An intelligent cascade utilization method for power lithium batteries based on car-connected cloud
By using the methods of Vehicle Cloud and deep learning algorithms in the cascade utilization of power lithium batteries, the SOH/SOC prediction model of power lithium batteries was established, and the problems of predicting the health status and life of power lithium batteries were solved, and the timely recovery of power lithium batteries was achieved. and cascade utilization have improved environmental protection benefits.
Patent Information
- Application Number
- CN202210318678.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-03-29
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2042-03-29
AI Technical Summary
The prior art is difficult to effectively predict the health status and life of power lithium batteries, resulting in the complex and long-term utilization process of power lithium batteries, and lacks time-consuming and environmental protection.
Using an intelligent cadrical utilization method based on Vehicle Cloud, we can collect driving environment information and power lithium battery related data in real time, establish SOH/SOC prediction model, and use deep learning algorithms such as GRU-RNN and CNN to perform data fusion and feature learning. Achieve life prediction and cascade utilization of power lithium batteries.
Effectively combining Internet of Vehicles technology and deep learning algorithms, a data-driven power lithium battery health prediction model is established to realize the timely recycling and cascade utilization of power lithium batteries, and improve environmental protection benefits.
Smart Images

Figure CN114976306B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of vehicle networking and new energy joint technology, and particularly relates to an intelligent cascade utilization method of power lithium batteries based on vehicle networking cloud. Background Art
[0002] As new energy vehicles become more and more popular among consumers, the recycling and reuse of retired power lithium batteries has become a hot topic of research today.
[0003] Due to the different working environments of power lithium batteries, there are differences in the remaining battery capacity and life after retirement. In the process of cascade utilization of power lithium batteries, we need to classify power lithium batteries according to the remaining battery capacity. The test and performance evaluation methods proposed by the current research, such as pulse test, constant power test and constant current test, only consider the characteristics of power lithium batteries under static environment, and cannot predict their health status. As for the research on the life of power lithium batteries, some factors that affect the life of power lithium batteries, such as charge and discharge rate, ambient temperature, etc., are mainly explained. It also considers how to reasonably and standardizedly use power lithium batteries under multiple influencing factors to better increase the utilization time of power lithium batteries. However, it does not judge the health status information such as SOC / SOH of power lithium batteries based on the analysis of power lithium battery data, and lacks rationality in the classification of power lithium batteries. Although some power lithium battery test rules have been formulated, obtaining a large amount of test data cannot derive the health status law of power lithium batteries. At the same time, the whole process is complicated and time-consuming, requiring a lot of manpower, which is not conducive to the timely recovery and cascade utilization of power lithium batteries. In the current research, there are still problems such as offline power lithium battery pack, variable external test environment, and limited test time.
[0004] The present invention adopts a combination of online and offline methods, and proposes an intelligent cascade utilization method for power lithium batteries based on the car-connected cloud. The SOH / SOC prediction model of the power lithium battery is established from the historical data and field test data of the power lithium battery, and the SOH / SOC health status information of the power lithium battery is estimated in real time. And according to the obtained SOH / SOC, the life prediction and cascade utilization of the power lithium battery are realized. Summary of the invention
[0005] The purpose of the present invention is to overcome the defects and shortcomings in the recycling and utilization technology of new energy power lithium batteries, and to propose an intelligent cascade utilization method of power lithium batteries based on the car-connected cloud.
