-in hybrids and range-extended hybrids. 12 months of operating data were collected in various scenarios (such as urban roads, highways, and mountain roads), with a total data volume of 5 million records. Each data record contains 12 input features and corresponding operating condition labels (manually labeled). The data was divided into training set and validation set in a 7:3 ratio, with 3.5 million records in the training set and 1.5 million records in the validation set. Training parameter settings: The optimizer is Adam, the initial learning rate is set to 0.001, and it is adaptively adjusted with each training iteration. The learning rate is halved when the validation set loss does not decrease for 5 consecutive iterations. The batch size is set to 256, and the number of iterations is set to 1000. Training is stopped early when the validation set accuracy reaches 96% or higher and remains stable for 10 consecutive iterations. The loss function used is the cross-entropy loss function, with the formula: ; in For the sample size, For the first The first sample The true label of each working condition, i.e., the working condition label (1 indicates that it belongs to the working condition, and 0 indicates that it does not belong to the working condition). for The first sample The predicted probability of each working condition.

[0034] Model evaluation and optimization: using accuracy, precision, recall, and... Scores are used as evaluation indicators; The accuracy formula is: ; in It is a true positive. It is a true negative. It was a false positive. It is a false negative; For the low-speed congestion-frequent start-stop condition with low recognition accuracy, the number of training samples for this condition was increased (from 500,000 to 800,000), and the weight parameters of branch one were adjusted to increase the attention to the engine start-stop frequency characteristics. After 5 rounds of optimization, the recognition accuracy for this condition increased from 82% to 94%.

[0035] Based on the characteristics, the hybrid vehicle's operating conditions are identified through a hybrid operating condition recognition model. Specifically, the vehicle's battery data, engine data, and environmental data are collected in real time, preprocessed according to the input layer feature processing method, and a feature matrix is ​​generated. The feature matrix is ​​input into the trained model, and features are extracted through a hidden layer dual-branch structure. After fusion through an attention mechanism, the probability values ​​of each operating condition are output by the output layer. The operating condition with the highest probability is selected as the current recognition result and output to the feature parameter extraction module.

[0036] Extract the corresponding feature parameters based on the operating conditions; Specifically, if the operating condition is cold start-battery assisted, then the extracted feature parameter is the start-up voltage drop amplitude. and peak duration of startup current ; Among them, the start-up voltage drop During a cold start of the engine, the battery experiences a sudden large current discharge, causing a sharp drop in voltage, reflecting the battery's instantaneous power supply capability. The operating procedure involves collecting the battery voltage at the first preset time before starting. and the lowest voltage after the second preset time after startup The calculation formula is: In this embodiment, the first preset time is 1 second, and the second preset time is 0.5 seconds. Startup current peak duration This reflects the battery's stability during high-current discharge; an excessively long duration indicates a decline in battery performance. The operation procedure involves recording the peak current during startup. The moment and the moment when the current drops to a preset percentage of the peak value The calculation formula is: In this embodiment, the preset percentage is 80%.

[0037] If the working condition is high-speed cruising-energy recovery, the extracted characteristic parameters are recovery current and vehicle speed synchronism coefficient and recovery efficiency decay rate ; The recovery current and vehicle speed synchronism coefficient , at high-speed cruising, the energy recovery current should maintain a certain synchronism with the vehicle speed, reflecting the matching degree of the energy recovery system and the vehicle driving state; the operation steps are to collect the vehicle speed and recovery current for a continuous third preset time , and the calculation formula is: ; Among them is the average vehicle speed, is the average recovery current; in this embodiment, the third preset time is 30 seconds, so the corresponding calculation formula can be expressed as: ; The recovery efficiency decay rate reflects the change of energy recovery efficiency during long-time high-speed cruising, and a too fast decay indicates that the battery or the recovery system is abnormal; the operation steps are to divide the high-speed cruising period into multiple first preset time interval intervals, calculate the energy recovery efficiency of each interval, and the calculation formula is: ; Among them is the number of intervals; in this embodiment, the first preset time interval is 5 minutes.

