Wireless Sensing Deployment Using Transfer Learning for Building Layouts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing wireless sensing devices require extensive on-site data collection and long training times to adapt to complex building environments, leading to inefficient deployment and high operational costs.
Innovation Solution
A deployment method and system utilizing transfer learning to optimize pre-trained neural networks for wireless sensing devices, incorporating spatial structure information and on-site data to rapidly configure and deploy the devices, reducing the need for extensive on-site operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If pre-trained neural network models are used for wireless sensing device deployment, then model adaptability to building environments is improved, but deployment time and training duration increase significantly
Solution Approach 1:
The patent applies preliminary action by pre-training neural network models on simulated line-of-sight data before actual deployment. This pre-training phase prepares the models in advance with fundamental sensing capabilities, so that when deployed in actual buildings, they only need minor fine-tuning on collected on-site data rather than complete retraining, thus reducing overall deployment time while maintaining adaptability.
Solution Approach 2:
The patent uses partial action by implementing a two-step training process where the first step (training on simulated data) provides sufficient foundational learning, and the second step (fine-tuning on measured data) provides only the necessary adjustments for specific environments. This partial approach avoids the excessive time consumption of complete retraining while achieving adequate model adaptability.
2Measurement precision
If complete model retraining is performed for each building environment, then measurement precision is improved, but productivity and deployment efficiency decrease
Solution Approach 1:
The patent performs preliminary training on simulated line-of-sight data to establish baseline sensing accuracy. This pre-prepared knowledge allows the model to achieve reasonable sensing performance without complete retraining for each new building, thus maintaining productivity while preserving adequate measurement precision through subsequent fine-tuning on actual environment data.
Solution Approach 2:
The patent creates a copy of the pre-trained model for each deployment scenario and performs only fine-tuning on the copied model rather than retraining from scratch. This copying approach preserves the valuable learned features from simulated data while adapting to specific building environments, balancing sensing accuracy with deployment efficiency.
3Reliability
If extensive on-site data collection is conducted for model fine-tuning, then model reliability in actual environments is improved, but deployment complexity and operational requirements increase
Solution Approach 1:
The patent applies partial action by collecting and using only a limited amount of on-site measured non-line-of-sight data for fine-tuning, rather than requiring extensive data collection. The pre-trained model from simulated data provides a strong foundation that requires minimal additional training data to achieve reliable performance in actual environments, thus reducing deployment complexity while maintaining reliability.
4Adaptability or versatility
If two-step training process is implemented, then model adaptability to different line-of-sight conditions is improved, but use of energy and computational resources increases
Solution Approach 1:
The patent uses partial action by implementing a two-step training process where the majority of training is performed on simulated data (first step), and only a small amount of fine-tuning is performed on actual measured data (second step). This approach achieves good line-of-sight adaptability while minimizing computational energy consumption compared to complete retraining on real data.
Solution Approach 2:
The patent creates a pre-trained model copy from simulated data that captures general line-of-sight characteristics, then applies minimal fine-tuning to adapt to specific deployment conditions. This copying and fine-tuning approach reduces computational energy requirements while maintaining adaptability to different line-of-sight scenarios.
Data Source
Figure 1~2
Figure 3
AI summary
A deployment method and deployment system for a wireless sensing device. The deployment method for a wireless sensing device includes S1: obtaining spatial structure information of a region to be deployed; S2: calculating a layout of the wireless sensing device; S3: calculating a signal strength according to the layout, and judging whether the signal strength meets requirements; S4: judging whether a default neural network model meets the requirements according to signals received by the wireless sensing device; if the requirements are not met, optimizing a pre-trained neural network model through transfer learning according to the signals received by the wireless sensing device; and S5: deploying the optimized neural network model into the wireless sensing device.