[0006] In order to achieve the above-mentioned purpose, the technical solution adopted by the present invention is a method for intelligent cascade utilization of power lithium batteries based on car-connected cloud, characterized in that it includes the following steps:
[0007] (1) Real-time collection of driving environment information, including road scene information, vehicle status information, and vehicle location information;
[0008] (2) collecting relevant data of the power lithium battery in real time, wherein the relevant data of the power lithium battery includes temperature, voltage and current information;
[0009] (3) uploading the driving environment information and the relevant data of the power lithium battery to the cloud, and pre-processing the driving environment information and the relevant data of the power lithium battery to form a driving environment history database and a power lithium battery history database;
[0010] (4) Building a power lithium battery test experimental platform to obtain test data, the test data including dynamic internal resistance, static capacity, HPPC and OCV of the power lithium battery;
[0011] (5) constructing a deep learning model for estimating the SOH / SOC of the power lithium battery, and using the driving environment history database, the power lithium battery history database and the test data as training data for feature learning based on the data fusion technology of the deep learning model to establish a power lithium battery SOH / SOC prediction model;
[0012] (6) The relevant data of the power lithium battery collected in real time is input into the power lithium battery SOH / SOC prediction model, and the SOH / SOC of the power lithium battery is estimated in real time to achieve life prediction and cascade utilization of the power lithium battery.
[0013] Furthermore, the driving environment information is obtained by a vehicle-mounted camera, a vehicle-mounted bus module and a vehicle positioning module. The vehicle-mounted camera collects the road scene information, the vehicle-mounted bus module collects the vehicle status information, and the vehicle positioning module collects the vehicle position information.
[0014] Furthermore, the relevant data of the power lithium battery is obtained by real-time detection by the vehicle online sensor device.
[0015] Furthermore, the driving environment information and relevant data of the power lithium battery are uploaded to the cloud through the DSRC wireless communication module of the V2X system.
[0016] Furthermore, the preprocessing of the driving environment information and the relevant data of the power lithium battery in step 3 includes:
[0017] (3.1) Using a convolutional neural network (CNN) to perform semantic segmentation on the road scene information;
[0018] (3.2) Processing the road scene information after semantic segmentation: including data review and screening, image scaling, data normalization, and image enhancement;
[0019] The data review and screening mentioned above: when reviewing data, errors in the data are discovered and corrected, and erroneous data or problematic data that cannot be corrected are eliminated;
[0020] Image scaling: adjusting the dimension value of the data by scaling so that the final data tensor is within the range of [0, 1] or [-1, 1];
[0021] The data normalization includes sample-by-sample mean reduction and feature normalization; the sample-by-sample mean reduction removes the average brightness value of the image; the feature normalization makes each dimension of the data have zero mean and unit variance;
[0022] Image enhancement: including shape change, cropping and color adjustment
[0023] (3.3) Tagging the driving environment information and the relevant data of the power lithium battery, wherein the road scene information is divided into a spatial database and an attribute database; the spatial database is the spatial location information of the road, including two spatial entities: points and lines; the attribute database is the road surface feature information, including road name and road width;
[0024] (3.4) Using ArcGIS to build a database of the driving environment information and the power lithium battery related data; obtaining the driving environment history database and the power lithium battery history database.
[0025] Furthermore, the semantic segmentation of the road scene information using a convolutional neural network (CNN) includes:
[0026] (3.1.1) Annotating the road scene information with 8 key points and filtering the road scene information, and dividing the filtered road scene information into a training set and a test set;
[0027] (3.1.2) Obtain a probability distribution map of the eight key points, take the maximum value of the probability distribution map, and obtain the key point coordinates in the pixel coordinate system;
[0028] (3.1.3) The detection model selects the residual network ResNet50 model for feature detection, with a video frame of size 256*256 as input and a feature map of size 32*32 as output;
[0029] (3.1.4) The feature map is input into a regression module consisting of two layers of the convolutional neural network CNN and optimized using the Euclidean distance loss function with L2 regularization:
[0030]
[0031] Where m is the number of key points, P i is the coordinate of the i-th key point predicted by the network, GT i represents the coordinates of the manually annotated i-th key point, w t are neural network parameters;
[0032] (3.1.5) After multiple iterations and parameter adjustments on the training set, select the model with the best effect on the validation set;
[0033] (3.1.6) Use the trained convolutional neural network (CNN) model to complete semantic segmentation of the road scene information.
