On December 23, 2020, China’s steam research successfully held the “2020 3rd New Energy Automobile Test Evaluation Technology International Forum”. China Steam Research will continue to push a wonderful speech, this paper, this article, “New Energy Automotive Safety Early Warning Algorithm Development and Test Evaluation” brought by the Minister of Data Products for China.
1. Development opportunities and challenges of new energy vehicles
As of June 2020, my country’s new energy car ownership has exceeded 4 million vehicles, accounting for more than 50% of the global ownership of new energy vehicles. According to the New Energy Automotive Industry Development Plan (2021-2035), in 2025, the sales volume of new energy vehicles will reach 20% of the total sales of automobile new cars. It is estimated that by 2035, my country’s new energy car insurance will exceed 160 million. At the same time, new energy cars faces 3 anxiety including mileage, security, protection, with the increase in the increase of the proportion and mileage improvement, but the security problem will be accompanied by the development of new energy vehicles. The investigation found that the focus of new energy vehicle safety issues is reflected in the battery, including in driving, collision, standing, charging, flooding, etc. New energy car safety is a full-industry issue, government, enterprises, industry and the public give a lot of attention. The government introduces relevant documents clearly proposes to strengthen supervision; enterprises strengthen their research and development capabilities; industry focus safety warning research.
2. New energy vehicle safety risk control main direction
Industry New Energy Automobile Safety Risk Control Technology is mainly expanded from two directions: First, the real-time security risk identification based on bicycle BMS (based on bicycle BMS real-time collection data, car enterprises use boundaries based on battery safety, formulate corresponding data control And alarm strategy); Second, cloud security risk identification based on cloud big data platform (based on historical data uploaded by the batch car, according to battery safety use boundaries and key security feature parameters historical data change regulations, etc.) . There are different data requirements and technical research directions, bicycles are based on real-time enterprise logistics data analysis, and the identification of battery safety uses boundary and thermal out-of control boundary conditions, but the risk identification is more lag, identifying security risk in advance Larger, focusing on processing burst security issues, prompting vehicle or driver making protection actions; batching vehicles based on historical data change law, achieving early identification and warning of security risks, but different systems and safety use boundaries Differences, influence precision, focusing on the advancement identification and early warning of safety risk vehicles, guidance car companies have made warning and maintenance. Both are relevant research and development in the industry, and China’s steam research has also carried out important work in new energy vehicle safety warning.
With the trust and support of the State Ministry of Industry and Information Technology, the Ministry of Science and Technology, the Market Supervision Administration, China Steam Research has carried out four national new energy automobile safety issues, gradually completed the construction of database construction, data analysis, model development, platform construction, Software develops all aspects of capacity building, and ultimately forms China’s auto research unique technical advantages. During the development of safety warning algorithm, it was found that there were two pain points: one is that the model verification test is difficult, and as the industry has gradually increased new energy automobile safety early warning algorithm model research, the verification cost of the model validity High; second, accident data is not used. Relying on a long-term accident vehicle data research on China’s steam research, the common risk characteristics of accident vehicles can be effectively extracted. To this end, combine a service platform for a safe warning algorithm.
3.TIVES Algorithm Service Platform
This platform is referred to as Test, Integrate, Verify, Evaluate, Spread platform, based on rich accident vehicle data and algorithm test technology, providing safety warning algorithm testing, verification and evaluation services, enhances the overall level, integration, and promotion of industries. Model algorithm. The function of the platform is 6 directions: provide experimental data management for the industry; data test set supports personalization customization; implementing external information platform docking; providing elastic computing services; providing elastic storage; providing multi-scenario data test set.
