Spectrum Prediction Handoff For Wireless Sensor Network Node In Fire Monitoring System

Cognitive Radio(CR)technology plays an important role in wireless communication by adjusting parameters of transmission to adapt to the changing radio environment.In this paper,a multi-priorities hybrid preemptive resume priority(HPRP)M/G/m handoff model is proposed for fire monitoring.The HPRP strategy allows high priority SU to preempt the channel of low priority users within a certain range and improves the transmission performance of high priority SU.At the same time,in order to avoid frequent switching,the preemption behavior of high priority SU will not be allowed after exceeding the set range,so as to maintain the overall transmission performance.M/G/m queuing model considers the load of multiple channels to avoid the channel imbalance.The proposed handoff model is compared with the existing handoff model.Simulation results show that the proposed model outperforms the mentioned methods in terms of extended data delivery time and packet loss rate.Furthermore,we apply the deep Q-Network(DQN)algorithm to our spectrum handoff scheme to maximize reward in the long-term.Deep reinforcement learning algorithm combines the feature extraction function of neural network and the decision-making function of reinforcement learning,which can maintain stable prediction effect under the conditions of multiple variables and continuous data input.The proposed algorithm is compared with Q learning algorithm.Simulation results show that the DQN based model outperforms in extended data transmission time and packet loss rate.Transfer Learning(TL)algorithm is introduced to accelerate the learning process of DQN.A multi-dimensional discriminant method is proposed to select the most similar neighbor user.We transfer the parameters getting from the selected SU to initialize the newly added SU.Simulation results show that the TL algorithm accelerates the learning process of SUs.