Over Voltage Prediction System for Long Cable Issue Using Artificial Intelligence

Introduction

Electricity is a backbone of our existence in today’s world. It’s quite hard to imagine a world without electricity in the modern times. We are dependent on it for a major part of our day to day activities. However, electricity does not exist in the natural form and also cannot be stored in usefully large quantities. It has to be harnessed from renewable or non-renewable resources and must be generated continuously to meet the demand from consumers. The electricity that is generated from its various sources are delivered to end users/consumers through the Transmission and Distribution(T&D) lines.

Transmission lines comprises of wires/cables that are hung on tall metal towers and carry electricity over long distances at high voltages. The electricity generated at power plants moves through this complex system, often referred to as the ‘grid’, which has powerlines, transformers and electricity substations that ultimately connect electricity producers with the consumers. A recent development in the transmission and distribution grids is the smart grid. A smart grid enhances the traditional transmission and distribution system through usage of digital technology and advanced instrumentation that enables utilities and customers to receive, up to date information from the grid and also enables communication with the grid. The grid is termed smarter, as it makes the electrical T&D system more reliable and also efficient in reducing transmission losses and enables the utilities to detect and fix problems quickly. The smart grid also enables end users to intelligently manage usage of electricity, especially during times of high demands or when system reliability requires lower energy demand. High voltage electricity is used in long distance electricity transmission, as they are associated with low currents and reduces the transmission losses, and hence more efficient and less expensive compared to transmission at low voltage state.

However, electricity at lower voltage is required at homes and businesses, viz. the end users. Transformers at various stages in the grid and those in substations, perform the job of increasing (stepping up) or reducing (stepping down) electricity voltages to adjust to the different stages of the movement of electricity, from the source of power plants through long-distance transmission lines and lower voltage distribution lines that carry electricity to homes and businesses. Transformer are set of 2 coils (3 coils for 3 phase system), tightly coupled by means of magnetic core. Developments in power electronics technology has improved the switching frequency of devices. Insulated Gate Bipolar Transistor (IGBT) technology currently is capable of switching frequencies in the range of 10 to 20 khz, along with rise times of 0. 1 µs[1].

The performance of PWM (Pulse-Width Modulation) voltage inverter as well as its output waveform have drastically improved because of this development thereby making them the popular choice for drives in induction motors with variable speeds. The induction motor and PWM inverter are mostly located in different location, especially in industrial applications and hence require long cables or motor leads. Fig. 1 shows a practical connection of induction motor fed via PWM inverter using long cable.

The electromagnetic pulses travel at half the speed of light (3*108 m/s), which is approximately(150-200m/µs). If the electromagnetic pules take longer than half the rise time to travel from inverter output to motor, then a full reflection of the pulse will occur at the motor terminal. This will cause the doubling of the amplitude of the voltage pulse. Increasing the rise time and fall time of inverter voltage output pulses will help to reduce this reflection of the pulse caused by fast switching transients. The factors affecting voltage reflection are the

• Rise time of the inverter output voltage pulses, tr
• Cable length used for connecting the inverter and motor, which affects the propagation time, tp of the pulse
• Surge impedance of cable and motorIf tp>tr/2, then a full reflection of the pulse occurs and cause doubling of the amplitude of voltage at the motor terminal[1][2].

To reduce the dv/dt gradient of the inverter output voltage, passive dv/dt filter with inductor, capacitor and resistor could be used. The major drawback of such filters is the large power loss. The reflected voltage amplitude depends on voltage reflection co-efficient(ℾm) of motor and is given by equation below: ℾm = Zm-Zc/Zm+Zc where Zm is characteristic impedance of the motor and Zc is the characteristic impedance of the cable. The peak voltage at the motor terminal, (Vm) is given by the following equation, Vm=Vs(1+ ℾm)where Vs represents source voltage. The reflection co-efficient varies with the size of the motor and its value reduces as the size of motor increases.

To reduce the losses in induction motor, fast switching devices are used in the inverter, but this in turn may cause overvoltage to appear at the motor terminals. The shorter rise time or steep slope of the voltage would cause the motor insulation to break down. J Desmat and et al [6] investigated and analyzed the impact of following parameters on overvoltage at the motor terminal

Inverter parameters

The influence of carrier frequency, voltage boost and U/f characteristics were analyzed. No difference in overvoltage or slope were detected while using carrier frequency of 2, 5, 8, 12 or 16 kHz. But increase in frequency leads to increase in number of over voltages per unit time. Changing motor frequencies to 10,30,50,60 and 70 Hz also did not cause overvoltage. However, an increase in number of over voltages per unit time with increase in motor frequency. Also starting voltage did not have an impact on the overvoltage. U/f characteristics had same effect on overvoltage as carrier frequency and motor frequency.

Cable parameters

The influence of cable length, cable type and cable section were analyzed. Motor voltage slope and rise when cable length increases. Difference in composition of the cable affects the damping. Increasing cable cross section increases voltage slope at motor terminals.

Motor parameters

Motor power has slight impact on motor reflection coefficient. For motor power less than 22kW this effect is not seen. It was observed that some of the parameters like rise time of the voltage at the inverter, the length, cross-section and type of the cable, power of the motor and carrier wave frequency of the inverter have a significant impact on the occurrence of overvoltage at the motor end. The project aims to implement a peak overvoltage prediction system using AI techniques. The test system consists motor driven by PWM inverter circuit via long leads or cables. Knowing the peak overvoltage is of significance for the insulation co-ordination of the long cables. Supply parameters and cable parameters are intended to be fed to the proposed system to provide an estimation of the overvoltage value.

