ANALYSIS OF THE BACKPROPAGATION ALGORITHM IN PREDICTING WATER VOLUME OF PDAM TIRTAULI PEMATANG SIANTAR CITY

: Increasing living standards cause an increase in the need for drinking water. However, current water supply estimates are still not optimal, with water production sometimes being more or less than requirements. To estimate the amount of water, an appropriate method is needed. The method used in this research is the back propagation algorithm artificial neural network method. When developing forecasts, past data is necessary to produce accurate results. This research aims to develop a predictive model that can estimate the volume of water that will be used by PDAM Tirtauli in the future. It is hoped that this predictive model can help PDAMs in planning more efficient water supply management and can reduce the potential for water supply shortages in the future. This research uses water distribution data for the 2015-2022 period. Training data starts in 2015-2021, testing data starts in 2016-2022. In this research, results were obtained using the Matlab R2011a application. In this research, the 5 architectures used are architecture 6-53-1, 6-58-1, 6-61-1, 6-81-1, 6-87-1. Based on these five architectures, the best architecture was obtained, namely architecture 6-87-1 with a root mean square error test value of 0.00010031 and an accuracy of 92%. The results achieved in 2023 are the total water volume of PDAM Tirtauli Pematangsiantar of 189,610,426.


INTRODUCTION
Water is the main source of supply and is very important for daily human needs and everyone has the right to use clean water [1] .Drinking water is a basic need that cannot be separated from human life.Its use is not only for household purposes but also for other installations.Along with population growth, development progress and increasing living standards, the need for drinking water continues to increase [2].Therefore, the community must be consistent in the quality of business management and drinking water services.is a company operating in the field of drinking water supply.One of the objectives of establishing PDAM is to meet the community's drinking water needs, including provision, development of infrastructure and services, as well as distribution of drinking water [3].PDAM as one of the BUMDs can make a full contribution as a community service and can contribute to local original income (PAD).This makes the service quality of PDAM Tirtauli Pematangsiantar City and the clean water management company very important for the community.However, current water supply estimates are still not optimal, with water production sometimes being more or less than requirements.High water consumption causes the need for drinking water to continue to increase, while the supply of drinking water continues to decrease every year, along with the large amount of empty green land being used for housing or buildings [4].This is of course a matter of wasting water because PDAM or consumers lack water.If this situation continues it will be a big problem for PDAM and the community [5].
Predictions are needed to calculate PDAM Tirtauli's monthly water distribution.When making predictions, you need to have the right way to complete them so that the predictions produce accurate results.Artificial neural networks are a great method that can be used to predict the problems described above.In artificial neural networks there are several methods, one of which is the method used in this field research practice report, namely the backpropagation algorithm.The backpropagation algorithm is a popular and widespread mathematical tool used to predict and estimate time, as well as determine the results of nonlinear functions [6].Backpropagation is an algorithm that uses the error value in the output to change the inverse weight and uses a differential activation function for the forward step [7].This algorithm is often used to solve complex problems.Indeed, this algorithm is trained using learning methods.By using this technique, it is hoped that a system can be created that can predict the distribution of PDAM water in PDAM Tirtauli, Pematangsiantar City and can help PDAM Tirtauli, Pematangsiantar City to calculate the amount of PDAM water distribution [8].
Some of the previous research that served as a guide for writing this article included research on predicting clean water needs in Malang district PDAMs.This research used training data and test data from 2013 to 2018 in the period January to December.This research resulted in a research accuracy rate of 89.72% [9].In research predicting the installation of the number of clean water supply facilities in PDAM Pematang Siantar, in this research if the data was checked the accuracy was 89% [10].

METHOD
This chapter discusses the systematic procedures or methods used by researchers to find the truth of a phenomenon through logical considerations and supported by factual data as real evidence (objective, not personal hypothesis).Following are the steps that will be taken: The first step is to collect data.In this research, data was obtained from PDAM Tirtauli Pematangsiantar.Then the literature review aims to complete the background knowledge and theories used in this research.Apart from that, literature study is also the first step in the research process carried out in this article.The literature study used comes from journals.Then Identifying the Problem.This stage is carried out after obtaining a collection of data that will be processed at the data transformation stage according to the bot that has been determined.After that Determine the Pattern.At this stage a backpropagation model will be generated which will determine the pattern.Testing Data Processing Results.After completing the model determination, testing will be carried out using the Matlab R2011a application.After finding a model that passes the Matlab application test, predictions will be carried out to compare the model results with the highest and most accurate results.Final Evaluation.
Has a weight that will strengthen or weaken the signal, to determine the output, each neuron uses an activation function that is applied to the amount of input received [11].ANN is determined by 3 things, namely [12] : (1) Pattern of connections between neurons (called network architecture), (2) Method for determining link weights (called train-ing/learning method)., (3) Activation function, which is the function used to determine the output of a neuron.

Backpropagation Algorithm
Backpropagation is an algorithm that uses the error value in the output to change the inverse weight and uses a differential activation function for the forward step [7].
This algorithm uses the output error to change the value of the weights in a backward direction.The Backpropagation method involves three layers, namely: the input layer, where data is introduced to the network; hidden layer, where data is processed; and the output layer, where the results of the given input are produced [13].

Prediction
Prediction is the process of estimating future needs.Predictions are also estimates of future needs to meet demand for goods and services that take into account the criteria of price, quantity, time, place and quality.[7].In predictions you don't have to give a definite answer about events that will happen in the future, but rather try to find accurate answers that might happen later [14].

Matlab
Matlab (Matrix Laboratory) is a high-level, closed, case-sensitive programming language in a digital computing environment developed by Math-Works [15].

RESULTS AND DISCUSSION
This research carried out several preparatory steps, including collecting data and distributing the data to the Matlab application, then looking for prediction values.

Data processing
The data used is data for 2015-2021, this data will be used as training data, test data and will be implemented in the Matlab application.As seen in Tables 1 and 2, the table displays the training data and test data before the data is normalized using the sigmoid function.

Data Normalization
Data normalization is carried out using the backpropagation algorithm process using a binary sigmoid activation function with a range of 0 < x < 1.The maximum value of the binary sigmoid activation function is 1, so that the original data in integer form is converted into a decimal number with a range of [0,1].[0.9] using the formula presented in equation 1 [16].In this research, before implementing the Matlab application, the input data will be determined first.We can see in the table below that the data entered for this search contains 8 variables.Initialize and assign values to variables that will be used during calculations.Input variables are variables that are used as input during the calculation process.

Output Determination
The backpropagation algorithm requires output to achieve the desired result.The results are necessary to determine accurate future results.In this study, the results were determined by the learning rate <=0.02 correct and >=0.02 incorrect.

Table 2 .
Test Data Before Normalization

Table 5 .
Input Data

Table 4 .
Test Data After Normalization