Application of automation technology in power system protection and control

Through continuous development, the power system has also been continuously upgraded, and intelligent protection technology, as an advanced technology, can realize real-time monitoring and control of the power system. The application of this technology in the power system can play a role in protecting the current power system. In the use of this technology, if there is a fault in the circuit, the protection device can cut off the fault part, reduce the scope of the fault, without affecting the normal part of the power system, so that other parts can still work normally. In the process of circuit operation, the application of the technology can also ensure the stability of the circuit, reduce the incidence of system failure. If the chain system malfunction, intelligent protection can make automated decisions, notify the technicians, and promote the rapid elimination of faults. The application of this technology can not only enhance the stability of the power system, but also improve the maintenance level of the system. The application of intelligent protection in the power system can continuously improve the automation level of the system and promote the long-term development of power engineering.

PLC technology is also a commonly used automation technology in power systems, PLC technology, also known as Programmable Logic Controller, is an industrial controller that uses program control [21]. It has functions such as input, output, logic programming and storage to achieve task-specific control functions. It is used to control various devices and mechanical equipment in the power system to realize automatic control. The technology integrates relay control and computer technology, and belongs to the microcomputer-based relay protection technology. Compared with the previous technology, the logic and reliability of this technology is more powerful, and has good anti-interference ability and adaptability, more efficient and convenient to use. The technology can recognize and calculate the main line signal, carry out automatic programming, and then generate instructions to control the operation of the system. The application of this technology in the power system can enhance the sensitivity of the system and reduce the loss of system operation.

Human production and life are inseparable from the application of electric energy. With the improvement of people’s living standards, the demand for electricity is also increasing. In order to improve the efficiency of the power system, automation technology should be optimized to improve the performance of the power system, so that people can get a better experience of using electricity. In the power grid, power system protection occupies a very important position. Therefore, automation technology should be reasonably applied to power system protection. When a fault occurs during the operation of the power system, the automatic start of equipment protection can ensure that the power system runs more smoothly. Automation technology can also provide technical support in equipment monitoring, so that the relevant technical personnel can quickly eliminate the obstacles. This can not only reduce the system failure rate, but also improve the quality of work of operators.

When PLC technology is applied to the power system, it can automatically control the transmission process of the power system and improve the transmission quality and efficiency of the power system. The development of the power system makes the system’s data acquisition ability, analysis ability and processing ability are upgraded synchronously. PLC technology has good compatibility, self-healing, security, interactivity and economy, the application of automation technology in grid scheduling can fully grasp the operation of the power system, according to the operation specification to control the operation state of the equipment, so as to make the transmission process of the power system smoother. PLC technology is also able to realize the diversified use of energy, such as wind energy, water energy, solar energy, etc., which effectively improves the utilization of resources, reduces the losses and costs of the power grid, and provides technical support for the development of the power system.

Common activation functions include ReLU, Sigmoid, Tanh, etc. Different activation functions are selected according to the needs of the actual task, and choosing the appropriate activation function is an important decision in the design of convolutional neural networks [23]. 1)

Sigmoid function

Sigmoid function, a common S-shaped function in biology, is also known as S-shaped growth curve. Its mathematical expression is shown below: (1)S(x)=11+e−x

Loss function is often called cost function or objective function, and its core concept is to mathematically measure the degree of deviation between the predicted value and the true value of the model [24]. Common loss functions include 0~1 loss function, absolute value loss function, and mean square loss function, which are described in detail in this subsection:

In routine cases, when the accuracy of the prediction results needs to be assessed, it is usually not necessary to consider the specific degree of error, but only need to determine whether the prediction results of the model coincide with the actual true value. If the prediction matches the true value exactly, then the loss value can be considered to be 0. Conversely, if there is any mismatch, then the loss value should be set to 1. The formula is shown below: (4)L(y,f(x,θ))={ 0,if(y=f(x,θ)) 1,if(y≠f(x,θ))

The absolute value loss function is generally used in regression problems to calculate the mean of the absolute difference between the predicted value and the true value, and its formula is shown below: (5)L(y,f(x,θ))=|y−f(x,θ)|

The mean-square loss function, which is the same as the absolute value loss function and is also more often used in regression problems, calculates the mean of the squared difference between the predicted and true values with the following formula: (6)L(y,f(x,θ))=12(y−f(x,θ))2

The cross-entropy loss function is one of the most common loss functions when dealing with classification problems. It measures the difference between the probability distribution predicted by the model and the probability distribution of the actual label. The cross-entropy loss function can effectively promote model learning and make its prediction results closer to the distribution of real labels. Mathematically, the cross-entropy loss function can be expressed as: (7)LCE=−∑i=1nyilog(y^i)

where y_i is the true probability distribution and y^i is the model predicted probability distribution.