[0038] If the working condition is low-speed congestion-frequent start-stop, the extracted characteristic parameters are start-stop cycle voltage fluctuation standard deviation and the ratio of start-stop frequency per unit time to SOC change rate ; The start-stop cycle voltage fluctuation standard deviation , in the frequent start-stop leading to frequent fluctuation of battery voltage, reflects the stability of the battery, and the larger the standard deviation, the worse the battery performance; the operation steps are to record the battery voltage data in the continuous start-stop cycle, calculate the average value of the voltage of each cycle, and then calculate the standard deviation of these average values, and the calculation formula is: ; Among them is the voltage average value of the th cycle, is the average voltage of the start-stop cycle; in this embodiment is 10, so the above formula can be expressed as: ; The ratio of the number of start-stop times per unit time to the SOC change rate , reflecting the influence degree of frequent start-stop on the battery SOC, and the ratio is too large indicating abnormal battery energy consumption; the operation steps are to count the number of start-stop times in the second preset time period , calculate the SOC change rate in the second preset time period , and the calculation formula is ; In this embodiment , the second preset time period is 1 hour.

[0039] If the working condition is sudden acceleration-engine dominated, the extracted characteristic parameters are the battery current mutation rate at the time of engine intervention and the battery temperature rise rate during acceleration ; Among them, the battery current mutation rate at the time of engine intervention , at the moment of engine intervention during sudden acceleration, the battery current will mutate, reflecting the smoothness of the power switching between the battery and the engine; the operation steps are to collect the current fourth preset time before engine intervention and the current fifth preset time after intervention , and the calculation formula is: ; The unit is , in this embodiment 0.5 seconds, 0.5 seconds; The battery temperature rise rate during acceleration , reflecting the heating condition of the battery during sudden acceleration, and the rate is too fast, which may exist a safety hazard; the operation steps are to record the temperature at the beginning of acceleration and the temperature at the end of acceleration, and the acceleration duration , and the calculation formula is: .

[0040] If the working condition is deceleration braking-energy recovery, the extracted characteristic parameters are the recovery voltage overshoot and the ratio of recovery energy to braking intensity ; Among them, the recovery voltage overshoot , at the beginning of deceleration braking, energy recovery may cause the battery voltage to overshoot instantaneously, reflecting the battery's ability to accept recovered energy; the operation steps are to collect the voltage and the highest voltage during the recycling process The calculation formula is: ; The ratio of recovered energy to braking intensity This ratio reflects the efficiency of the energy recovery system in utilizing braking energy; a higher ratio indicates better recovery performance. The operating procedure involves calculating the recovered energy during a single braking process. (Calculated by integrating current, voltage, and time) and braking intensity The average value of the brake pedal opening is calculated using the following formula: .

[0041] If the operating condition is idle-battery sustain, then the extracted feature parameter is the battery self-discharge rate at idle. Consistency coefficient of single cell voltage at idle speed ; Among them, the battery self-discharge rate at idle speed This reflects the battery's leakage current under idling conditions; an excessively high self-discharge rate indicates a battery malfunction. The operating procedure is to record the SOC (State of Charge) at the start of idling. and idling SOC after hours The calculation formula is: ; Single cell voltage consistency coefficient at idle speed This reflects the voltage balance of each individual battery cell under idling conditions; a smaller coefficient indicates poorer consistency. The operation procedure involves collecting the voltage of all individual battery cells. ( , (Number of individual units), calculate the maximum voltage. and minimum voltage The calculation formula is: ; in This represents the average unit voltage.