[0034] Furthermore, the convolutional neural network CNN consists of a fully connected layer, four convolutional blocks, an input layer and an output layer; the relationship between the input A and the output B is: B = F(A|Θ) = f L {…f2[f1(X|θ1)|θ2]|θ L}.
[0035] Furthermore, the power lithium battery test experimental platform in step 4 is composed of host computer software, a computer, a constant temperature box and a lithium battery charge and discharge test system;
[0036] The dynamic internal resistance test: the dynamic internal resistance is tested by intermittent discharge method;
[0037] The static capacity test is obtained by calculating the average value of multiple static capacity tests;
[0038] The HPPC and OCV tests: a variety of pulse current combinations are used to test HPPC and OCV.
[0039] Further, the step 5 includes the following process:
[0040] (5.1) constructing a deep learning model for estimating the SOH / SOC of the power lithium battery, wherein the deep learning model is a recurrent neural network GRU-RNN based on a gated recurrent unit.
[0041] (5.2) inputting the voltage, current, temperature in the power lithium battery history database and the information in the driving environment history database into the gated recurrent unit recurrent neural network GRU-RNN to estimate the corresponding SOC / SOH under the dynamic scene;
[0042] (5.3) The dynamic internal resistance, static capacity, HPPC, and OCV in the test data are input into the gated recurrent unit recurrent neural network GRU-RNN to estimate the corresponding SOC / SOH in the static scenario;
[0043] (5.4) After multiple training iterations, the training of the deep learning model is completed.
[0044] Furthermore, the gated recurrent unit recurrent neural network GRU-RNN consists of 1 input layer, 1 hidden layer, 1 fully connected layer and 1 output layer; the forward propagation process of the gated recurrent unit recurrent neural network GRU-RNN is as follows:
[0045] z t =σ(w z ·[h t-1 , x t ]+b z )
[0046] r t =σ(w r ·[h t-1 , x t ]+b r )
[0047]
[0048]
[0049] Among them, h t-1 is the output of the hidden layer node in the previous sequence, x t is the hidden layer node input of the current sequence, h t is the hidden layer node output of the current sequence, z t and r t are the outputs of the update gate and the reset gate respectively, w is the weight parameter, and b is the bias parameter.
[0050] Beneficial effects of the present invention:
[0051] 1. The present invention can effectively combine Internet of Vehicles technology and deep learning algorithms, and adopt an online and offline approach to establish a data-driven power lithium battery health prediction model to facilitate the cascade utilization, timely recycling and environmental protection of retired power lithium batteries of new energy vehicles.
[0052] 2. The main advantage of the joint estimation model is that GRU-RNN has a good data feature extraction technology with high accuracy, while CNN has fewer parameters and fast calculation speed. The use of this joint estimation model can better realize the life prediction and cascade utilization of power lithium batteries. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] Figure 1 This is a flow chart of the intelligent cascade utilization method of power lithium batteries based on the car-connected cloud of the present invention.
[0054] Figure 2 The functional design block diagram of the V2X vehicle-mounted system of the present invention.
[0055] Figure 3 This is a combined estimation model of SOC and SOH for a power lithium battery of the present invention.
[0056] Figure 4 This is a schematic diagram of the CNN network semantic segmentation of the present invention.
[0057] Figure 5 This is a process diagram of cascading utilization of power lithium batteries by combining vehicle-connected cloud data and data fusion technology in the present invention.
[0058] Figure 6 It is a bar chart of the cascade utilization of power lithium batteries in the present invention. DETAILED DESCRIPTION
[0059] The intelligent method for recycling power lithium batteries based on the car-connected cloud of the present invention will be further described in detail below in conjunction with specific embodiments.
