Application of Genetic algorithm and Artificial Neural Network in Water Consumption Forecasting and Driving Factors Determination
| dc.contributor.advisor | Mebruk Mohammed (PhD) | |
| dc.contributor.author | Terefe Tilahun | |
| dc.date.accessioned | 2025-03-05T09:32:10Z | |
| dc.date.available | 2025-03-05T09:32:10Z | |
| dc.date.issued | 2024 | |
| dc.description.abstract | Accurate understanding of water consumption is paramount for effective resource management, infrastructure planning, and ensuring a reliable water supply. Understanding and determining the driving factors of water consumption has become a key challenge. In order to address this issue, the study employs Genetic Algorithm (adopted for both linear and nonlinear regression) along with an ANN model consisting of one hidden layer with either one or five nodes. The GA and ANN models were used to predict water consumption in Addis Ababa city and analyze the driving factors behind water consumption. The model was developed using input data such as water consumption time series, average temperature, population, construction activity, relative humidity, economic development, number of livestock, industrial development, and holiday/festival. Monthly data on water consumption and meteorology (from 2015 (June) to 2023(August)) were gathered from Addis Ababa city water and sewerage authority, as well as the National Meteorological Agency. Sensitivity analysis is used in the forecasting process to choose the most important explanatory factors. Four different models were developed and their performances were assessed using two metrics: root mean squared error (RMSE) and Normalized root mean squared error (NRMSE). The linear regression GA model achieved an RMSE value of 0.355 Mm3/month and an NRMSE value of 0.0451. On the other hand, the nonlinear regression GA model yielded an RMSE value of 0.339 Mm3/month and an NRMSE value of 0.0430. Moving on to the ANN model with one hidden node, it achieved an RMSE value of 0.325 Mm3/month and an NRMSE value of 0.0413. Lastly, the ANN model with five hidden nodes achieved the lowest RMSE value of 0.3195 Mm3/month and the lowest NRMSE value of 0.0405. Among the four models, the ANN model with five hidden nodes performed the best, according to these results. The study conducted using ANN model with five hidden nodes revealed that the primary factors influencing water consumption in Addis Ababa city are population, relative humidity, and industrial activity. Population was identified as the most significant factors, with an effect size of 18.73%, followed by relative humidity at 16.65% and industrial activity at 12.77%. The additional factors play a substantial role (ranging from 1.1% for average temperature to 11.3% for livestock), to the point that neglecting them in water consumption calculations could result in inaccuracies when forecasting future demand trends.It is recommended that future predictions of water consumption in Addis Ababa city take into account nine driving factors: population, average temperature, construction activity, relative humidity, economic development, agricultural activity, industrial development, holiday/festival, and precipitation. | |
| dc.identifier.uri | https://etd.aau.edu.et/handle/123456789/4466 | |
| dc.language.iso | en_US | |
| dc.publisher | Addis Ababa University | |
| dc.subject | Addis Ababa City | |
| dc.subject | Artificial Neural network | |
| dc.subject | Genetic Algorithm | |
| dc.subject | Water Consumption Driving factors | |
| dc.title | Application of Genetic algorithm and Artificial Neural Network in Water Consumption Forecasting and Driving Factors Determination | |
| dc.type | Thesis |