Vehicle time series data analysis

Cloud Computing represented by hadoop only provides an algorithm runtime environment, and provides the best (approximate) method for Parallel Computing of big data at the current software and hardware level, it cannot solve all problems in big data applications. For specific applications, data providers that access the it circle through the Internet of Things are facing the primary problem of data analysis algorithms, followed by parallel computing of algorithms.

Taking car manufacturers (OEM, tire1, vendor, TSP) as an example, the big data problems faced are in 4 V (volume, velocity, variety, veracity/value, the most prominent difference is velocity, that is, real time. Some signals have an update cycle of 10 ms. Of course, from the perspective of application sampling and algorithm processing, there may be no need for such confidential data, which involves the differences between the system architecture and which functions are run on the final end, which functions are run on the backend server. Taking the engine speed signal as an example, the signal cycle on the bus is generally 10 ms ± 5%. If the entire IOV system only needs Driver Behavior Analysis (reflecting the vehicle running status ), it is impossible to use such a high-frequency sampling period, and it is possible to package and send data to the background once every 10 seconds. However, if the application of IOV is engine fault diagnosis or anti-theft alarm, the accuracy is different. If the normal start speed is lower than 500rpm, the engine is almost certainly abnormal, if the driver is notified after 30 s, the engine should smoke. For some event-triggered signals, such as the abnormal startup of the engine when the car is locked, the background server has a higher requirement for determining the time when the car is stolen.

When the IT industry evaluates whether the system uses nosql or SQL, how to search for data processing on automobiles is the first challenge. Unlike text data in the traditional internet industry, the Internet of Things (IOT) or vehicle network is faced with time series data. In this regard, you can see the various dazzling curves on the stock market trend chart. When a signal sample is defined as a fault mode, there are still similar curves in the historical data, which is a problem of similarity search in time series. If a signal curve is always cyclical and shows a certain upward or downward trend, whether the signal can be predicted in the future is a data prediction problem. Other data-related analysis, data clustering, and other techniques collectively referred to as data mining are built on structured data to reduce the data dimension (variety ), currently, there are limited applications in the automotive control and analysis fields.

Unfortunately, there are almost no mature tools and methods for analyzing and processing time series in the IOV field, even if professional mathematical tools such as MATLAB, R, and wolframalpha provide few algorithm libraries. This is because the accumulated data in the IOT industry is not rich enough and the application prospects are unclear. More importantly, the processing of time series data involves technologies and methods in various specialized fields, making it very difficult to process. Taking the car speed data as an example, in the era of Mechanical Industry, everyone is concerned about the unit of hour, in the era of electronic and information industry, the unit is s, and in the whole process of the IOT industry and big sample scenarios, the discussion is ms.

Vehicle time series data analysis