[0060] Figure 1 This is a flow chart of the intelligent cascade utilization method of power lithium batteries based on the car-connected cloud of the present invention, which mainly includes the following steps:
[0061] S101: collecting driving environment information in real time, wherein the driving environment information includes road scene information, vehicle status information and vehicle position information;
[0062] Road scene information is collected through the on-board camera, and the GPS module collects vehicle location information; the on-board bus module collects vehicle status information; vehicles interact with each other or with roadside units through vehicle networking communications; each vehicle interacts with the driving environment information it has collected through vehicle networking communications to confirm driving safety. Clearly collecting driving environment information can facilitate the formation of a historical database of driving environments and can also confirm driving safety. At the same time, collecting road scene information can determine the road conditions, such as uphill and downhill. The vehicle status and vehicle location information obtained when the vehicle is driving uphill and downhill are different, and these need to be collected. Collecting vehicle status information and vehicle location information can obtain the temperature, voltage, current data, etc. required by the power lithium battery in dynamic scenes when the vehicle accelerates, decelerates, goes uphill, goes downhill, etc. After uploading these data to the cloud to form a historical database, they can be used as training sets for feature learning.
[0063] S102: Collecting relevant data of the power lithium battery in real time, wherein the relevant data of the power lithium battery includes temperature, voltage and current information;
[0064] The temperature, voltage and current of the power lithium battery pack are detected in real time through the on-board online sensor device to obtain relevant data of the power lithium battery when the vehicle is driving. As the vehicle is driving, in dynamic (online) scenarios, the on-board online sensor device can detect the temperature, voltage and current of the power lithium battery in real time. When the vehicle accelerates, decelerates, goes uphill or downhill, the temperature, voltage and current required by the power lithium battery can be detected by its online sensor device. Specifically, the on-board main control module determines the current driving state of the vehicle according to the opening of the accelerator pedal and the brake pedal. When it is found that the opening of the accelerator pedal becomes larger, the vehicle is in the normal driving stage, and the power lithium battery supplies power to the motor. When it is found that the opening of the brake pedal becomes larger, the vehicle enters the deceleration stage, and the motor uses the braking energy of the vehicle to generate electricity and stores the energy in the energy storage device. When the current generated during braking reaches a certain threshold, the power lithium battery meets the charging conditions. In this process, the main control module decides whether to perform brake energy recovery based on the SOC of the power lithium battery and the current vehicle speed. If energy is to be recovered, the on-board main control module sends relevant instructions to the motor controller to recover part of the braking energy. During driving, the vehicle main control module sends power requirements to the battery controller via the CAN line, and collects temperature, voltage and current data of the power lithium battery. These data are fed back to the vehicle main control module via the CAN line.
[0065] S103: uploading the driving environment information and the relevant data of the power lithium battery to the cloud, and preprocessing the driving environment information and the power lithium battery information to form a driving environment history database and a power lithium battery history database;
[0066] Data is uploaded to the cloud via the DSRC wireless transmission module of the V2X vehicle network. Figure 2 This is a functional diagram of each module of the V2X vehicle networking system. The vehicle-mounted system mainly consists of three parts: data acquisition, information interaction, and data upload. The data acquisition part is the power lithium battery data previously collected by the online sensor device of the power lithium battery, and the road scene data collected by the vehicle-mounted camera. The information interaction part includes V2V communication and V2I communication. When the vehicle enters the radio frequency range of the roadside basic communication unit, the collected data is uploaded to the cloud through the DSRC link. The V2X vehicle networking system mainly includes a CAN bus module, a DSRC wireless communication module, a GPS positioning module, a main control module, etc. The CAN bus module is used to transmit the acquired vehicle status data to the V2X vehicle-mounted system, such as acceleration, steering and braking current data.
[0067] In the cloud, the convolutional neural network (CNN) is first used to semantically segment the road scene information in the driving environment information. The road scene in the driving environment affects the driving state of the vehicle in a dynamic scene, thereby affecting the use and consumption of the power lithium battery, and indirectly determines the real-time collection of power lithium battery related data. High-precision real-time segmentation can be completed efficiently with fewer layers.
