DEVELOPMENT OF AN ALGORITHM FOR ANALYZING CALCULATIONS FOR DETERMINING THE LOCATION OF DAMAGE

ABSTRACT

The article discusses the methods used to determine the location of damage on the experience of operation of the Rostov enterprise of the main electric networks and describes their shortcomings. An algorithm based on a neural network is proposed to improve the accuracy of determining the location of damage on power lines.

Keywords: algorithm; neural network; power lines.

There are topographic and remote methods for determining the location of damage. Topographical methods include electromechanical and induction methods, remote methods include pulse methods and methods based on the parameters of the emergency mode. A variety of methods for the parameters of the emergency mode are one-way and two-way methods for determining the location of damage, which are widely used in Russia due to the availability of the necessary software systems for determining the location of damage. Such PCs as WinBres, "ARM SRZA", etc. they are most often used to calculate the distance to the short circuit location using the parameters of a damaged power line.

Simplicity of implementation, availability of source data - these are the main advantages of this method, but there are also features that affect the accuracy of determining the short-circuit point. The main factors that have a negative impact on the error of determining the location of damage by the parameters of the emergency mode are:

• inaccurate power line parameters;
• ambient temperature;
• the presence of other lines in the vicinity;
• unstable value of the transient resistance;
• weather conditions (season of the year).

To more clearly reflect the existing problems associated with determining the location of damage, it is necessary to consider an example of calculating one of the power lines in emergency mode and evaluate the data obtained using modern PCs. As an example, consider a certain power line in the Rostov region, the beginning of which is designated as "P", and the end as"B". 500 kV overhead line with a length of 433.6 km, where an emergency occurred. Taking all the necessary data obtained from the devices, you can start using the calculation methods for determining the location of the damage.

The results of the fault location determination calculation shown in Figure 1 were obtained by the fault location detection devices, calculated by the «WinBres» PC, the «ARM SRZA» PC, and manually by the oscillograms of the zero and reverse sequences currents and voltages. For a qualitative analysis, it is necessary to include all the methods in the study in order to obtain a greater variety of the location of the short-circuit point in the damaged area. [1]

Using a graphical program to display the results obtained, we get a complete picture of the methods under the conditions of this damage (Figure 1).

In the figure above, the first digit is responsible for the method used to determine the location of the damage, and the second for the variety or device:

• 1.х – according to the protocol of the PC "ARM SRZA" of the zero sequence;
• 2.х – on devices for determining the location of damage;
• 3.х – by zero-sequence formulas;
• 4.х – according to the protocol of the PC "ARM SRZA" of zero sequence and reverse sequence;
• 5.х – according to the protocol of the PC «WinBres».

For ease of perception, some methods are marked with different colors, and the red vertical is the place of the short circuit.

Based on the calculations, it is impossible to single out a method or a software package that will easily cope with the solution of the task. The spread of the results is quite large, and considering this problem in the conditions of power lines, the length of which is more than 400 km, it is necessary to find the most accurate option, where the error will be minimal.

Figure 1. Graphical representation of the calculation result

In order to approach the working algorithm of an artificial neural network, it is necessary to correctly set the tasks that the artificial intelligence must cope with as an assistant to determine the location of damage. [2]

The damage location is determined using software packages and other methods, which means that the input data for the neural network will be the output data obtained from these sources. In addition, you need to take into account the length, voltage, and short-circuit currents. It is not uncommon for a calculation to result in any one result noticeably different from others that are in a uniform range, therefore, to exclude it at an early stage is a good solution, so as not to mislead the program during the subsequent analysis. Since it is quite difficult to identify the most true result from the remaining results under the conditions of this damage, additional data is needed, which is preloaded into a certain database, which will guide the neural network throughout the entire time to achieve the task. The contents of the repository should be pre-filled with information related to previous calculation experience, and it is desirable to take into account all the factors that can contribute to the correct decision. Weather conditions, incidents of any nature, the temperature of the environment, the time of year, and more. Based on the above, the algorithm of an artificial neural network that will satisfy the conditions should look like this:

1. Input of initial data: at this stage, the neural network receives information about the calculations obtained using the methods used;

2. Primary analysis of input data: artificial intelligence removes a deliberately incorrect value, which can be detected by various errors in the calculation of programs or other methods due to incorrectly entered data, inaccuracies in calculations or instrument readings;

3. Secondary analysis: due to the need for the most accurate data selection, the artificial neural network compares the remaining results of damage location calculations after the primary analysis with the accumulated experience (database);

4. Output of the result: selecting the most appropriate option or options, if there are several of them from the results of the secondary analysis;

5. Updating the database: after receiving the true result about the location of the short-circuit, the neural network independently supplements the information library.

Figure 2 shows a visual representation of the operation of the artificial neural network algorithm.

Figure 2. The scheme of the artificial neural network

Figure 2 shows the scheme of the neural network, where X1, X2..., Xn – the input data obtained by using known programs to determine the location of the damage. The data then goes to the processing area - P, where it is analyzed in order to exclude deliberately incorrect calculations, if any, obtained by this or that method. The remaining results go to the M processing location, which includes additional input data: Xr-external factors (temperature, weather conditions, etc.) and Date – the date and time of the incident. Taking into account the accumulated experience, the library-Lib, the artificial neural network completes the algorithm and receives the output data-N. P2 combines actions related to external factors and the library, since this is an interconnected process: Date and Xr are compared with the data in the Lib.

Conclusion: the use of this algorithm by artificial intelligence as an assistant in determining the location of damage makes it possible to find new patterns between the factors affecting the calculations and the calculations themselves. In the future, this can serve as a starting point for creating optimized software packages that take into account the maximum amount of source data that affects the final result, which will reduce the error.

References:

1. Derevyanko R. D., Fakhrutdinov A. Sh., Dolgov N. A., Pavlenko A. A. Fedoseev N. A. Problems of DLD on lines of 220 kV and higher. Experience of operation of the Rostov EMEN / / Actual problems of science and technology. 2020. [Electronic resource]: materials of the National Scientific and Practical Conference : (Rostov-on-Don, March 25-27, 2020) / ed. by N. A. Shevchenko; Don State Technical University. univ. - Rostov-on-Don : DSTU, 2020. - p. 818-820 - URL: https://ntb.donstu.ru/;
2. Rashid, Tariq Create a neural network. : Translated from English - Saint Petersburg: Alfa-kniga LLC, 2017. - 272 p.: il. - Paral. tit. eng.