[0042] Set the baseline thresholds and correction factors for each feature parameter, and construct a feature library by combining the operating conditions, feature parameters and their corresponding baseline thresholds and correction factors; The feature library uses a distributed database, such as HBase or MongoDB, and the data structure stores feature parameters according to a three-level index based on hybrid vehicle model, mileage range (every 50,000 kilometers), and operating condition type. The correction factor is a dynamically changing value, the reference threshold corresponding to each characteristic parameter and the correction factor are used to dynamically adjust the threshold of the characteristic parameter, and the threshold of the characteristic parameter is equal to the reference threshold multiplied by (1+the correction factor); and the reference threshold and the correction factor of different characteristic parameters are different, and the correction factors change in different ways; For example, in the cold start-battery assisted working condition, the threshold of the voltage difference is calculated according to the following formula: ; Wherein is the threshold of the voltage difference, is the reference value, is the correction factor of the ambient temperature, and the coefficient is 0.2 when the ambient temperature is-20℃; When the driving mileage exceeds 100,000 kilometers, the threshold of the start-stop interval-voltage fluctuation coefficient is increased by 15%, so the correction factor is 15%, and the threshold of the start-stop interval-voltage fluctuation coefficient is calculated according to the following formula: ; Wherein is the threshold of the adjusted start-stop interval-voltage fluctuation coefficient, is the reference threshold, is the correction factor.

[0043] In addition, if a preset number of new working conditions are identified, the statistical update of the characteristic parameters in the feature library is triggered; Specifically, in the embodiment, the preset number is 1000, that is, the statistical update of the characteristic parameters in the feature library is automatically triggered every 1000 new working conditions, such as updating the quartile interval of the threshold. If a new working condition type is identified, such as an extremely low temperature-engine forced start, a subset of the characteristic parameters is automatically added in the feature library.

[0044] The embodiment also provides an application method of the hybrid battery data feature library, which uses the feature library constructed by the automatic construction method of the hybrid battery data feature library. Vehicle data of the hybrid vehicle are acquired in real time, and features are extracted; wherein the features include: battery dynamic features, engine coordination features, and environment and vehicle state features. According to the features, a hybrid working condition recognition model is constructed to recognize the working condition of the hybrid vehicle. According to the recognized working condition, corresponding characteristic parameters are extracted, and reference thresholds and correction factors of the corresponding characteristic parameters are extracted in the feature library. According to the extracted reference thresholds and correction factors, the threshold of the characteristic parameter is determined. If the characteristic parameter exceeds the corresponding threshold value, a safety warning is given; for example, when the current peak attenuation rate at the start-up moment exceeds the threshold value, a battery aging warning is output; According to the extracted characteristics, the cycle life of the hybrid battery is predicted.

[0045] Wherein the cycle life of the hybrid battery is predicted, comprising: establishing a hybrid battery cycle life prediction model through the change trend of the SOC jump value of the energy recovery stage with the mileage, and the prediction formula is: ; Wherein is the predicted life, is the initial life, is the attenuation coefficient, is the total of the SOC jump value. The feature library provides the core data basis for life prediction, and the threshold value provides a reference basis for the effectiveness of the prediction result and the battery health state judgment.

[0046] Embodiment two This embodiment is basically the same as the above-mentioned embodiment, the difference is: it also includes: verification and screening of characteristic parameters: for each extracted characteristic parameter, 1000 groups of data of different battery health states (new battery, used for 1 year, used for 3 years) are collected; the difference of the characteristic parameters under different health states is tested by using the analysis of variance (ANOVA) method, if , it indicates that the parameter can effectively distinguish the battery health state, the parameter is retained, otherwise it is rejected; the correlation analysis is carried out on the retained characteristic parameters, and the redundant parameters with a correlation coefficient greater than 0.8 are removed to ensure the independence of the parameters. In this embodiment, through the above steps, finally 12 characteristic parameters (2 for each working condition) are retained.

[0047] The above-mentioned is only an embodiment of the present application, and the specific structure and characteristics of the scheme which are well known are not described too much, the ordinary skilled in the art knows all the ordinary technical knowledge in the field of the application before the application date or the priority date, can know all the prior art in the field and has the ability to apply the conventional experimental means before the date, the ordinary skilled in the art can improve and implement the scheme under the inspiration of this application, some typical known structures or known methods should not be an obstacle for the ordinary skilled in the art to implement the present application. It should be pointed out that for those skilled in the art, without departing from the structure of the present application, a number of modifications and improvements can be made, which should be regarded as the protection scope of the present application, which will not affect the effect and practicality of the patent. The protection scope of the present application should be subject to the content of its claims, and the specific implementation mode in the specification can be used to explain the content of the claims.