TIVES algorithm service platform application related data processing technology, complete test set build, compatible with different scenarios and model tests. From the architectural diagram, it is divided into four phases. First, data collection, first to carry out non-standard data labeled work, access a variety of data sources, then perform data processing, including data analysis, decoding, school Introduction, storage, etc.; Second, data management, first data TTL cleaning, including time format, charge and discharge, encoding format, column name uniform, and data cleaning work, etc., but also data quality Verification, including data check, user-related check, etc .; data collection and data governance are important fundamental works for TIVES platforms. The third is the data warehouse, that is, according to different accident types, battery types, utilities, regional dimensions, mumens, etc., these labels are facing individualized processing of subsequent data; four is data intelligence, according to the test The direction of the collection, such as different testing scenarios, such as safety warning, life prediction, SOH evaluation, etc., providing different test services. The test set is a data foundation for model testing, and is equipped with data intelligence engines, efficient accumulation of data assets, empowering different test scenes, helps enterprises, colleges, research institutions test validation model effectiveness and feasibility. The construction of the entire test data set, the most important two points: data verification, including time sorting, uniform data, easy to provide more accurate input for subsequent testing; data desensitization, including model, region, use, type desensitization, pass Deactivation can better solve the balance between privacy protection and test services, and test data sets itself is also proposed corresponding rules and algorithms.
Core data items, there are many data in the data acquisition process of the entire new energy vehicle. After a large number of verification, there are about 23 degrees of relocation data, this 23 data covers the prediction and related correlation analysis. Aspect, including items, information about the entire battery pack, and different voltage sensors, temperature sensors, mileage, insulation resistance, vehicle speed, positioning, etc. How different manufacturers solve the problem, discovery through existing accident vehicle data: According to the statistics of the accident vehicle data tag characteristics, the standard test concentrated as much coverage of accident vehicle characteristics and data.
Through practice research, in the algorithm evaluation, it is divided into four core indicators. First, the ratio is the proportion of the real hazardous cars in the result of the hazardous cars; the second is to check the rate, indicating the predicted dangerous cars account for the actual accident. The ratio in the car; the third is the runtime, the same test set and the calculation resource conditions, the runtime of different algorithms calculates the runtime; four is the prediction advance time, the same test set, different algorithms are advanced in advance. At the same time, there are also some extension indicators, and these evaluation indicators can provide a reference to the user to test, on the one hand, according to the indicator, the algorithm of the other party can be verified.
TIVES system provides testing, integration, verification, evaluation, and promotion services. The full process of business services is divided into two phases of algorithm access and algorithm testing. From the user experience, it is divided into account login, start testing, selecting data, creating a program, executing program, and viewing a full range of processes. TIVES Algorithm Service Platform Some Functions Interface Reference: Account Login, Sample Data List, Sample Data Details, Console Initial Sandbox, Favorites Data, and Editors.
High performance processing technology of the TIVES algorithm service platform: Heterogeneous data capability (by adaptation of heterogeneous data, unified format data, speeding up the speed, improve cooperation efficiency); data integration capacity (integrated mutually exclusive sample set, satisfaction Scene, multi-working model test); data calculation capabilities (support batch vehicle calculations, complex model calculations, black box calculations); data storage capacity (data storage, professional vehicle data structure). Integrate corporate data, platform data implement unified batches of data storage and analysis, meet data and calculation requirements of different models and application scenarios, deep hardware and hardware collaborative optimization design, forming sustained efficient data acquisition, processing, computing power, and analyzing model testing capabilities.
Information security of the TIVES algorithm service platform, by setting up a separate account, sandbox environment, data isolation, black box test, etc. to ensure the safety of algorithms and data, users enter the data platform through independent accounts, each user has an independent sandbox environment, Preview data information, can not copy download, programming environment supports compiler, and guarantee the security. To ensure relevant test work in reasonable, controllable, safe environments.
4.TIVES Algorithm Service Platform More Applications
In the future, based on the TIVES algorithm service platform, the accident vehicle is conducting a typical operating condition data, third-party safety warning algorithm evaluation, safety warning algorithm tuning service, help the vehicle enterprise to identify external model applications, battery enterprises, model development team model development optimization, Make China to act as an excellent model of industry, expand an early warning service. I hope that based on the data capacity, platform capabilities of China’s steam research, the identity of third parties, can provide more support for new energy vehicle safety warning algorithms, especially in the field of test.