Genetic algorithm optimized BP neural network for fast reconstruction of three-dimensional radiation field

Area of Science:

• Nuclear Engineering
• Artificial Intelligence
• Radiation Detection

Background:

• Accurate 3D radiation field mapping is crucial for nuclear facility safety and radiation protection.
• Current 2D reconstruction methods and grid-based measurements are limited.
• Artificial intelligence, particularly neural networks, offers potential for advanced radiation field reconstruction.

Purpose of the Study:

• To develop an efficient method for reconstructing 3D radiation dose rate fields at any spatial position.
• To address the limitations of existing 2D and grid-dependent reconstruction techniques.
• To leverage AI for faster and more accurate radiation field analysis.

Main Methods:

• An improved Genetic Algorithm Optimized Backpropagation (GA-BP) neural network was developed.
• The method reconstructs radiation dose rates in 3D space, even with low sampling rates.
• Effectiveness was validated using Monte Carlo simulations (shielded and unshielded) and real experimental data.

Main Results:

• The GA-BP network achieved rapid reconstruction, processing nearly 500 spots in 0.01 seconds.
• Reconstruction of the 3D dose rate field used less than 6% of data with low error (3-8%) in simulations.
• Experimental validation showed a 15% overall error and <30% point error in most areas.

Conclusions:

• The proposed GA-BP neural network method enables efficient and accurate 3D radiation field reconstruction.
• The AI-driven approach significantly reduces data requirements and processing time.
• This method enhances radiation protection strategies in nuclear facilities.