[0068] The convolutional neural network CNN consists of a fully connected layer, four convolutional blocks, an input layer and an output layer. We set the input and output of the convolutional neural network to A and B, and the relationship between the input and output can be expressed as shown in the formula:
[0069] B=F(A|Θ)=f L {…f2[f1(X|θ1)|θ2]|θ L}.
[0070] Convolutional neural network CNN can output several channels of feature maps with smaller resolution, which is equal to the size of the previous image. A part of the decoder is connected after the encoder. The feature maps of each layer in the network structure of convolutional neural network CNN are as follows: Figure 4 As shown. The previously collected road scene sample images are annotated with 8 key points. The training data set includes training images and test images. The test images can be used as a test set to test the model after the model is built. Cut and rotate the annotated images. Mirror transformation can be used to enrich the data set. Manually screen out valid frames, and remove frames with serious vehicle occlusion and a small number of key points where the annotation difference exceeds 30% of the pixel. The accuracy of the training data annotation is ensured by cross-validation by multiple annotators. ResNet50 is selected as the feature detection model, and the corresponding key point detection model should be built for the reference key point detection network. The basic framework of the vehicle key point detection module uses the convolutional neural network CNN. After learning through the feature extraction network, a feature map of size 32*32 is output, and the feature map is input into the regression module. The regression module consists of two layers of convolutional neural network CNN. The network is optimized by the Euclidean distance loss function with L2 regularization, as shown in the following formula:
[0071]
[0072] Where m is the number of preset key points of the vehicle, p i is the coordinate of the i-th key point predicted by the network, GT i represents the coordinates of the manually annotated key point. i are neural network parameters.
[0073] After that, the obtained driving environment information and the relevant data of the power lithium battery are preprocessed, including data review and screening, image scaling, data normalization, and image enhancement. When reviewing the data, the problems and errors in the data should be corrected. Then the data is screened and the data that cannot be corrected is directly deleted. The normalization of the data consists of simple scaling, sample-by-sample mean reduction, and feature standardization. After simple scaling and adjusting the dimension value of the data, the final data tensor is within the range of [0,1] or [-1,1]. In image processing, the sample-by-sample mean reduction operation removes the average brightness value of the image. Feature standardization can make each dimension of the data have zero mean and unit variance. Image enhancement methods generally include shape change, cropping, and color adjustment. Road scene data in the driving environment can be classified into spatial databases and attribute databases. In the spatial database, the spatial position of the road is two spatial entities, points and lines, so that the spatial position of the road can be more accurately represented. The attribute database contains road surface feature information, such as road name, road width, etc. This type of information can represent all information about the road. ArcGIS is used to collect and build a database of driving environment information and power lithium battery information data. The data is converted to ArcGIS format and then sorted and classified. The ArcTool conversion method is used to convert different CAD layers into feature classes, and an empty PersonalGeoDatabase is created using the function module ArcCatalog. The database is renamed and the spatial data is imported using the wizard. Before entering the attribute data, the corresponding attribute structure must be established first. Finally, the driving environment history database and the power lithium battery history database can be obtained.
[0074] S104: Building a power lithium battery test experimental platform to obtain test data, where the test data includes dynamic internal resistance, static capacity, HPPC, and OCV of the power lithium battery;
[0075] The power lithium battery pack test experimental platform is mainly composed of host computer software, computer, constant temperature box and lithium battery charge and discharge test system. The power lithium battery charge and discharge test system collects power lithium battery data and transmits it to the computer through the CAN bus, and the host computer software visualizes the data. The power lithium battery charge and discharge test system is equipped with 8 independent control test channels, which can test 8 groups of power lithium batteries at the same time. Each test channel is connected to the power lithium battery with a special fixture, and the maximum sampling frequency of each channel is 3HZ; the temperature acquisition accuracy is ±0.5℃; the current measurement accuracy is ±0.2%; the voltage measurement accuracy is ±0.1%. The host software can receive and upload the data obtained from the power lithium battery test in real time, process the historical data of the power lithium battery, and analyze and display the basic performance of the power lithium battery. During the test, the power lithium battery is first placed in a constant temperature box with a temperature control range of -10℃~100℃ and a temperature control accuracy of ±0.1℃. Then the dynamic internal resistance, static capacity, HPPC, and OCV of the power lithium battery are tested, and the field test data are obtained.