The structure of the convolutional neural network is shown in Fig. 1. The convolutional neural network is a typical deep learning model, which learns each local feature in the input through the convolutional kernel and extracts the deep abstract feature information layer by layer, and then fuses and analyzes each local information through the fully connected layer. When a two-phase ground fault occurs on line L2, each sample consists of 16 channels such as du_al and di_al, and each channel is set to have a duration of half a cycle, i.e., 100 points at 10 kHz. The waveform of a channel can be regarded as a local characteristic of the whole waveform, while a set of small fragments of equally spaced samples can reflect the morphological characteristics of that channel. For example, a du_al-channel characteristic can be reflected by 5 samplelets fragments, each containing 5 sample points, and this can correspond to a convolutional kernel. The relevant parameters of the CNN network structure chosen in this paper are as follows. 1)

Convolutional layer: the size of the corresponding convolutional kernel for C1 is 5×5 and for C2 is 3×3. The number of convolutional kernels for both C1 and C2 is 20, and the step size is 1. The ReLu function is used as the activation function.

Figure 1.

Convolutional neural network structure

In this paper, the three transmission line two-end power model shown in Fig. 2 is used as a case study and simulation experiments are performed in PSCAD/EMTDC. The system frequency is 50 Hz, the rated voltage is 220 kV, and the capacity of power supply at both ends is 100 MVA. The length of lines L1 and L2 is 200 km, and the length of L3 is 350 km, and the three-phase transmission lines are not phase switched. The positive-sequence impedance of transmission lines L1 and L2 is Z112=5.38+j84.55Ω and the zero-sequence impedance is Z01.2=64.82+j209.53Ω , and the positive-sequence impedance of transmission line L3 is Z13=9.415+j147.96Ω and the zero-sequence impedance is Z03=113.435+j366.68Ω . The power supply impedance on both sides is ZS1.2=9.19+j74.76Ω . In Fig. 2, the protection elements installed at both ends of the three transmission lines are labeled Relay 1 to Relay 6 in sequence.

Figure 2.

The case study representing sources at both line terminals and three lines

Line 2 is analyzed using data collected from Relay3 and Relay4. For the protection of Line 2, faults occurring on Line 1 and Line 3 are external faults, and faults occurring on Line 2 are internal faults, and internal and external faults on the line are each categorized into 10 specific fault types including single-phase grounded short-circuits, two-phase short-circuits, two-phase grounded short-circuits, and three-phase grounded short-circuits (three-phase short-circuits and three-phase grounds are considered to be the same type of fault). Thus, including no faults, Line 2 can be categorized into a total of 21 status categories.

In this chapter, the proposed multi-channel feature fusion-based power system transmission line fault identification and protection scheme is implemented by convolutional neural network with specific training and testing strategies as described below:

After online testing, the flow of the proposed corresponding power system protection scheme is shown in Fig. 3, considering the online performance characteristics of the recognition algorithm. The three subsets are named as dataset 1-training, dataset 1-validation, and dataset 1-testing, respectively. The network with the highest accuracy in dataset 1-validation is selected as chosen as the final CNN, and dataset 1-test is used to test the performance of the CNN. In order to validate the generalization ability and online performance of the CNN after evaluating the training completion, the online test is performed with dataset 2. Setting the duration as half a week wave, at a sampling frequency of 10kHz, after the fusion model configuration in dataset 2, the corresponding segments are selected at each fusion data point, each segment is 100 points in length for online testing of the trained CNN, and these segments obtained are also directly referred to as the segments in dataset 2. The protection scheme consists of a fault initiation determination scheme and a fault type discrimination scheme, and the two sub-schemes in the power system protection scheme will be subsequently validated and analyzed. When an internal fault is recognized, the protection is initiated first, and after 1/4 cycle, if the same fault type is detected at c consecutive points, a trip instruction A is sent to the processor, and in the next small period of time, for example, 1/8 cycle of time, the processor compares the instruction A with the trip instruction B obtained from the existing protection scheme, and the protection is tripped if the trip instruction B is a trip signal. If the signal B has been a non-trip signal, the decision of whether to trip is made according to the importance of the protection for security and dependability in different scenarios, for example, in scenarios with high security, the protection is not tripped on the basis of the instruction B. In scenarios with high dependability requirements, the protection is tripped on the basis of the tripping instruction A from the CNN.

