رگرسیون خطی و روش های عصبی برای پیش بینی تقاضای آب در مناطق آبیاری با سیستم های تله متری

Information regarding water demand is key to managing consumption in irrigation districts. Forecasting water demand is one of the main problems for designers and managers of water delivery systems. This paper evaluates the performance of linear multiple regressions and feed forward computational neural networks (CNNs) trained with the Levenberg–Marquardt algorithm for the purpose of irrigation demand modelling. The models are established using data recorded from an irrigation water distribution system located in Andalusia, southern Spain, during two irrigation seasons (2001/2002, 2002/2003). A commercial telemetry system was installed on 28 farms of the irrigation network to record water volumes in real time. The input or independent variables used in various CNN and multiple regression models are: (a) water demands from previous days; (b) water demands and climatic data (rainfall, maximum, minimum and average temperatures, relative humidity and wind speed) from previous days. Good predictions were obtained when water demand original data were modified in the calibration period by a smoothing process to reduce the noise in the data acquisition during the start-up of the research project. The best predictions were obtained when water demand recorded during the two previous days was used as input data.

Demands for water are increasing in both quantity and quality; a phenomenon that is conditioned by social, political and environmental factors. The pressures to meet growing demands have led to greater competition for available water resources among traditional water consumers, namely agriculture, industry and cities. These competing interests are already limiting social, industrial and rural development actions of many countries. Furthermore, the fact that this growing demand for water is not coupled or sychronised with increased resources is giving rise to greater competition between regions or countries for access to water (FAO, 1993; Ohlsson, 1995; Sumpsi et al., 1998). The current concern for environmental protection has given rise to a new factor affecting competition for water. Certain non-consumptive uses for recreational, ecological or landscaping purposes are now being considered when assigning water for consumptive uses. Thus, not only has competition increased in terms of the amount demanded, but also the quality. As a result of these different factors affecting competition for water resources, water is increasingly considered a scarce and valuable resource requiring rigorous management and extreme caution to prevent its depletion. One of the keys to solving these problems lies in the agricultural sector given that irrigated agriculture is the largest user of water throughout the world, accounting for 87% of consumptive uses (ONU, 1997; Sumpsi et al., 1998). The improvement of water management in an irrigation district requires the analysis of water demand in order to determine ways in which it may be modified and rationalised with a view to making water management policy more efficient. Information regarding water demand in irrigated areas is key to the development of policies on irrigation water consumption. These data can provide us with information regarding the marginal value of water and the response level to different irrigation water rates. This also provides reference data for the design, modernisation and exploitation of water-delivery systems (Kadra & Lamaddalena, 2006). Daily water requirements for crop irrigation can be estimated by the rates of percolation and evapotranspiration and used for irrigation planning. Many models have been used to simulate these water requirements, from empirical or functional (Doorenbos & Pruit, 1977; Doorenbos & Kassam, 1979; Allen et al., 1998) to mechanistic (Van Aelst et al., 1988). However, water requirements calculated for irrigation planning are not always suitable for predicting actual use (i.e., consumer demand) due to changes in the field environment such as weather conditions and farm management practices, which can influence the actual amounts of water needed. Additionally, to facilitate data acquisition and irrigation system management and operation, recently developed tools, such as remote sensing and geographic information systems GIS (Hartkamp et al., 1999; Kite, 2000; Kite & Droogers, 2000; Lorite et al., 2004), and monitoring and controlling systems (Leib et al., 2003; Mareels et al., 2005; Miranda et al., 2005), have been combined with hydrologic models to assess and to improve the behaviour of irrigation schemes. Despite these advances water management in some irrigation districts is carried out using only the experience and knowledge of the administrator although there is always a need to forecast daily water demand. Significant progress in the field of forecasting has recently been made possible through advances in a branch of nonlinear system theory modelling called artificial or computational neural networks (ANNs or CNNs). The neural approaches are increasingly being applied in many fields of science and engineering and usually providing highly satisfactory results. Some of the applications of CNNs for the management of water resources include modelling the rainfall-runoff process (Hsu et al., 1995; Lorrai & Sechi, 1995; Mason et al., 1996; Abrahart et al., 1999; Tokar & Johnson, 1999; Thirumalaiah & Deo, 2000; Tokar & Markus, 2000; Chiang et al., 2004; Moradkhani et al., 2004; Anctil & Rat, 2005; Agarwal et al., 2006), short-term river stage forecasting (Thirumalaiah and Deo, 1998 and Thirumalaiah and Deo, 2000; Abrahart and See, 2000 and Abrahart and See, 2002; See & Openshaw, 2000; Cameron et al., 2002; Nayebi et al., 2006; Pulido-Calvo & Portela, 2007), rainfall forecasting (French et al., 1992; Zhang et al., 1997; Kuligowski & Barros, 1998), groundwater modelling (Roger & Dowla, 1994; Yang et al., 1997), predicting the soil water contents (Givi et al., 2004) and nitrate-nitrogen in drainage water (Sharma et al., 2003), and drought analysis (Shin & Salas, 2000), among others. Previous works on water demand forecasting both in urban supply systems and irrigation districts (Griñó, 1992; Pulido-Calvo et al., 2002 and Pulido-Calvo et al., 2003) show that the use of CNNs provide very satisfactory results. The objective of this paper was to forecast consumer demands of an irrigation area using on-farm water-use information from supervisory control system and approaches based on linear regression, traditional forecasting methods, and on computational neural networks; heuristic models included in the knowledge field known as soft-computing. The purpose of forecasting is the real-time control of the daily water uses at the farm-scale for the various crops, as proposal of improvement water supply management in on-demand irrigation districts.

In this paper, consumer water demand forecasting systems that can support decision making of the irrigation district administrators are proposed using multiple regressions and computational neural networks (CNNs). Determination coefficients higher to 92%, error magnitudes of ES lower to 20% and efficiency coefficient E higher to 0.91 have been obtained in the validation period, when water demand of the 2 days prior to forecasting is used as input or independent variables to the neural network or multiple regression, respectively. In this situation, the multiple regressions linear models and the neural approaches provide similar results. However, the CNNs performed better than the regressions when water demand and climatic variables were considered as input data. Short-term demand modelling can be used as input in methods and/or programs for the management of water-delivery systems in real time. Furthermore, this approach achieves a better fit of the pumped volumes and the real demand of the distribution network, thereby leading to a more rational use of water and energy resources. It would be of interest to broaden the methodology developed in this paper and implement it in other areas in order to establish a general model for the efficient management of water in irrigated areas. This model constitutes a first step in the analysis and forecasting of water demands and should be of aid in decision-making processes to develop efficient water management policy.