Literature survey

Many researchers have worked on the topic of overvoltage in transmission lines. Some research work focuses on the overvoltage in power lines and others on overvoltage on motor connected via long leads on inverter fed systems. The researchers have identified the factors causing overvoltage. Artificial Neural networks (ANN) based solutions are proposed by few researchers to predict overvoltage on power lines [1][2]. Support Vector Machines (SVM) based classification of supply voltage to induction motor was also explored by [7] researchers in the past. A. V. Jouanne and et al. [1] examines how long motor leads affect AC motor drives that are fed by high frequency PWM inverters. It is observed that, though high switching speed leads to an improvement in the performance parameters of the PWM inverters, it has negative effect on motor insulation.

Also, long cable length tends to cause excess voltage at the motor terminal and this puts further stress on the motor insulation. The voltage reflection is analyzed/investigated and the theory of cable transmission as well as the analysis of cable capacitance are presented. The paper also illustrates as to how the length of the cable and the rise time of the inverter pulse output affects the magnitude voltage at the motor terminals. A. Acharya B and et al. [2] designed specifically for motor drives an output dv/dt filter. The paper has proposed a new procedure to design an LC clamp filter. The output voltage of PWM inverters has high dv/dt, which can cause a doubling of peak voltage at the motor terminal connected using long leads. By using the proposed filter, the phenomenon of voltage doubling at the motor terminal can be reduced. There are several mitigation techniques available that are for implementation at the motor end. They are, use of a bearing that is insulated, use of an electrostatic shield between the rotor and the stator, increase of the grade of insulation as well as a termination that matches the impedance of cable and motor terminal. At the inverter end mitigation techniques that can be adopted are usage of filter, reduction of common mode voltage and resonant switching inverter. The proposed filter addresses components that are of common mode or differential mode.

In addition, in the filter, the resonant frequency is selected leading the induction motor to behave as inductive load at high frequency which in turn has the LCL effect of a higher order filter. Effectiveness of this is verified using Pspice simulation. The filter is designed to be compact and can be easily included/placed in the inverter package. Vitor F. Couto and et al. [3] proposes an alternate strategy to model a transmission line which analyzes the three phase conductors separately. Usually, power transmission lines are modelled as a single quadrupole circuit in an arrangement of parameters which is similar to the shape of the Greek letter, Pi. A transmission line of 230 kV and 197 km is adopted to illustrate the importance of the proposed method. The results from the simulation clearly show that there are overvoltage occurrences along the line although the values of voltage at both ends of the line are at acceptable levels. Need for new procedures for insulation co-ordination of power transmission lines is highlighted by the researchers. D. Thukaram et al. [4] presents an approach that is based on ANN in which peak over voltage at switching transients during line energization is estimated. In this approach, to train the multilayer perceptron the Levenberg–Marquardt method is used.

The insulation level of the Extra High Voltage system (EHV) ac system is determined by the switching overvoltage. The computation of temporary overvoltage as well as switching overvoltage is done using the Electro Magnetic Transients Program (EMTP) tool. The aspects considered to influence the transient overvoltage are the length of the transmission line, the switching angle, the strength of the source and the reactor at the receiving end. The proposed tool would be helpful to the operator to predict the overvoltage. During the line energization, the wave that travels starts along the line towards the destination which is the receiving end and at the open-end doubles to an overvoltage near to 2 p. u. The proposed methodology is explained considering a 400kV EHV network. Since overvoltage can be caused due to switching during energization, a controlled switching is adopted on the transmission line in order to keep this effect to the minimum. Increase in length of the transmission line, increases the charging current and creates overvoltage at the receiving end bus. In the proposed methodology, two tracks are adopted in order to get the estimate of overvoltage. In the first track, the value of the reactor at the receiving end is kept as fixed. The parameters that have an influence on peak voltage under this track were the switching angle, strength of the source and length of the transmission line.

In the second track, the reactor at the receiving end is kept as switchable and has various values. The parameters that have an influence on the peak voltage under this track are value of the reactor at the receiving end, the switching angle, the strength of the source and the length of the transmission line. The proposed methodology for both the tracks uses a Multi-Layer Perceptron (MLP) which has one hidden layer along with ten hidden units. Supervised learning of ANN is adopted for the proposed methodology. The input is then supplied to EMTP to return the peak values of transient overvoltage. The data values thus obtained are used to train the ANN. Seyed Abbas Taher et al. [5] uses ANN for nonlinear input-output mapping, to predict temporary overvoltage due to transmission line energization. Multi-Layer Perceptron (MLP) is trained using second order of Levenberg–Marquardt method for obtaining small mean square error(MSE). In extra high voltage lines(EHV) lines, for insulation co-ordination, the primary parameter is the switching overvoltage.

The MATLAB/Simulink tool includes Power system blockset(PSB). This PSB is utilized for calculation of overvoltage in the proposed approach. The proposed methodology is explained using a sample system consisting of a 400 kV EHV network. This methodology considers the following parameters of the network to predict the transient overvoltage.

• Voltage at source
• Resistance of the equivalent network
• Inductance of the equivalent network
• Capacitance of the equivalent network
• Circuit breaker poles closing times
• Length of the cable
• Capacitance of the line
• Capacity of Shunt reactor

The scale and duration of the overvoltage is not dependent on any single isolated parameter as the changes in values of one parameter has the ability to alter the impact of another parameter. This prevents the derivation of simple and straight forward formulae that is applicable to all cases. So, in proposed methodology an ANN is used to predict the magnitude and duration of the peak overvoltage during line energization. A feed forward MLP which has hidden units of 10 numbers and 2 hidden layers is used in the proposed methodology. Energization of both single and three phase transmission line is analyzed as part of this methodology.