CN112904219B - Big data-based power battery health state prediction method - Google Patents

Abstract

The invention discloses a method for predicting the health state and the residual life of a power battery based on big data, which comprises the following steps: 1, acquiring a large amount of real-time running condition data from an electric vehicle, and cleaning and filling the data; 2, extracting four working condition characteristic values of temperature, speed, current and mileage value by processing discharge data to express increase of equivalent cycle times; 3, processing the charging data to obtain an accurate average capacity curve of the battery; 4, obtaining the relation between the SOH and the equivalent cycle times based on a battery charge-discharge characteristic experiment, and establishing a battery health state evaluation model based on the driving condition-charging calculation; and 5, constructing an integrated neural network machine learning model by taking the characteristic working condition as input and the difference of the equivalent cycle times as output, thereby realizing accurate estimation and prediction of the SOH of the battery.

Description

A prediction method of power battery health status based on big data

technical field

The invention is applied in the field of electric vehicles, and is specifically a method for predicting the state of health of a power battery based on big data, which is suitable for accurate estimation of the state of health of a battery vehicle.

Background technique

In recent years, with the rapid development of lithium-ion battery technology, the electric vehicle industry has gradually entered a new stage. Battery state of health (SOH, State Of Health) estimation, as one of the key technologies in the battery management system (BMS, Battery Management System), plays a vital role in the mileage and life prediction of electric vehicles. However, since the battery is a highly nonlinear electrochemical system, the identification and estimation of its internal state is still a huge problem.

Since the SOH cannot be directly measured, in order to accurately measure the SOH of the battery, a large number of SOH estimation methods have been proposed. The most commonly used methods are model-based estimation methods and big data-based estimation methods. The model-based estimation method simulates the internal working principle of the battery by establishing an equivalent circuit, and combines corresponding algorithms, such as particle filter, Kalman filter, sliding mode observer, etc., to estimate SOH, but this type of method relies too much on model accuracy, and Algorithm design is more complicated.

Contents of the invention

In order to solve the shortcomings of the above-mentioned prior art, the present invention proposes a prediction method of power battery health status based on big data, in order to avoid the modeling problem and parameter identification problem in the model-based estimation method, so as to realize Accurate estimation and prediction of battery SOH.

The present invention adopts following technical scheme in order to achieve the above-mentioned purpose of the invention:

A method for predicting the state of health of a power battery based on big data of the present invention is characterized in that it comprises the following steps:

Step 1: Collect real-time driving condition data on the electric vehicle, including: electric vehicle speed, accumulated mileage, voltage data, current data, battery state of charge data and temperature data;

Step 2, data preprocessing;

Step 2.1. Clean the real-time driving condition data respectively, and remove the points with large errors in the data, so as to obtain the effective driving condition data set;

Step 2.2, based on the real-time driving condition data of the electric vehicle, a temperature model is established, which is used to fill the temperature data in the effective driving condition data set, so as to obtain evenly distributed temperature data;

Step 2.3, fitting the change curve of the cumulative mileage in the effective driving condition data set;

Step 3, battery capacity calculation during charging;

Step 3.1, using the current data and battery state-of-charge data after preliminary data cleaning to calculate the average capacity data of the battery in a single state-of-charge;

Step 3.2, taking temperature data and charging current data as input, taking the average capacity of the battery in the single state of charge as output, constructing a random forest regression model, and bringing the standard temperature and standard current into the random forest regression model Perform calculations to obtain standard values;

Step 3.3. Input the evenly distributed temperature data and the current data in the effective driving condition data set into the random forest regression model for calculation, obtain the original battery average capacity value, and make a difference from the standard value, and The obtained difference is added as a gain to the battery average capacity data, thereby obtaining the regression-processed battery average capacity data;

Step 3.4, drawing the battery average capacity data after random forest model regression processing into a scatter diagram;

Step 3.5, performing cluster fitting on the scatter diagram to obtain the charging capacity curve;

Step 4. Establish a battery health status evaluation model based on driving conditions-charging calculations;

Step 4.1, using the charging capacity curve, combined with formula (1) to calculate the battery state of health SOH _t of the battery on day t:

In formula (1), cap _t is the battery capacity on day t, and cap _r is the rated capacity of the battery;

Step 4.2, convert the battery state of health SOH _t of the battery on day t into the equivalent cycle number value of the battery, and measure the battery life of the car in one day by calculating the difference between it and the equivalent cycle number value on day t-1. Consumed battery life criteria;

Step 4.3, perform feature extraction on the discharge data in the effective driving condition data set, obtain the average speed, equivalent mileage value, average temperature, and average current value and use them as the driving condition characteristics of the electric vehicle for the discharge process description of the working conditions;

Step 5. Establishment and training of the machine learning model;

The driving condition feature is used as the input of the machine learning model, the difference of the equivalent number of cycles is used as the output of the machine learning model, and the effective driving condition data set is brought into the machine as a training set In the learning model, a trained machine learning model is obtained;

Step 6, output prediction results;

Perform feature random sampling processing on the effective driving condition data set, obtain the sampled driving condition data, and input it into the trained machine learning model for prediction calculation, obtain the prediction result of the equivalent number of cycles, and according to etc. The mapping relationship between the effective cycle times and the battery health status is obtained to obtain the final predicted battery health status.

The method for predicting the state of health of the power battery based on big data of the present invention is also characterized in that: the conversion steps of the equivalent number of cycles in the step 4.2 are as follows:

Step 4.2.1, based on the battery charge and discharge characteristic test, obtain the relationship curve of the equivalent number of cycles of the battery and the state of health SOH of the battery, thereby according to the relationship curve, establish the relationship shown in formula (2):

In formula (2), a ₁ and b ₁ are the attenuation factors of the set electrode materials, a ₂ and b ₂ are the attenuation factors of the set electrolyte materials; x represents the equivalent cycle number of the battery;

Step 4.2.2, use the genetic algorithm to identify the parameters of the four attenuation factors in formula (1), and obtain the values of the four attenuation factors, so as to obtain the mapping relationship between the final equivalent cycle number and the battery state of health SOH.

The steps of the characteristic random sampling processing of the step 6 are as follows:

Step 6.1, respectively calculating the daily driving condition characteristics of the electric vehicle in the effective driving condition data set, and the maximum and minimum values of each driving condition characteristic in a day;

Step 6.2, determine the sampling time, if there is at least one piece of data corresponding to the sampling time in the valid driving condition data set, then randomly generate a number from the maximum and minimum values of the day as the characteristic value of the driving condition of the day; if If there is no data corresponding to the sampling time in the effective driving condition data set, the day closest to the sampling time is selected to perform the random number generation operation; if there is only one piece of data corresponding to the sampling time in the effective driving condition data set, then Use the corresponding data as the characteristic value of the driving condition;

Step 6.3, repeatedly execute step 6.2, so as to obtain the sampled driving condition data.

Compared with prior art, the beneficial effect of the present invention is:

1. The method of the present invention overcomes the difficulty of estimating the state of the battery, utilizes the big data of the real-time operation of the electric vehicle, and selects appropriate working condition characteristics, establishes a battery health state evaluation model based on the driving condition-charging calculation, and combines the machine The learning algorithm digs out the change law of the battery health state, can accurately estimate and predict the SOH value of the battery, has high precision, strong robustness, and is easy to implement.

2. The present invention establishes the relationship between equivalent cycle times and SOH through battery charging and discharging experiments, and indirectly estimates and predicts battery SOH, which is intuitive and accurate.

3. The algorithm of the present invention has a simple structure, does not require additional equipment, and only needs to use BMS and vehicle sensors to collect real-time data, and can realize accurate prediction of battery SOH through algorithm programming.

Description of drawings

Fig. 1 is the overall algorithm block diagram of the method for predicting the state of health of a power battery based on big data in the present invention;

Fig. 2 is the relational graph of SOH and equivalent number of cycles established by the present invention;

Fig. 3 is the integrated neural network schematic diagram that the present invention uses;

Fig. 4 is the genetic algorithm flow chart that the present invention uses;

Fig. 5 is the scatter diagram of battery capacity before and after random forest naturalization in the present invention;

Fig. 6 is the curve diagram of the battery capacity obtained by the cluster fitting algorithm along with the vehicle mileage in the present invention;

FIG. 7 is a histogram of battery SOH distribution predicted by a machine learning algorithm in the present invention after one year.

