Design and Simulation of an Intelligent Current Monitoring System for Urban Rail Transit

Many urban rail transit (URT) systems adopt the DC traction power supply system. Because of the impedance and incomplete ground insulation of the running track, it is inevitable for a part of the traction current to flow into the ground from the track, creating the stray current. This type of current causes great safety hazards to the metal structures in and near the URT system. Considering the power supply mode of the URT, this paper explores the different resistances in each power supply section under unilateral power supply and bilateral power supply. Then, the defects of the current discharge method were identified in the context of stray current protection. To solve these defects, the backpropagation neural network (BPNN) was adopted to build a discharge flow prediction model. On this basis, an intelligent current monitoring system was established for the URT. Finally, the authors simulated the impact of each factor on stray current, and verified the reliability and stability of the proposed monitoring system. Compared with predicted values and the actual values, the prediction agrees with the actual data very well.


I. INTRODUCTION
With the boom in urban rail transit (URT), the stability and safety of the URT system have become a major concern. In the URT, the train cannot operate safely and efficiently without a robust current system.
The rated voltage of traction power supply in the URT falls on multiple levels in the traditional range of DC 500-3,000V. During train operations, the traction current flows to the train via the catenary, and returns through the running track to the negative pole of the rectifier at the traction substation. Because of the impedance and incomplete ground insulation of the running track, it is inevitable for a part of the traction current to flow into the ground from the running track, and then back to the track and the substation. This part of current is called the stray current [1].
The stray current poses many threats to the URT, such as eroding the metal structures in and near the system. Calculations show that a stray current of 1A can erode 9.1kg metal within a year. The normal operation of the URT generates a large volume of stray current. It only takes a few years before The associate editor coordinating the review of this manuscript and approving it for publication was Dalin Zhang. the buried metal structures falling into disrepair, owing to the corrosion by the stray current.
The running track of the URT is often interlaced with underground municipal pipelines, ranging from gas pipeline, water supply pipeline to communication pipeline, which makes matters worse. To protect the URT operations and other underground facilities, it is imperative to identify the distribution law of stray current in the URT, and formulate effective measures to monitor the stray current.
In the light of the above, this paper attempts to design an intelligent monitoring system for the stray current in the URT. Firstly, the different resistances in each power supply section were examined under two modes of traction power supply. Next, the authors identified the defects of the current discharge method for stray current protection, and then built a discharge flow prediction model based on the backpropagation neural network (BPNN). On this basis, an intelligent current monitoring system was established for the URT, and verified through example analysis.

II. LITERATURE REVIEW
Many scholars have attempted to control the current in the URT. For instance, Shen and Jiang [2] [3] differentiated between the front and rear tractions of the electric drive system, through analysis on rectification, inversion, modulation and demodulation.
Li et al. [4] divided the current monitoring system of the URT into several subsystems, and determined the basic elements of that system. The stray current is the focus of the current monitoring system for the URT. Chen et al. [5] treated the distribution law of stray current as a complex current field problem, and held that the stray current distribution is affected by the design size of the URT system and the location of the train. Sheridan et al. [6] highlighted the complexity of the traction power supply system in the URT, as multiple traction stations need to supply power to multiple trains at the same time. Under the unilateral power supply mode, Jia et al. [7] built an analytical model of stray current distribution in the track-earth return system, using the following hypotheses: the catenary does not produce stray current onto the earth, and the positive pole of the traction substation at the receiving end of the train is not coupled with the earth. To reduce the total traction power and track potential, Chen et al. adjusted the time of each substation by the genetic algorithm (GA) to prevent the simultaneous starting and braking of multiple trains [8]- [10].
Currently, the URT parameters are measured either online or offline [11]- [13]. Wang et al. [14] proposed the field test method after analyzing the impacts of track-earth transition resistance, track potential and longitudinal track resistance on the return system. Liu et al. [15] put forward an approach to test the track-earth transition resistance under AC power supply, but the test approach faces large errors. Drawing on Newton's iteration, Daude et al. [16] developed an online measurement method for track-earth transition resistance. Using first-order differential equation, Haidar et al. [17] solved the continuous model of the return system with DC traction power supply, and discussed about the ways to test the track-earth transition resistance and track resistance.
In the URT, the stray current is mainly generated as the current returns to the DC traction power supply via the track. The monitoring and prevention of stray current require the synergy between various disciplines. Some scholars have investigated the harms of stray current and developed various countermeasures that integrate prevention, discharge and interception. For example, Zhou and Xing [18] explored the stray current monitoring system based on Foundation Fieldbus (FF), and detailed the architecture, software and hardware of the system. Di [19] compared the merits and defects of centralized and decentralized strategies of current monitoring, developed a distributed current monitoring strategy, and verified its effectiveness through field test. Johnston [20] proposed the array electrode method to measure the potential distribution at the interface of concrete reinforcement and at different depths. Cai et al. [21] probed into the monitoring methods of stray current on corroded long-distance pipelines.
Zhao et al. [22] tackled the IR-drop error through polarization potential test, theoretically derived the double reference electrode method to eliminate the IR-drop, and experimentally proved that the method can eliminate the IR-drop, which exists in the stray current monitoring of the URT.

