Adaptive neuro-fuzzy inference system based evolving fault locator for double circuit transmission lines

Received Dec 10, 2019 Revised Feb 12, 2020 Accepted Apr 4, 2020 Evolving faults are starting in one phase of circuit and spreading to other phases after some time. There has not been a suitable method for locating evolving faults in double circuit transmission line until now. In this paper, a novel method for locating different types of evolving faults occurring in double circuit transmission line is proposed by considering adaptive neuro-fuzzy inference system. The fundamental current and voltage magnitudes are specified as inputs to the proposed method. The simulation results using MATLAB verify the effectiveness and correctness of the protection method. Simulation results show the robustness of the method against different fault locations, resistances, time intervals, and all evolving fault types. Moreover, the proposed method yields satisfactory performance against percentage errors and fault location line parameters. The proposed method is easy to implement and cost-effective for new and existing double circuit transmission line installation.


INTRODUCTION
The growth of a country depends on availability of continuous electrical power supply. Most of the research work has been carried out for protection of double circuit three phase transmission lines (DCTPTL) [1][2][3][4][5]. The location of faults by considering fundamental component of currents and voltages in transmission line has been widely studied in Ref [6]. Further, different fault location algorithms have been presented in DCTPTL based on fuzzy inference system (FIS) [7][8], support vector machines [9], artificial neural networks (ANN) [10] and k-nearest neighbor algorithm [11]. Presence of a variety of individual computing methods [7][8][9][10][11] available for locating the DCTPTL is still having certain problems which limit their applicability. Hybridization of technologies can solve the Recently, ANFIS has been found highly successful in applications such as evaluation of drought indices [12] and prediction of solar thermal energy system [13].
In the last few decades, ANFIS networks have been extensively used by researchers for location of transmission line faults [14][15][16][17][18][19][20]. Although, all the ANFIS location schemes [14][15][16][17][18][19][20] are designed to line to line faults (normal shunt faults) occurring at single time but are not applicable for line to line faults (evolving faults) occurring with time gap. Evolving faults include two sequential faults, resulting in the change of fault phase and time in DCTPTL. Some notable works have been reported in the literature on the location of evolving faults in transmission lines [21][22][23][24][25][26][27]. The complexity arising due to the time gap of two faults which adversely affect the performance of transmission line relaying. However, there is still a need to design fault location method to handle evolving faults. Hence, it can be concluded that, hitherto none of the earlier reported papers can locate evolving faults using ANFIS and DCTPTL current and voltage magnitudes.
In this regard, a novel method to locate evolving faults in the DCTPTL using ANFIS is presented in this paper. The novelty and highlights of the proposed work can be summarized as 1. Development of ANFIS based fault locator using single end fundamental current and voltage magnitudes and thus avoids the use of communication channel for sending and receiving terminals 2. Development of ANFIS without using classification method 3. Detects the evolving faults correctly 4. Further, another notable contribution of the proposed method is that it improves the accuracy of fault location and 5. Effectively distinguish the two successive faults different time positions on transmission lines.
The rest of this paper is structured as follows: Section 2 describes the DCTPTL simulation and training pattern generation. Section 3 presents the fault location algorithm for evolving faults using ANFIS. Section 4 reports the simulation results of fault location method. Section 5 provides conclusions. Figure 1 depicted the simplified diagram of the chosen 100 km, 400 kV, 50 Hz, DCTPTL. The MATLAB software is employed for DCTPTL and evolving faults simulation. DCTPTL is constructed using distributed parameter line block. Different evolving faults considered in this paper are 1 phase to ground fault changed to 2 phase to ground, 2 phase to ground faults changed to 3 phase to ground, 2 phase fault changed to 3 phase to ground fault and lastly 1 phase to ground fault changed to 2 phase to ground which again changed to 3 phase to ground.

PROPOSED METHOD
The fuzzy and ANN have their own benefits and drawbacks. ANFIS has capability to capture the merits of both in a single construction. It is a type of adaptive network that is equivalent to FIS functionally. In this, to improve the accuracy of FIS model, neural network is used. The benefits of ANFIS are excellent explanation facilities through IF-Then rules, fast, easy to implement, accurate learning, easy to integrate both linguistic and numeric knowledge and strong generalization abilities. ANFIS can be adopted in a variety of applications of designing, signal processing, control and decision making. It can also be employed for both nonlinear and linear of input-output parameters.
For training of the ANFIS fault locator, various fault cases are simulated with change in evolving fault type, time interval, fault location and fault resistance. Three-phase voltages and currents in circuit-1 and circuit-2 are collected from the source point for all fault cases. The fault parameters used for training of the ANFIS are listed in Table 1. It can be observed from the Table 1 that the number of fault cases for training is 1728. After simulating various fault cases with changing fault parameters, the voltage and current magnitudes measured at the source point are processed.
The ANFIS has constructed in MATLAB software with the DCTPTL simulation data. A single ANFIS has been developed to locate all types of evolving faults in both the circuits under different DCTPTL operating conditions. Nine inputs and single output are taken for developing ANFIS. The fundamental component of three phase current and voltage amplitudes are fed to ANFIS for locating the evolving faults. The fuzzy system inputs symbolize IA1f, IB1f, IC1f, IA2f, IB2f, IC2f, VAf, VBf and VCf. Its output symbolizes location of fault Lf. All input current values are classified as I1 to I5. Similarly, each voltage inputs are classified as V1 to V5 and output is classified as Lf1 to Lf5. The inputs and outputs are classified based on Gaussian membership functions. An array of five IF-Then rules is made for fuzzy system as explained below. Int J Artif Intell