[0076] The dynamic internal resistance is tested by intermittent discharge. The effective capacity of the power lithium battery is obtained by calculating the average value of the static capacity tested five times. A variety of pulse current combinations are used to test HPPC and OCV. The power lithium battery is placed in a constant temperature box and allowed to stand for 1 hour to allow the power lithium battery to reach thermal equilibrium. The power lithium battery is charged to the maximum value with a standard current of 0.5C constant current, and then five discharge current rates of 0.3C, 0.5C, 3C, 3.5C, and 2.6C are selected to discharge to the cut-off voltage. The standing time after charging and discharging is more than 1 hour to obtain OCV, and the voltage after 1h of standing is used as OCV.
[0077] S105: constructing a deep learning model for estimating the SOH / SOC of the power lithium battery, and using the driving environment database, the power lithium battery history database and the test data as training data to perform feature learning based on the data fusion technology of the deep learning model, to establish a power lithium battery SOH / SOC prediction model;
[0078] A deep learning model based on a gated recurrent unit recurrent neural network GRU-RNN is constructed to estimate the SOH / SOC of the power lithium battery. The gated recurrent unit recurrent neural network GRU-RNN consists of 1 input layer, 1 hidden layer, 1 fully connected layer and 1 output layer. The forward propagation process of GRU-RNN is expressed as follows:
[0079] z t =σ(w z ·[h t-1 , x t ]+b z )
[0080] rt =σ(w r ·[h t-1 , x t ]+b r )
[0081]
[0082]
[0083] Among them, h t-1 is the output of the hidden layer node in the previous sequence, x t is the hidden layer node input of the current sequence, h t is the hidden layer node output of the current sequence, z t and r t are the outputs of the update gate and the reset gate respectively, w is the weight parameter, and b is the bias parameter.
[0084] The implementation process of the power lithium battery SOC / SOH joint estimation model is as follows:
[0085] (1) Build a deep learning model for power lithium battery SOC / SOH estimation, set the objective function and optimization algorithm, set the number of training times, sampling interval, and initialize the network weights and bias;
[0086] (2) Input the voltage, current, temperature in the power lithium battery history database and the information in the driving environment history database into the gated recurrent unit recurrent neural network GRU-RNN to estimate the corresponding SOC / SOH under dynamic scenarios;
[0087] (3) The dynamic internal resistance, static capacity, HPPC, and OCV in the test data are input into the gated recurrent unit recurrent neural network GRU-RNN to estimate the corresponding SOC / SOH in the static scenario;
[0088] (4) After multiple training iterations, the training of the deep learning model is completed.
[0089] S106: Inputting the relevant data of the power lithium battery collected in real time into the power lithium battery SOH / SOC prediction model, estimating the SOH / SOC of the power lithium battery in real time, and realizing life prediction and cascade utilization of the power lithium battery.
[0090] After the model training is completed, the actual test set data is input into the model to quickly and accurately estimate the SOH / SOC of the power lithium battery. And based on the obtained SOH / SOC, the life prediction and cascade utilization of the power lithium battery are realized. Figure 5 and Figure 6As shown in the figure, the new power lithium batteries with the same capacity originally have different capacities after long-term use. After car-connected cloud data analysis, on-site rapid testing and model prediction, the power lithium battery capacity differences are reduced by using data fusion, and they are allocated to electric devices, energy storage devices, and low-end customers according to the remaining capacity. Those with 0 remaining capacity are dismantled and recycled. The online historical data and offline field test data are used as the training set of the deep learning joint estimation model composed of the convolutional neural network CNN and the gated recurrent unit recurrent neural network GRU-RUU. The data fusion technology of the deep learning joint estimation model is used to train and output the power lithium battery SOH / SOC parameters. After the model training is completed, the actual power lithium battery data can be input to quickly and accurately estimate the SOH / SOC of the power lithium battery, thereby realizing the life prediction and cascade utilization of the power lithium battery.
