Application of multi-sensor fusion positioning in high-speed railway system

In 2009, the application of C3 train control system in passenger dedicated lines such as Wuhan-Guangzhou and Zhengzhou-Xi'an put forward higher requirements for high-speed railway train positioning technology. The main technical principles of C3 train control system clearly state that the operating speed of the train reaches 350 km/h, the minimum tracking interval is 3 minutes, and the EMUs of 300 km/h and above are not equipped with train operation monitoring devices. On lines of 300 km/h and above, the speed tolerance of the on-board equipment of CTCS-3 level train control system is stipulated as 2 km/h overspeed alarm, 5 km/h overspeed triggers normal braking, and 15 km/h overspeed triggers emergency braking. These technical principles require that the high-speed railway train operation control system must be able to determine the accurate position of the train at any time and any place, including the relevant interval, speed, acceleration and allocation of trackside equipment and on-board equipment resources for the train's driving safety. This information is used to determine whether braking measures need to be taken to ensure safe intervals. At present, sensors such as gyroscopes, accelerometers, odometers, and GPS receivers have been widely used in train speed measurement and positioning systems.

1 Development of high-speed railway technology

The development of high-speed railway technology is diverse, and different countries use different positioning technologies according to their road conditions, terrain, and operational needs. The French AS-TREE system uses Doppler radar for speed measurement and positioning; the North American ARES, PTC, and PTS systems use GPS (Global Positioning System) for positioning; the European ETCS and Japanese CARAT systems use query/responders and speed sensors for positioning; the German LZB system uses track cables for train positioning; and the American AATC system uses wireless ranging for positioning.

According to the complex conditions of China's railway terrain and lines and the requirements of high-speed railways for train positioning technology, this paper proposes a multi-sensor combined positioning solution, selects the GPS/DR/MM combined positioning method, utilizes the information complementarity of multi-sensor combined positioning technology, and uses Kalman filtering to fuse the obtained information to obtain more accurate positioning data than single sensor positioning.

2 Train Positioning System Solution

This solution uses the characteristics of DR's autonomous positioning to ensure that the train can output positioning information anywhere and at any time. The use of GPS can provide DR with initial position data, and the use of MM meets the system's demand for positioning accuracy. The fusion algorithm uses the federated filtering algorithm to solve the shortcomings of other filtering algorithms, such as heavy calculation burden and poor fault tolerance.

2.1 Selection of train positioning method

2.1.1 Introduction to positioning methods

GPS is a radio navigation system. As the earliest high-tech applied to navigation and positioning systems, it has the characteristics of providing users with continuous high-precision three-dimensional position, speed and time information at any time and in any weather conditions around the world. When using GPS positioning, you only need to install a receiver on the locomotive, but the positioning accuracy of the train will be affected in places with many obstacles around. In addition, GPS is very sensitive to satellite failures. Once a satellite fails, GPS performance will deteriorate. Therefore, GPS positioning information cannot be used as the position parameter of the train control system alone.

DR is a relatively classic algorithm used in vehicle positioning and navigation. It consists of a sensor for measuring heading and a sensor for measuring distance. In this solution, an odometer is used as a sensor for measuring distance, and a gyroscope is used as a sensor for measuring heading. The odometer outputs a pulse signal. Every time the wheel rotates one circle, the odometer outputs a fixed number of pulse signals. By accumulating the number of pulses of the odometer within a certain period of time, the distance traveled by the vehicle during this period of time can be calculated, and the speed of the vehicle can also be calculated. The gyroscope outputs the angular velocity information of the heading angle. Integrating the angular velocity information output by the gyroscope can obtain the relative turning angle of the train. Compared with GPS, DR can locate autonomously, and there is no loss of train positioning information caused by problems such as occlusion. However, the initial position of the DR system cannot be obtained autonomously, and the track calculation is an accumulation process. The measurement errors and calculation errors at different times will accumulate. As time goes by, the DR error is a divergent process. [page]

2.1.2 Positioning scheme

Based on the above analysis of the characteristics of GPS and DR positioning, this solution adopts multi-sensor combined positioning technology, that is, a solution in which various positioning technologies complement each other. In the railway line section, when the GPS information is continuous, the GPS receiver installed on the head of the locomotive sends the GPS information to the positioning system. The GPS information is used as the main information, and the DR information and the query transponder information are used as the verification information. The three are combined and filtered to give the optimal positioning estimation information.