Detailed ways

In this embodiment, a method for predicting the health state of a power battery based on big data is to establish a battery health state evaluation model based on driving conditions-charging calculations by using the big data of electric vehicles running in real time, combined with machine learning algorithms, Accurately estimate and predict the SOH value of the battery, thereby solving the problem of difficult estimation and low accuracy of the battery SOH; at the same time, the relationship between the battery SOH and the equivalent cycle number is established by using the battery charge and discharge experiment, and indirectly estimated by the equivalent cycle number The battery SOH makes the prediction method not only practical, but also highly accurate, and intuitive; specifically, as shown in Figure 1, the method is carried out as follows:

Step 1: In this embodiment, the real-time operating data on the electric vehicle, the vehicle speed, accumulated mileage, voltage data, current data, battery state of charge data and temperature data of the electric vehicle are collected through the on-board sensor and BMS;

Step 2, preprocessing of battery charging and discharging data;

Step 2.1. Clean the battery charge and discharge data respectively, and remove the points with large errors in the data, so as to obtain an effective battery charge and discharge data set;

Step 2.2, based on the data of the real-time driving condition of the electric vehicle, fill in the temperature data; in this embodiment, adopt the automatic segmentation fitting method to realize the segmentation of the temperature, adopt the root mean square error as shown in formula (1) The formula controls the polynomial accuracy of each segment, thus establishing a temperature model, and realizing the filling of temperature data in the data;

In formula (1), y _i represents the approximate value of the temperature in the piecewise polynomial, y represents the actual value of the temperature in the piecewise polynomial, and N represents the number of temperature sampling points on a certain day.

Step 2.3, fitting the change curve of the accumulated mileage in the valid driving condition data set.

Step 3. Calculation of battery capacity during charging:

The capacity of the battery is mainly affected by SOH, charging current and temperature. In order to obtain a more accurate change of capacity with SOH, it is necessary to normalize the temperature and current of all data to the same condition, so as to ensure that they will not interfere with the battery. This algorithm, so this embodiment performs data processing through the following steps.

Step 3.1, using the current data and battery state-of-charge data after preliminary data cleaning to calculate the average capacity data of the battery in a single state-of-charge;

Step 3.2, taking temperature data and charging current data as input, and taking the average capacity of the battery in a single state of charge as output, construct a random forest regression model, and bring the standard temperature and standard current into the random forest regression model for calculation, and get standard value;

Step 3.3. Input the evenly distributed temperature data and the current data in the effective driving condition data set into the random forest regression model for calculation, obtain the average capacity value of the original battery, and make a difference with the standard value, and use the obtained difference as the gain and Add the battery average capacity data to obtain the battery average capacity data after regression processing;

Step 3.4, drawing the battery average capacity data after random forest model regression processing into a scatter diagram; the capacity scatter diagram before and after processing is shown in Figure 2;

Step 3.5, perform cluster fitting on the scatter diagram, and obtain the final curve of charging capacity changing with mileage, as shown in FIG. 3 .

In this embodiment, the charge and discharge characteristic experiment was carried out on the battery, and the corresponding battery data was obtained, and the relationship curve between the number of charge and discharge cycles x and SOH of the battery was obtained, as shown in Figure 4, which is used to establish a charging cycle based on driving conditions. Computational battery state of health evaluation model. The specific steps are as follows:

Step 4. Establish a battery health status evaluation model based on driving conditions-charging calculations;

Step 4.1, using the curve of charging capacity changing with mileage, combined with formula (2) to obtain the battery state of health value SOH _t on day t;

In formula (2), cap _t is the battery capacity on day t, and cap _r is the rated capacity of the battery.