[0091] The present invention can effectively combine Internet of Vehicles technology and deep learning algorithms to establish a data-driven power lithium battery health prediction model, which facilitates the cascade utilization, timely recycling and environmental protection of retired power lithium batteries of new energy vehicles.
[0092] The specific embodiments of the present invention are described above in conjunction with the accompanying drawings. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art may make various modifications or variations within the scope of the claims, which do not affect the essence of the present invention.
Claims
1. An intelligent cascade utilization method of power lithium batteries based on car-connected cloud, characterized in that: The following steps are involved: (1) Real-time collection of driving environment information, including road scene information, vehicle status information, and vehicle location information; (2) collecting relevant data of the power lithium battery in real time, wherein the relevant data of the power lithium battery includes temperature, voltage and current information; (3) uploading the driving environment information and the relevant data of the power lithium battery to the cloud, and pre-processing the driving environment information and the relevant data of the power lithium battery to form a driving environment history database and a power lithium battery history database; (3.1) Using a convolutional neural network (CNN) to perform semantic segmentation on the road scene information; (3.2) Processing the road scene information after semantic segmentation: including data review and screening, image scaling, data normalization and image enhancement; The data review and screening mentioned above: when reviewing data, errors in the data are discovered and corrected, and erroneous data or problematic data that cannot be corrected are eliminated; Image scaling: adjusting the dimension value of the data by scaling so that the final data tensor is within the range of [0, 1] or [-1, 1]; The data normalization includes sample-by-sample mean reduction and feature normalization; the sample-by-sample mean reduction removes the average brightness value of the image; the feature normalization makes each dimension of the data have zero mean and unit variance; The image enhancement includes shape change, cropping and color adjustment; (3.3) Tagging the driving environment information and the relevant data of the power lithium battery, wherein the road scene information is divided into a spatial database and an attribute database; the spatial database is the spatial location information of the road, including two spatial entities: points and lines; the attribute database is the road surface feature information, including road name and road width; (3.4) Using ArcGIS to build a database of the driving environment information and the relevant data of the power lithium battery to obtain the driving environment history database and the power lithium battery history database; (4) Building a power lithium battery test experimental platform to obtain test data, the test data including dynamic internal resistance, static capacity, HPPC and OCV of the power lithium battery; (5) constructing a deep learning model for estimating the SOH / SOC of the power lithium battery, and using the driving environment history database, the power lithium battery history database and the test data as training data for feature learning based on the data fusion technology of the deep learning model to establish a power lithium battery SOH / SOC prediction model; (5.1) constructing a deep learning model for estimating the SOH / SOC of the power lithium battery, wherein the deep learning model is a recurrent neural network GRU-RNN based on a gated recurrent unit; (5.2) inputting the voltage, current, temperature in the power lithium battery history database and the information in the driving environment history database into the gated recurrent unit recurrent neural network GRU-RNN to estimate the corresponding SOC / SOH in a dynamic scenario; (5.3) The dynamic internal resistance, static capacity, HPPC, and OCV in the test data are input into the gated recurrent unit recurrent neural network GRU-RNN to estimate the corresponding SOC / SOH in the static scenario; (5.4) After multiple training iterations, the training of the deep learning model is completed; (6) The relevant data of the power lithium battery collected in real time is input into the power lithium battery SOH / SOC prediction model, and the SOH / SOC of the power lithium battery is estimated in real time to achieve life prediction and cascade utilization of the power lithium battery.
