Abstract
Taking the western Qinling Mountain, in the southern Shaanxi Province of China, as an example, based upon comprehensive analysis of geological data for 20 debris flow gullies, the author has put forward a series of indices system and has developed one evaluation system called "dangerous degree evaluation system of mine debris flow based on IGA-BP". This system adopts Visual Basic 6.0 and Access technology to manage database, adopts immune genetic algorithm to optimize the hidden layer structure and network parameters of BP neural network and adopts sample model of mine debris flow whose dangerous degree has been known to realize the BP neural network evaluation of the debris flow risk which to be determined. The calculating results show that this evaluation method has high reliability and simplicity of operation, and it can make comprehensive evaluation precisely. The evaluation results have important guiding significance in the prevention and reduction mine debris flow.
Introduction
In mining regions such as China's western Qinling Mountains, the combination of steep terrain, loose waste slag (tailings), and heavy seasonal rainfall creates a high risk for catastrophic debris flows. Historically, evaluating the risk of these events has relied on subjective methods like "torrent classification" or fuzzy comprehensive evaluation. These approaches often suffer from "man-made scoring," where human experience heavily influences the weighting of environmental factors, leading to potential inconsistencies.
While standard artificial neural networks (ANNs) offer a way to model these non-linear relationships, they frequently encounter the "local optima" problem, where training gets stuck on sub-optimal solutions. To address these challenges, this research introduces a hybrid system integrating an Immune Genetic Algorithm (IGA) with a Back Propagation (BP) neural network to automate and refine risk assessment.
Methodology
The proposed IGA-BP system integrates three distinct components to create a robust evaluation framework:
- BP Neural Network: A standard multi-layer network maps input geological data to specific risk levels. Through iterative testing, a 33-8-3 architecture (33 input indices, 8 hidden neurons, and 3 output neurons) was determined to be optimal for this geological context.
- Immune Genetic Algorithm (IGA): This is the core optimization engine. Unlike standard Genetic Algorithms that may suffer from premature convergence, the IGA incorporates "immunological memory" and "diversity maintaining" functions. This ensures the search for network parameters explores a wider range of solutions to find a global optimum.
- System Implementation: The software was developed using Visual Basic 6.0 with an Access database to manage 33 distinct input indices, including terrain physiognomy, lithological character, and mining development intensity.
Results and Discussion
The system was trained and tested using geological data from 20 debris flow gullies with known danger levels. The IGA-BP model demonstrated exceptional precision during the distinction examination phase. The average relative error was recorded at 5.1%, with an average absolute error of approximately 0.015 to 0.051 depending on the sample set.
Crucially, the system achieved a 100% success rate in correctly categorizing the danger levels of test samples into three categories: No risk, Subsidiary risk, and Risk. When applied to 10 "unknown" gullies, the system successfully identified high-risk areas like "Sigou" and low-risk areas like "Tangjiagou," providing objective data for disaster prevention planning.
Conclusion
The IGA-BP evaluation system represents a significant step toward objective, automated disaster risk assessment in mining regions. By leveraging immune theory to optimize neural network training, the system overcomes the limitations of subjective scoring and local optima. While the model currently requires high-quality on-site geological data, its high reliability and precision make it a valuable tool for guiding mine safety and debris flow prevention efforts.
Summary updated