III. STRAY CURRENT MODEL OF THE URT
The URT has a very complex DC traction power supply system, which mainly consists of a regional substation (or gridconnected step-down substation), an external power supply system, a special electrical system (including a DC traction substation and a step-down substation), and a track area. The architecture of the DC traction power supply system is shown in Figure 1 below. Before setting up a current monitoring system of the URT, it is necessary to clarify the operation mode of the traction power supply system. In general, there are two types of operation modes for the said system, namely, unilateral power supply and bilateral power supply. For a train running in a power supply section, it is powered only by a nearby traction substation under the former operation mode, and powered by two front traction substations nearby under the latter operation mode.
In the DC traction power supply system of urban rail transit, the traction substation is fed to the catenary, and the current flows back to the negative pole of the traction substation through the running rail after passing through the electric locomotive. In the process of rail return, part of the current flows into the underground through the track bed, and two corrosion cells will be formed underground. The metal in these two areas will be corroded by electrolysis. In the process of corrosion, the chemical reaction process is also different according to the surrounding acid and alkali environment. Hydrogen evolution corrosion occurs when metals are corroded in acid environment, and oxygen absorption corrosion occurs when metals are corroded in alkaline environment.
On this basis, the following constrains were put forward for the stray current distribution under the mode of unilateral power supply: (1) The track has a uniform longitudinal resistance; (2) The track-earth transition resistance is uniform; (3) The earth has a uniform longitudinal resistance, which is infinitely close to zero; (4) The system impedance of traction substation(s), feeder, and catenary is negligible.
Under the above constraints, the track-earth resistance under the mode of unilateral power supply was modelled as Figure 2, where R g is the track-earth resistance, R is the longitudinal resistance of the track, and I is the traction current of the train. Note: R g represents track-earth transition resistance, R represents rail longitudinal resistance, I represents traction current of electric locomotive.
The current flow path of track-earth resistance is shown in Figure 3, where x is the distance to the traction substation, and i(x) is the track current at x.
where i(x) represents current, R represents the longitudinal resistance of the rail, u(x) represents rail potential at position x. Equation (1) can be simplified as: Similarly, the following equation can be derived from Figure 3 R g represents rail-ground transition resistance. Solving (3), we have: According to (2), we have: Substituting (4) into (5), we have: Assuming that α 2 = R R g , (6) can be transformed as: The general solution of (7) can be expressed as: where, ϕ and ω are indefinite coefficients. Taking the derivatives of (8) on both sides, we have: Substituting (9) into (2), we have: According to (10), the indefinite coefficients ϕ and ω can be solved as: In the URT, the stray current distribution is affected by many environmental factors. To disclose the basic law of stray current distribution, the stray current model was applied for the URT simulation. According to the value ranges in Table 1, VOLUME 8, 2020 the model parameters were configured as follows before the simulation: R s = 0.02, R d = 20, R l = 0.015, R t = 5, R b = 0.015, D = 3.5 and I = 3, 000.
First, the URT was simulated under the two power supply modes, and the distributions of track potential and stray current were observed. The distance between the train and the traction substation was 1km and 2km under the unilateral power supply mode and the bilateral power supply mode, respectively. According to the simulation results, the effects of power supply modes on track potential and stray current are presented in Figures 4 and 5, respectively.  The simulation results show that the distributions of track potential and stray current between the train and the traction substation were similar between the two power supply modes. The track potential decreased monotonically from the positive maximum at the train towards the negative maximum at the traction substation, passing through a point of zero track potential. The stray current distribution takes the shape of a parabolic curve between the train and the substation.
The tack potential and stray current were higher under the bilateral power supply mode than under the unilateral power supply mode. This is because the train is powered by the traction substations on both sides under the former mode and by only one substation under the latter mode. Since the distributions of track potential and stray current follow the same trend under the two modes, the unilateral power supply mode was adopted to analyze the influence of other factors in subsequent simulations.