Claims

1. An automated method for constructing a hybrid battery data feature library, characterized in that, Includes the following: Acquire vehicle and environmental data of hybrid vehicles and extract features; these features include: battery dynamic features, engine coordination features, and environmental and vehicle state features. Based on the characteristics, the operating conditions of hybrid vehicles are identified through the constructed hybrid operating condition identification model; Extract the corresponding feature parameters based on the operating conditions; Set the baseline thresholds and correction factors for each feature parameter, and construct a feature library by combining the operating conditions, feature parameters and their corresponding baseline thresholds and correction factors.

2. The method for automated construction and application of the hybrid battery data feature library according to claim 1, characterized in that, The battery dynamic characteristics include: voltage change rate, peak current, SOC change rate, and temperature gradient; The engine cooperative features include: current response delay time during engine start-up, engine speed to battery current ratio, and engine start-stop frequency. The environmental and vehicle status characteristics include: ambient temperature, altitude, vehicle load, driving speed, and accelerator pedal opening.

3. The method for automated construction and application of the hybrid battery data feature library according to claim 1, characterized in that, The hybrid operating condition identification model includes: The input layer is used to input several features; The hidden layer adopts a dual-branch structure to process the operating conditions in which the engine participates and the operating conditions in which the engine does not participate, respectively. It captures the high-level features of the operating conditions in which the engine participates and the high-level features of the operating conditions in which the engine does not participate in the output of the input layer, and fuses them to generate the original feature values. The output layer transforms the original feature values ​​output by the hidden layer, outputs the probability value of each hybrid-specific operating condition, and uses the operating condition with the highest probability as the recognition result.

4. The method for automated construction and application of the hybrid battery data feature library according to claim 3, characterized in that, The dual-branch structure includes: The first branch is the transmitter participation condition branch, containing 64 neurons and using the ReLU activation function to handle the transmitter participation condition. The input weight matrix is... The dimension is 12×64, and the bias vector is... The dimension is 1×64; the output is the high-level features of the analysis. The second branch is the pure electric drive mode branch, containing 32 neurons and using the Tanh activation function to handle the mode where the engine is not involved. The input weight matrix is... The dimension is 12×32, and the bias vector is... The dimension is 1×32; the output analysis shows high-level features. Attention vectors are also set in the hidden layer. This is used to fuse the high-level features from the first branch output and the high-level features from the second branch output to generate the original feature values, where The weight of the first branch, Let be the weight of the second branch, and The weights are calculated as follows: ； in For the first The similarity between the output features of each branch and the engine state features; When the engine is running Increase Reduce; when the engine is stopped. Increase Decrease.

5. The method for automated construction and application of the hybrid battery data feature library according to claim 1, characterized in that, The step of extracting corresponding feature parameters based on the working conditions includes: If the operating condition is cold start-battery assisted, then the extracted feature parameter is the start-up voltage drop amplitude. and peak duration of startup current ; If the operating condition is high-speed cruising with energy recovery, the extracted characteristic parameter is the synchronization coefficient between the recovery current and vehicle speed. and recovery efficiency decay rate ; If the operating condition is low-speed congestion with frequent start-stop cycles, then the extracted feature parameter is the standard deviation of voltage fluctuation during the start-stop cycle. The ratio of the number of start-stop cycles per unit time to the rate of change of SOC. ; If the operating condition is rapid acceleration with engine dominance, then the extracted feature parameter is the battery current mutation rate when the engine intervenes. and the rate of battery temperature rise during acceleration ; If the operating condition is deceleration braking-energy recovery, then the extracted characteristic parameter is the recovery voltage overshoot. The ratio of recovered energy to braking intensity ; If the operating condition is idle-battery sustain, then the extracted feature parameter is the battery self-discharge rate at idle. Consistency coefficient of single cell voltage at idle speed .

