Multi Objective Optimization Framework for Irrigation Systems Based on Deep Learning: A Robust Decision-Making Method Integrating CNN BiLSTM and NSGA II
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Climate variability and nonstationary inflows make irrigation planning a data-intensive decision problem. From a computer-science perspective, we formulate end-to-end irrigation optimization as a pipeline that tightly couples learning and search: (1) an attention-augmented CNN–BiLSTM forecaster for soil-moisture time series, (2) a GA-accelerated mixed-integer layout optimizer for pipes and tanks, (3) a Monte Carlo–driven bilevel robust program for survival assessment under stochastic inflow reduction, and (4) an NSGA-II multi-objective scheduler balancing operating cost and irrigation satisfaction. On real deployments, our forecaster improves R2from 0.74 to 0.87 and reduces MSE from 0.072 to 0.067 (≈7%); the layout module attains the minimum build cost of 215,320.97 CNY while satisfying coverage and connectivity; the robust component quantifies resilience and recommends emergency storage ratios that rise with drought probability (e.g., ≈8.38% at 50%, 10.64% at 80%, 32.52% at 100%); and the scheduler yields diverse Pareto solutions, with the selected plan meeting seasonal demands using river withdrawals alone (no reservoir drawdown). Quantitatively, these results demonstrate algorithmic gains in predictive accuracy, cost efficiency, and robustness. Methodologically, the paper contributes a generalizable "learn-then-optimize" framework that integrates deep sequence modeling with combinatorial, stochastic, and multi-objective optimization for infrastructure decision-making at scale.
Fields of Study & Subjects
Water resources management and optimization
Irrigation Practices and Water Management
Water Systems and Optimization
Publication details
- Journal / source
- IEEE International Conference on Power, Electronics and Green Energy (ICPEGE)
- Pages
- 1808–1813
- Publication date
- 28 January 2026
- Language
- EN