IF-THEN Rules
The fuzzy inference is computed by Takagi Sugeno model. The FIS and training data are given to the ANFIS. The ANN is training by back-propagation method. The most excellent performance is achieved by using back-propagation algorithm with 80 epochs, 5 Gaussian membership functions in each input and output with 5 IF then rules has been used. The structure of the developed ANFIS model is shown in Figure 4. In this way, ANFIS is developed, which provides only one output for location of evolving faults. Now ANFIS is ready to collect the current and voltage and can find the location of evolving faults. When the current and voltage magnitudes are given to the developed method, one of the aforementioned rules is fired and the corresponding location of evolving faults is obtained.  1 phase to ground fault changed to 2 phase to ground fault A1-G to A1B1-G, B1-G to B1C1-G, C1-G toC1A1-G, A2-G to A2B2-G, B2-G to B2C2-G, C2-G to C2A2-G.
1 phase to ground fault changed to 2 phase to ground which again changed to 3 phase to ground fault A1-G to A1B1-G to A1B1C1-G, B1-G to B1C1-G to A1B1C1-G, C1-G to C1A1-G to A1B1C1-G, A2-G to A2B2-G to A2B2C2-G, B2-G to B2C2-G to A2B2C2-G, C2-G to C2A2-G to A2B2C2-G. Fault location (km) [

RESULTS AND DISCUSSION
The performance of the proposed ANFIS based fault location method has been examined and the results have been analyzed. In order to examine the response of proposed approach, it is necessary to get the relevant data which is never used during training process. Different parameters have been selected for checking the proposed ANFIS. Total 14000 test samples of evolving faults, with change in fault parameters like faults locations, fault resistances, fault types and time intervals have been collected. After the training process, the achieved optimal network configurations are checked with the MATLAB software tool samples. The error (%) in estimated fault location is calculated using (1). Some results of the evolving faults and proposed method are explained below. (1)

Performance changing evolving fault types
In addition to ten types of shunts faults, DCTPTL is also prone to evolving faults due to animal/human and tree branch falling on live conductor normally begins off as single phase to ground faults. Most of the methods have been developed and unable to located especially evolving faults. Thus, the developed method is verified for different evolving faults in both the circuits of transmission lines. Performance changing evolving fault types, keeping invariable time interval (20 ms), constant resistance (80 Ω), and fixed location of fault (65 km) are listed in Table 2. The error is up to 0.281% and detection time is up to 0.06 ms for all types of evolving fault. As seen in Table 2, evolving faults have no influence on the performance of the proposed method even for different evolving faults occurring in the line.

Performance changing time intervals
The faults can take place with evolving fault type at any time; therefore, to study its effect on the developed method, a variety of time intervals for evolving faults have been applied. In order to explain the impact of time interval, the fault in the fixed location value (43km) with constant resistance value (50Ω) for the evolving fault type (A2-G to A2B2-G) and various time intervals has been considered. Some of results of fault location changing time intervals during evolving faults are illustrated in Table 3. The error is up to 0.290% and detection time is up to 0.07 ms for all cases. The maximum error of the fault location in DCTPTL line remains acceptable regardless of the variation in the time intervals.

Performance changing fault resistance
The fault resistance has a significant effect on the study of evolving fault analysis. Various fault resistances have been used in the simulation study and the results with typical fault resistances are reported here. To check its performance the evolving fault (A1B1-G to A1BA1C1-G), kept constant location (18km) and time interval (15ms), and resistances are varied from 10Ω to 120Ω. The test results of the developed method for evolving faults in DCTPTL under changing resistances are illustrated in Table 4. The error is up to 0.230% and detection time is up to 0.06 ms for all types of fault. Results give a satisfactory error of the proposed method for different fault resistances. What's more, the method stays unaffected by fault resistance. Even if the fault occurs on the vicinity of resistances, the method can locate evolving faults.

Performance changing fault location
In this part, the simulation results with different fault locations are presented. The developed method is tested by keeping fixed time interval (10ms), constant resistances (20Ω), different locations of evolving fault (B1-G to B1C1-G) from the relaying point ranging from Lf = 1 km to 99 km of the line. Some of the results of evolving faults in DCTPTL under changing locations are given in Table 5. The maximum absolute error percentage was 0.292% under evolving faults and maximum detection time taken by the proposed method is 0.06 ms. As can be seen in Tables 5 the developed method is able to achieve a reasonably high accuracy in locating fault. Also, the performance of the developed method is not affected by variations in the fault location.

CONCLUSION
This paper has outlined a new ANFIS method for locating evolving faults as well as shunt faults in the DCTPTL. System designing and simulations have performed in MATLAB software package for capturing samples of faulty signals by varying fault parameters. Low pass filter, sampler, and DFT have utilized to derive the features of raw current and voltage signals, which are further given as inputs for the fault locator. The proposed technique has overcome problems such as information from two ends of the transmission line and knowing the fault type. The simulation results showed that the method can be applied within the entire line and the accuracy of the fault location is high. Furthermore, the proposed method correctly located the evolving faults and tested with all fault events. Thus it can be applied for protection of real power systems as well.