6. The method for automated construction and application of the hybrid battery data feature library according to claim 5, characterized in that, The start-up voltage drop The operation steps are as follows: collect the battery voltage at the first preset time before startup. and the lowest voltage after the second preset time after startup The calculation formula is: ; The peak duration of the starting current The operation steps are to record the peak current during startup. The moment and the moment when the current drops to a preset percentage of the peak value The calculation formula is: ; The synchronization coefficient between the recovered current and the vehicle speed The operation steps are as follows: collect data for the third consecutive preset time. speed and recycled current The calculation formula is: ； in Average vehicle speed The average recovery current; The recovery efficiency decay rate The operation steps are as follows: divide the high-speed cruising period into multiple first preset time intervals, and calculate the energy recovery efficiency of each interval. The calculation formula is: ； in The number of intervals; The standard deviation of voltage fluctuation during the start-stop cycle The operation steps are to record consecutively The battery voltage data during each start-stop cycle is used to calculate the average voltage for each cycle, and then the standard deviation of these average values ​​is calculated using the following formula: ； in For the first The average voltage over one cycle, for Average voltage over one start-stop cycle; The ratio of the number of starts and stops per unit time to the SOC change rate The operation steps are as follows: Statistical analysis of the second preset time period The number of start-stop cycles within a given time period is used to calculate the rate of change of SOC within the second preset time period. The calculation formula is: ； The battery current mutation rate during engine intervention The operation steps are as follows: collect data at the fourth preset time before engine intervention. current and the fifth preset time after intervention current The calculation formula is: ； The rate of battery temperature rise during the acceleration process The operation steps are to record the temperature at the start of acceleration. and the temperature at the end of acceleration and the duration of acceleration The calculation formula is: ； The recovery voltage overshoot The operating steps are as follows: voltage is collected before the recovery begins. and the highest voltage during the recycling process The calculation formula is: ； The ratio of the recovered energy to the braking intensity The operating procedure is to calculate the recovered energy during a single braking process. and braking strength The calculation formula is: ； The battery self-discharge rate at idle speed The operating procedure is to record the SOC at the start of idling. and idling SOC after hours The calculation formula is: ； The voltage consistency coefficient of individual cells at idle speed The operation steps are as follows: collect the voltage of all individual cells. Calculate the maximum voltage and minimum voltage The calculation formula is: ； in This represents the average cell voltage.

7. The automated construction method for a hybrid battery data feature library according to claim 1, characterized in that, The feature library uses a distributed database, where the data structure stores feature parameters according to a three-level index of hybrid vehicle model, mileage range, and operating condition type.

8. The automated construction method for a hybrid battery data feature library according to claim 1, characterized in that, The correction factor is a dynamically changing value; The baseline threshold and correction factor dynamically adjust the threshold of the feature parameter, including: the threshold of the feature parameter = baseline threshold × (1 + correction factor).

9. A method for applying a hybrid battery data feature library, characterized in that, The feature library is constructed using the automated construction method for the hybrid battery data feature library as described in any one of claims 1-8; Real-time acquisition of vehicle data for hybrid vehicles and extraction of features; these features include: battery dynamic features, engine coordination features, and environmental and vehicle status features. Based on the characteristics, the operating conditions of hybrid vehicles are identified through the constructed hybrid operating condition identification model; Based on the identified working conditions, extract the corresponding feature parameters, and extract the basic threshold and correction factor of the corresponding feature parameters from the feature library; Based on the extracted base threshold and correction factor, determine the threshold of the feature parameters; If the feature parameter exceeds the corresponding threshold, a security warning will be issued.

10. The application method of the hybrid battery data feature library according to claim 9, characterized in that, Also includes: Based on the extracted features, predict the cycle life of the hybrid battery; The prediction of hybrid battery cycle life includes: establishing a hybrid battery cycle life prediction model based on the trend of SOC jump value during the energy recovery stage with mileage, and the prediction formula is as follows: ； in To predict lifespan, For the initial lifespan, The attenuation coefficient is... This is the sum of SOC jump values.

Images