Figure 3.

The flow of power system protection scheme

In the face of harsh environmental influences, for power system control is a small test, must have a good reliability and stability. PLC as a processor with excellent logic control and extraordinary anti-interference ability, to solve the traditional controller large size, poor reliability and other shortcomings. Therefore, this design selects SIEMENSS7-1200PLC as the core controller, the controller has a compact structure, powerful, flexible and adaptable to harsh environments, etc., able to complete the calculation, logic control, motion control, network communication and other tasks. It represents the future development direction of small PLC and leads the automation trend.

This design selects 1214CDC/DC/DC module as CPU module, its power supply is DC 24V, with 14 digital inputs and 10 digital input ports and 2 analog input and output ports. It supports PROFINET interface and can communicate with other devices. As many as up to 6 high-speed counters for calculation and timing. Featuring 3 communication modules for serial communication and 8 signal modules for I/O expansion, the CPU has a strong self-diagnostic fault function and is easy to maintain on a daily basis.

In addition to ensure that the design requirements need to add an additional digital module SM1223DC/DC and CM1241RS485/422 communication module, the former through the 16 digital input 16 digital output channels to meet the system’s digital demand, the latter supports point-to-point communication, using RS485 as a transmission medium to communicate with the DSP, the DSP through the transmission medium will be the voltage, RS485 is used as the transmission medium to communicate with DSP, and DSP transmits power parameters such as voltage, current and power factor to PLC through the transmission medium.

In order to PLC work stably and reliably, all PLC power supply in this design comes from 4NIC-K600 aerospace power supply, which has the advantages of high reliability, simple installation, humidity resistance, salt spray resistance, etc., and has the control functions of short circuit, overcurrent, overvoltage, etc., and it can be applied to the power system. Its specifications for the input AC 220V, 50Hz, output DC 0 ~ 30V or 0 ~ 10A, the maximum power output up to 300W, can work in harsh environments. This design utilizes a PLC digital output module to drive a relay to control the system’s generator set as well as the display light and main switch. This design uses OMRON small relay, OMRON relay has 4 pairs of contacts inside; it has more than 80 million times of mechanical life and more than 200,000 times of electrical life; it has the function of display light when the relay is in action; it can operate normally under the temperature of -55°C~70°C; it has high reliability and is widely applicable to various uses and other characteristics.

When performing hardware configuration in TIA portal, the model number, built-in program, and mounting location of each module configured must be consistent with the actual module installed to ensure that the designed program is downloaded into the PLC smoothly. Figure 4 shows the PLC hardware configuration of the power system. From left to right, the PLC communication module, PLC CPU module, PLC digital module, the connection of each module is consistent with the actual situation. Based on the experimental platform, in connection with the specific functions of power system automation, the controller PLC in this design is responsible for all digital input and output signals.

Figure 4.

Hardware configuration diagram of PLC

Let the capacity of a group of capacitors is Q_C, Q_C = P tan φ, the active power transmitted by the power system is P, and the power factor angle is φ; let the power factor transmitted by the power system is λ₁. When a group of capacitors is put in, the power factor after putting in a group of capacitors is obtained by the calculation of Equation (8) and Equation (9) so as to get the power factor of λ₂ after putting in a group of capacitors: (8)tanφ2=tanφ1−QCP (9)λ2=cos1(1+tan2φ2)

where φ₁ and φ₂ are the power factor angles before and after voltage control by dropping the capacitor bank.

According to Eqs. (8) and (9), it can be found that dropping the capacitor can produce a change in the power factor, and this change is related to P. The active power is constantly changing during the operation of power system, and the range of its change is large, so it is necessary to include the relationship between the voltage control effect and voltage loading into the criterion.

PLC carries out voltage reactive power compensation of the power system according to the bus voltage. First, through the collected bus voltage and standard voltage, the fuzzy PID controller outputs the regulated voltage value according to the real-time voltage data and the standard voltage value, combines with the collected power factor to judge the overrun region of the voltage value, and sends out the capacitor group casting and cutting commands corresponding to this region to realize the voltage regulation and control of the power system. 1)

PLC-based PID controller

In order to improve the power system voltage regulation and control effect, this paper adds the fuzzy set theory on the basis of the traditional PID controller, uses the fuzzy logic in the fuzzy set theory to deal with the error signals in order to better adapt to a variety of control scenarios, digitizes the PID to get the discrete PID equation, and designs the voltage control program through this equation.