Step 4.2, combined with formula (3), convert the SOH of the battery into the equivalent cycle value of the battery, and measure the battery consumed by the car in one day by calculating the difference between it and the equivalent cycle value on day t-1 life standard;

In formula (3), a ₁ and b ₁ are the attenuation factors of the set electrode materials, a ₂ and b ₂ are the attenuation factors of the set electrolyte materials, and x represents the equivalent cycle number of the battery;

Step 4.3, use the genetic algorithm to identify the parameters of the four attenuation factors in formula (1), and obtain the values of the four attenuation factors, so as to obtain the final mapping relationship between the equivalent number of cycles and the battery state of health SOH.

Genetic Algorithm (GA) is a heuristic algorithm that simulates the natural evolution process of organisms to search for the optimal solution. It transforms the problem to be solved into the process of chromosome gene selection, crossover, mutation, and recombination in biological evolution, and then obtains the problem the optimal solution of . This method is used to solve the problem in step 4.3, and the parameters to be identified can be quickly obtained with high precision. The logic block diagram is shown in Figure 5.

Step 4.4. Extract the features of the discharge data in the effective driving condition data set, obtain the average speed, equivalent mileage value, average temperature, and average current value, and use them as the driving condition characteristics of the electric vehicle for the working condition of the discharge process. Describe the situation; where the equivalent mileage value is divided by the single-day mileage divided by the SOH value.

Step 5. Establishment and training of machine learning model based on integrated neural network

In order to make the model have good learning and generalization capabilities, an integrated neural network is used in this example to build a machine learning model. The artificial neural network has the advantages of high classification accuracy, strong learning ability, low sensitivity to noise interference data, good generalization and expansion ability, and can approach any nonlinear relationship. Its structure is shown in Figure 6. The integrated artificial neural network refers to the process of combining multiple simple neural networks into a classifier based on the artificial neural network. It can overcome the problem that the model is prone to dead spots and divergence due to insufficient training data sets. , using an ensemble learning method to strengthen the stability of the network. The specific implementation steps are as follows:

The characteristics of driving conditions are taken as the input of the machine learning model, the difference of the equivalent number of cycles is taken as the output of the machine learning model, and the effective driving condition data set is brought into the machine learning model as a training set, so as to obtain a good training result. machine learning model.

Step 6. Output prediction results

In this example, random sampling of features is performed on valid battery charge and discharge data sets through the following steps:

Step 6.1, respectively calculate the daily driving condition characteristics of the electric vehicle in the effective driving condition data set, and the maximum and minimum values of each driving condition characteristic in a day;

Step 6.2, determine the sampling time, if there is at least one piece of data corresponding to the sampling time in the valid driving condition data set, then randomly generate a number from the maximum and minimum values of the day as the characteristic value of the driving condition of the day; if If there is no data corresponding to the sampling time in the effective driving condition data set, the day closest to the sampling time is selected to perform the random number generation operation; if there is only one piece of data corresponding to the sampling time in the effective driving condition data set, then Use the corresponding data as the characteristic value of the driving condition;

Step 6.3, repeatedly execute step 6.2, so as to obtain the driving condition data set after feature random sampling.

The data set after feature random sampling is input into the trained machine learning model for prediction calculation, and the prediction result of the equivalent number of cycles is obtained, and according to formula (3), the final predicted battery state of health value SOH is obtained.

In this example, the battery state of health value SOH predicted by the machine learning algorithm after one year is displayed by the SOH distribution histogram with a total frequency of 100, as shown in Figure 7.

Claims (3)

1. A big data-based power battery state of health prediction method is characterized by comprising the following steps:

the method comprises the following steps: gather the real-time operating mode data that traveles on the electric motor car, include: the speed, accumulated mileage, voltage data, current data, battery state of charge data and temperature data of the electric automobile;

step two, data preprocessing;

step 2.1, respectively cleaning the real-time running condition data, and removing points with larger errors in the data to obtain an effective running condition data set;

2.2, establishing a temperature model based on real-time driving condition data of the electric automobile, wherein the temperature model is used for filling temperature data in the effective driving condition data set so as to obtain uniformly distributed temperature data;

step 2.3, fitting a change curve of the accumulated mileage in the effective driving condition data set;

step three, calculating the battery capacity in the charging process;

step 3.1, calculating to obtain average capacity data of the battery in a single charge state by using the cleaned current data and the cleaned charge state data of the battery;