2. The intelligent cascade utilization method of power lithium batteries based on the car-connected cloud according to claim 1 is characterized in that: The driving environment information is obtained by a vehicle-mounted camera, a vehicle-mounted bus module and a vehicle positioning module. The vehicle-mounted camera collects the road scene information, the vehicle-mounted bus module collects the vehicle status information, and the vehicle positioning module collects the vehicle position information.
3. The intelligent cascade utilization method of power lithium batteries based on the car-connected cloud according to claim 1 is characterized in that: The relevant data of the power lithium battery is obtained by real-time detection by the vehicle online sensor device.
4. The intelligent cascade utilization method of power lithium batteries based on the car-connected cloud according to claim 1 is characterized in that: The driving environment information and relevant data of the power lithium battery are uploaded to the cloud through the DSRC wireless communication module of the V2X system.
5. The intelligent cascade utilization method of power lithium batteries based on car-connected cloud according to claim 1 is characterized in that: The using of a convolutional neural network (CNN) to perform semantic segmentation on the road scene information comprises: (3.1.1) Annotating the road scene information with 8 key points and filtering the road scene information, and dividing the filtered road scene information into a training set and a test set; (3.1.2) Obtain a probability distribution map of the eight key points, take the maximum value of the probability distribution map, and obtain the key point coordinates in the pixel coordinate system; (3.1.3) The detection model selects the residual network ResNet50 model for feature detection, with a video frame of size 256*256 as input and a feature map of size 32*32 as output; (3.1.4) The feature map is input into a regression module consisting of two layers of the convolutional neural network CNN and optimized using the Euclidean distance loss function with L2 regularization: Where m is the number of key points, P i is the coordinate of the i-th key point predicted by the network, GT i represents the coordinates of the manually annotated i-th key point, w t are neural network parameters; (3.1.5) using the training set for multiple iterations and adjusting parameters, and then selecting the model with the best effect on the test set; (3.1.6) Use the trained convolutional neural network (CNN) model to complete semantic segmentation of the road scene information.
6. The intelligent cascade utilization method of power lithium batteries based on car-connected cloud according to claim 1 is characterized in that: The convolutional neural network CNN consists of a fully connected layer, four convolutional blocks, an input layer and an output layer. The relationship between input A and output B is: B = F(A|Θ) = f L {…f2[f1(X|θ1)|θ2]|θ L }.
7. The intelligent cascade utilization method of power lithium batteries based on car-connected cloud according to claim 1 is characterized in that: The power lithium battery test experimental platform in step 4 is composed of host computer software, computer, constant temperature box and lithium battery charge and discharge test system; The dynamic internal resistance test is conducted by intermittent discharge. The static capacity test is performed by calculating and testing the static capacity multiple times and obtaining an average value; The HPPC and OCV tests: a variety of pulse current combinations are used to test the HPPC and OCV.
8. The intelligent cascade utilization method of power lithium batteries based on the car-connected cloud according to claim 1 is characterized in that: The gated recurrent unit recurrent neural network GRU-RNN consists of 1 input layer, 1 hidden layer, 1 fully connected layer and 1 output layer; the forward propagation process of the gated recurrent unit recurrent neural network GRU-RNN is as follows: z t =σ(w z ·[h t-1 ,x t ]+b z ) r t =σ(w r [h t-1 ,x t ]+b r ) Among them, h t-1 is the output of the hidden layer node in the previous sequence, x t is the hidden layer node input of the current sequence, h t is the hidden layer node output of the current sequence, z t and r t are the outputs of the update gate and the reset gate respectively, w is the weight parameter, and b is the bias parameter.
Citation Information
Patent Citations
Lithium battery state-of-health estimation method based on geometrical characteristics of charging voltage curve
CN112098873A
Lithium battery safety degree estimation method and device based on random forest
CN113466713A
Battery data adjusting method and battery management unit manufacturing method as well as battery management unit and server
JP2021092464A