When encountering obstacles such as "urban canyons", the GPS signal will disappear or weaken. At this time, DR information is used as the main information. The position before GPS failure can be used as the initial position of DR. After the initial position is obtained, the odometer and gyroscope can be used to estimate the position of the train at the next moment.

After the train enters the station, due to the small distance between the track lines, the positioning accuracy of GPS and DR can no longer well reflect the difference between the tracks, so the query transponder is used to obtain the positioning information of the train in the station. At this time, the query transponder information is used as the main information, and the GPS information and DR information are used as verification information.

2.2 Data fusion method

The core issue of this solution is the design of the positioning algorithm based on data fusion. The data fusion methods used in the field of train speed measurement and positioning include judgment detection theory, estimation theory, data association, etc., and the most widely used is the Kalman filter method in estimation theory. Compared with other estimation algorithms, Kalman filtering has significant advantages: the state space method is used to design filters in the time domain, and the state equation can be used to describe the dynamic characteristics of any complex multidimensional signal, avoiding the trouble of decomposing the signal power spectrum in the frequency domain. The design of the filter is simple and easy, and a recursive algorithm is used. Therefore, Kalman filtering can be applied to the estimation of any stationary or non-stationary random vector process, and the resulting estimation has the best accuracy in linear estimation. The filtering algorithms that have been developed include linear Kalman filtering, extended Kalman filtering, and federal Kalman filtering. This solution uses federal Kalman filtering for data fusion.

2.2.1 Joint Kalman Filter Model for Data Fusion

In this filtering algorithm, βm=0, that is, the main filter has no information input, which further optimizes the system and reduces the amount of calculation.

2.2.2 System filtering algorithm steps

(1) The local filter l processes the train position information output by the GPS receiver and gives the covariance matrix p1 of the state estimate x1 and the estimation error;

(2) Local filter 2 processes the angle information x2 and train running distance information output by the gyroscope and odometer, and gives the covariance matrix p2 of the state estimation and estimation error;

(3) Local filter 3 processes the route length and other information output by the query transponder, and gives the state estimate x3 and the covariance matrix p3 of the estimation error;

(4) x1, x2, x3, and p1, p2, p3 are sent to the main filter and fused with the state estimate of the main filter according to equations (1) and (2) to obtain the global optimal estimate and covariance matrix

(5) Use the optimal estimate of the main filter to reset the state estimates of the three sub-filters.

[page]

2.2.3 Selection of information distribution parameters

Different information allocation coefficients can obtain different structures of the federated filter and different fault tolerance performance, filtering accuracy and computational complexity. In this scheme, an adaptive federated Kalman filter is designed. The P value is automatically adjusted based on the parameters reflecting the positioning accuracy output by the GPS receiver. This positioning system determines the value based on the p value of the GPS receiver. The specific adaptive algorithm is

2.3 Map Matching

The combined navigation of GPS and DR systems has improved the accuracy and reliability of the positioning system to a certain extent, but there are still certain errors in the positioning data, and when GPS data is lost, the error of the DR system will accumulate and become larger. In actual systems, map matching algorithms are usually used to further improve the accuracy of GPS and DR systems.

The basic idea of ​​map matching is to link the vehicle positioning trajectory with the road network information in the digital map, and thus determine the vehicle's position relative to the map. The map matching algorithm is divided into two relatively independent processes: one is road selection, which mainly segments the road, extracts road feature information, and then uses appropriate search rules and matching algorithms to find the most likely road in the map database based on the vehicle information given by the current sensor; the other is road matching, which matches the vehicle's current position and displays it on this road to eliminate the positioning error of the sensor.

3 Conclusion

In order to solve the problem of train positioning in the high-speed railway train operation control system, a GPS/DR/MM combined positioning solution is proposed. The Kalman filter is used to fuse the data information of multiple sensors and then match it with the electronic map to provide real-time train positioning information. Compared with the single sensor positioning method, the speed measurement and positioning accuracy of the train can be further improved, ensuring the safe and reliable operation of high-speed trains.