The input and output relationship of the PID controller is: (10)M(t)=KC[e(t)+1T1∫0edt+TDdedt]+M0

In the formula, M(t) and M₀ for the PID controller output of the regulated voltage value and PID controller output of the initial results; T₁ and T₀ for the integral time constant and differential time constant, respectively; e(t) = sp(t) − p_V(t) and K_c for the standard voltage value of sp(t) and real-time voltage value of pN(t) error signals and proportionality coefficients, c(t) for the output of the PID controller.

Set T_s as the sampling period, when t = 0 is the moment of reactive power compensation system operation, respectively, using rectangular integral and differential to approximate the exact integration and differentiation, the equation (10) discretization can be obtained when the nth sampling PID controller output: (11)Mn = Mn−1+Kc[e(n)−e(n−1)]+K1e(n) +KD[e(n)−2e(n−1)+e(n−2)]

In the formula, the error value of the nst sampling is expressed as e_n; sp_n, pv_n and M_n are the nth sampling digital quantities of sp(t), pv(t) and M(t), and the differential and integral coefficients are expressed as K_D and K₁ respectively.

In order to improve the robustness of the reactive power compensation device, better stability, the use of PID controller in voltage reactive power compensation at the same time to add a fuzzy voltage controller, the composition of the fuzzy PID controller so that the reactive power compensation device in the face of different situations through the intelligent coordinator to switch between the fuzzy controller and the PID controller, the switching conditions of the reactive power compensation device operation based on the indexes of the switching conditions are as follows: 1)

The fuzzy controller is usually used when the reactive power compensation device has a node voltage change or when the reactive power compensation device has overshooting and oscillation.

1)

Input and output data composition and description

For CNN, its input data is generally a two-dimensional matrix. In order to use CNN for fault diagnosis, the fault data needs to be matrixized and reorganized. Since the electrical quantity data of the power system has a natural time sequence, it can be matrixically reorganized by taking the data from one time window. There are 3 types of short circuit faults, positive ground fault, negative ground fault and bipolar short circuit fault. For the 3 outgoing lines there would be 9 fault categories. Together with the normal operation category, the final fault identification problem with CNN is transformed into a 10-classification problem. For the multiclassification problem, the output layer generally uses the ReLu function, which sets the output probability of the largest to 1, and sets the rest to 0. Therefore, for different types of faults, the output result is a one-dimensional vector composed of 0 and 1. The output results of different fault types are shown in Table 1. In the table, the fault types 1PG, 1NG, 1PN indicate line 1 positive grounding, line 1 negative grounding, line 1 bipolar short-circuit, line 2 and line 3 have similar naming rules, and the absence of faults is indicated by NF.

The trained model is used for test data, and the power system fault test results are shown in Table 3. The rows in the table indicate the real categories of each category, and the columns indicate the predicted categories of the CNN, based on Table 3, the prediction accuracy and recall of each category can be calculated, for example, in the case of Line 1 Positive Grounding (1PG), there are 5,924 samples in which the predicted categories are the same as the real categories, and there are 6,000 IPGs in the actual number of the IPGs, so the accuracy of the 1PG is 5,924/6,000 = 98.73%, and the remaining few By analogy, the prediction accuracy, recall, and F1 values of each category are shown in Table 4. Taking NF as an example, the values of accuracy and recall are known to be 100.00% and 98.32%, respectively, and according to the formula for calculating the F1 value: F1=2*accuracy*recall/(accuracy+recall), which calculates the F1 value to be 99.15%, and the remaining 9 items are the same. For 60,000 test samples with trained neural network, the calculation time is about 18.6s, so the calculation time of a single sample is about (1000*18.6/60, 000=0.31)0.31ms, and since the time window is 2ms before and after the fault, the fault judgment time is around 3.1ms, which can satisfy the rapidity requirement. The test results show that the CNN-based power system line fault protection method has a high accuracy and recall rate, and can accurately determine the fault line and fault category. Of course, there are many factors affecting the training effect of CNN, the filter size, the number, the number of iterations is only a part of it, and there are other influencing factors, which are not verified one by one in this paper. The adjustment of parameters has also been a major problem in the neural network training process, there is no specific method. Therefore, there is no best model, only a better model under the premise of meeting the requirements.