IV. BPNN-BASED URT CURRENT MONITORING
In the URT, several parameters are the key targets of stray current monitoring, including longitudinal track resistance, track-earth transition resistance and track potential. The weights of these parameters can be determined by their sensitivity to stray current corrosion.
In the URT current monitoring system, there is no exact relationship between the discharge flow and polarization potential. The stray current flowing from track to earth can be captured by the current discharge protection method.
To enhance the discharge protection effect, it is necessary to build an artificial neural network (ANN)-based model that predicts the discharge flow, according to the polarization potentials of multiple monitoring points.
Here, the discharge flow is modelled based on the BPNN [23], which is good at solving nonlinear problems with no exact relationship between the input and output. The BPNNs generally have three or more layers, each of which has multiple neurons.
In the URT, the current discharge mainly occurs under the polarization potential of the metal structure. For each power supply section, a total of 16 reference electrodes were deployed at different positions to capture the polarization potential of every metal structure.
According to the multi-point evaluation principle of discharge protection, the polarization potential should be evaluated at five points at the least. In this paper, the number of evaluation points is set to 6, and these points have different impacts on discharge flow. The mean value of the forward propagation of the 6 monitoring points was taken as the input of the BPNN, and the proportion factor µ of the discharge flow was taken as the output. Thus, the structure of the BPNN-based discharge prediction model is presented in Figure 6.
where, c is the number of input layer neurons; d is the number of output layer neurons; t is a constant in [1], [10]. Through calculation, the number of neurons on the input layer, hidden layer and output layer were set to 6, 10 and 1, respectively.
The BPNN can memorize the patterns of the problem through data training. Hence, the training data must cover all the characteristic patterns of the problem. In our model, the polarization potentials measured at the six monitoring points are taken as the inputs of the sample data, and the discharge factor was taken as the output.
The sample data were imported to the BPNN-based model, with the error of 0.00001 as the training target. After 29 iterations, the error of the BPNN was 0.00008961, which achieves the training target.
To verify its performance, the BPNN-based model was tested by the sample data in Table 2. The predicted values and the actual values are compared in Figure 7. The comparison shows that the prediction agrees well with the actual data.  As shown in Figure 7 and Table 2, the predicted values of the BPNN-based model only deviated from the actual values of the ten groups of sample data by 0.05, indicating that our model can effectively predict the discharge flow after training.

V. CONCLUSION
This paper introduces the architecture of the transit power supply system in the URT, analyzes the formation of stray current in URT operations, and simulates the degrees of impacts from various factors on the stray current. On this basis, the authors established a current monitoring system for the URT. First, the measurement method for monitoring parameters was discussed in details. Next, the defects of current discharge method were analyzed, in the context of stray current protection. To solve the defects, a stray current monitoring system was developed based on the BPNN. Finally, the effectiveness of the proposed system was proved through data verification. Therefore, the analysis of the distribution law of the current and the adoption of effective current monitoring methods and protective measures will have positive significance for the design, construction and safe and reliable operation of urban rail transit in the future.