P3TRTA-RDQL: a crowdsensing task allocation scheme integrating privacy protection protocol and low-dimensional reinforcement learning

Crowdsensing, as an emerging data collection mode, demonstrates great potential in Internet of Things (IoT) applications. However, it faces a critical trilemma: accurate task allocation depends on node proximity to the task location, but disclosing location data risks privacy leakage, while concealing it reduces allocation precision. Existing solutions either incur high computational overhead (encryption), rely on unfeasible trusted third parties (anonymization), or degrade data utility (obfuscation), failing to balance privacy, accuracy, and efficiency. To address these issues, this paper proposes the P3TRTA-RDQL scheme, combining a symmetric encryption-based privacy protection protocol (P3TRTA) with a low-dimensional Q-learning algorithm (RDQL). The P3TRTA protocol uses a location-based symmetric key generator (LSKeyGen) to protect node/task location privacy and proxy re-encryption to secure task content, eliminating reliance on trusted third parties. The RDQL algorithm reduces state dimensionality by 60% compared to traditional reinforcement learning, enhancing large-scale task allocation efficiency. Experimental results show that P3TRTA-RDQL outperforms existing methods by 30% in privacy protection strength, achieves 98% task allocation accuracy, and reduces allocation time for 1000 tasks by 40%. This work provides technical support for crowdsensing’s widespread IoT applications.