After using PSCAD simulation software to verify the effectiveness of the CNN-based power system fault identification algorithm, the power system protection scheme constructed by it will be verified and analyzed from several aspects. The specific analysis process is shown below: 1)

Simulation analysis of fault initiation determination scheme

When the power system is running, CNN detects the positive DC voltages U_12p and U_13p of line LINE12 and LINE13 under positive ground fault and inter-pole short-circuit fault, respectively, when the fault occurs in f₁ and f₂, fault f₁ is set at the midpoint of the bus side, and fault f₂ is set at the position of 40% of the total length of the DC line, with the fault resistor R_f of 50 Ω, and the flat-wave reactor L_z of 100 mH, and the changes of the two voltages at different faults are shown in Fig. 5. The changes of the two voltages at different faults are shown in Fig. 5, where (a)~(d) are bus-side single-pole ground fault (X1), bus-side inter-pole short-circuit fault (X2), line-side inter-pole short-circuit fault (X3), and line-side single-pole ground fault (X4), respectively. Under different faults, the voltage of each fault line all decreases instantaneously, in order to determine the fault start threshold D_udc, select the voltage instantaneous drop value under the condition of different voltage changes, it can be known that the instantaneous drop value of fault voltage on the line side is higher than that on the bus side, and the drop value of line voltage U_13p outside the area is much greater than the drop value of line voltage U_12p in the area, take the maximum value U_dec therein is 326kV, and the voltage change amount ΔU_dc is 184kV, in order to guarantee fault start, leave a margin of 10%, can take D_udc as 165kV, can be used as an alternative threshold.

Figure 5.

The variation of two voltages when different faults occur

Figure 6.

Change curve of current difference I_D under different faults

Figure 7.

I_D Change curve under different fault resistances

Figure 8.

I_D change curve under different flat wave reactors

Figure 9.

The whole process of fault protection removal

In order to verify the effectiveness of PLC-based fuzzy PID controller in the application of power system control, the PSCAD simulation software described above is used to analyze the power system voltage under the action of PLC technology, and the power system voltage data and waveforms are analyzed to verify whether the method of this paper has applicability. Under the automatic voltage regulation and control strategy based on reactive power compensation, the voltage waveform of the power system output by simulation test is shown in Fig. 10, in which U_ab, U_bc and U_ac indicate the external terminal voltage of the equipment, the internal resistance voltage of the power system and the power system voltage, respectively. From the figure, it can be seen that the output voltage is distorted to a certain extent, and the burr phenomenon of the voltage curve is also more obvious, in the absence of the role of the fuzzy PID controller in the PLC, the voltage quality is significantly reduced.

Figure 10.

Voltage output waveform

The fuzzy PID controller is introduced into the power system, and the output voltage waveforms are shown in Fig. 11 through the power system voltage simulation test. From the figure, it can be seen that the introduction of fuzzy PID controller power system output voltage burrs are significantly reduced, and the quality of the power system voltage output waveform is improved to a large extent.

Figure 11.

The voltage output waveform is controlled by PLC

According to the constructed system model, a 48KW frequency converter governor experimental platform was established, the RMS value of the configured voltage was 24KV, and the maximum output value of the current was 2KA, which was subjected to Fourier analysis, and before and after the optimization of the PLC-based fuzzy PID controller, the voltage waveform data of the power system was shown in Table 5. From the data in the table, it can be seen that the maximum, minimum and RMS output voltage growth rate of the power system are 4.92%, 0.57% and 9.13% respectively, which proves that the PLC-based fuzzy PID controller has a coordinating effect on the power system and ensures that the power system operates at a higher speed of performance state.

The equipment speed of the power system is analyzed as shown in Fig. 12. In order to further verify the effectiveness of this paper’s control method in the application of power systems, the speed change of the motor tracking, respectively, before and after the use of fuzzy PID controller, when the motor speed of 509R / s, as well as when the motor reaches a stable speed from 509R / s down to 408R / s when the speed of the motor changes, the use of this paper’s control method, the motor’s dynamic performance and steady state performance is better! The synchronization accuracy is also significantly improved, indicating that the method in this paper effectively improves the real-time and responsiveness of the power system. Overall, under the joint effect of CNN-based protection scheme and PLC-based fuzzy PID controller, the power system operates stably in a more efficient and low-consumption state.

Figure 12.

Comparison result of motor speed curve