3.2, taking charging current data contained in the current data in the uniformly distributed temperature data and effective driving condition data set as input, taking the average capacity of the battery in the single charge state as output, constructing a random forest regression model, and substituting the standard temperature and the standard current into the random forest regression model for calculation to obtain a standard value;

step 3.3, charging current data contained in the uniformly distributed temperature data and the current data in the effective driving condition data set are input into the random forest regression model for calculation to obtain an original battery average capacity value, the original battery average capacity value is subjected to difference with the standard value, and the obtained difference is used as gain to be added with the battery average capacity data, so that the battery average capacity data after regression processing are obtained;

step 3.4, drawing the average capacity data of the battery subjected to the regression processing of the random forest model into a scatter diagram;

step 3.5, performing cluster fitting on the scatter diagram to obtain a charging capacity curve;

step four, establishing a battery health state evaluation model based on driving condition-charging calculation;

step 4.1, calculating the battery state of health (SOH) of the battery on the t day by using the charging capacity curve and combining the formula (1) _t ：

In the formula (1), cap _t The battery capacity at day t, cap _r Is the rated capacity of the battery;

step 4.2, the SOH of the battery on the t day _t Converting the equivalent cycle number value into an equivalent cycle number value of the battery, and calculating the difference value between the equivalent cycle number value and the t-1 day equivalent cycle number value to measure the service life standard of the battery consumed by the automobile in one day;

4.3, performing characteristic extraction on discharge data contained in the current data in the effective running condition data set to obtain an average speed, an equivalent running mileage value, an average temperature and an average current value, and using the average speed, the equivalent running mileage value, the average temperature and the average current value as running condition characteristics of the electric automobile for describing the working condition of the discharge process;

establishing and training a machine learning model;

taking the driving condition characteristics as the input of the machine learning model, taking the difference value of equivalent cycle times as the output of the machine learning model, and taking the effective driving condition data set as a training set to be brought into the machine learning model, thereby obtaining the trained machine learning model;

step six, outputting a prediction result;

and performing characteristic random sampling processing on the effective driving condition data set to obtain sampled driving condition data, inputting the sampled driving condition data into a trained machine learning model for prediction calculation to obtain a prediction result of equivalent cycle times, and obtaining a final predicted battery health state according to a mapping relation between the equivalent cycle times and the battery health state.

2. The big data based power battery state of health prediction method of claim 1, characterized in that: the conversion steps of equivalent cycle times in the step 4.2 are as follows:

step 4.2.1, obtaining a relation curve of the equivalent cycle number of the battery and the SOH of the battery based on a battery charging and discharging characteristic test, and establishing a relation shown as a formula (2) according to the relation curve:

in the formula (2), a ₁ 、b ₁ Is a set attenuation factor of the electrode material, a ₂ 、b ₂ A set attenuation factor for the electrolyte material; x represents the equivalent cycle number of the battery;

and 4.2.2, performing parameter identification on the four attenuation factors in the formula (2) by using a genetic algorithm to obtain values of the four attenuation factors, so as to obtain a final mapping relation formula of the equivalent cycle number and the SOH (state of health) of the battery.

3. The big data based power battery state of health prediction method of claim 1, characterized in that: the feature random sampling processing of the step six comprises the following steps:

6.1, respectively calculating the running condition characteristics of the electric vehicle in the effective running condition data set every day and the maximum value and the minimum value of each running condition characteristic in one day;

step 6.2, determining sampling time, and if at least one piece of time data corresponding to the sampling time exists in the effective running condition data set, randomly generating a number from the maximum value and the minimum value of the current day as a running condition characteristic value of the current day; if the effective running condition data set does not have data corresponding to the sampling time, taking the day closest to the sampling time to execute random number generation operation; if only one piece of time data corresponding to the sampling time exists in the effective driving condition data set, taking the corresponding data as a driving condition characteristic value;

and 6.3, repeatedly executing the step 6.2 to obtain the sampled running condition data.

Images