Research On The Comfort Control In Smart Home Environment Based On Deep Learning

The comfort of the indoor environment is an important branch of smart home research.The comfort of the indoor environment directly affects the physical and mental health of the people working and living in it.A comfortable indoor environment makes people happy,physically relaxed,focused and efficient;a bad indoor environment can make people anxious,affect mood and mental conditions,and even endanger health.The current research on indoor comfort control are mainly focused on learning the user’s usage habits and equipment status to build models.This part of the research will hinder model learning when there are multiple users,users do not understand operations and errors,and lack effective objective control standards.Therefore,the thesis proposes to establish a comprehensive evaluation standard for indoor comfort,and to predict changes in the indoor environment as a model learning object,and achieve comfort control through deep learning and predictive control.The research started with the evaluation of the comfort of smart homes,and pointed out that the current smart home environment has the problem of difficulty in obtaining some parameters when calculating thermal comfort.The research proposes to use BP neural network to fit the thermal comfort calculation,and increase the discrete features to enhance its effect,and finally realize an improved thermal comfort evaluation model,which has a good evaluation effect on a certain error range.Through studying the LSTM model,using the self-attention mechanism to improve the learning of the model’s long-term dependence,and introducing outdoor climate and environmental characteristics to enhance the prediction effect,the time series prediction model of indoor environmental factors is realized.By comparing with the ARIMA model and the Light GBM model,it is found that the prediction effect of the research model of the thesis is better in some aspects,and the fluctuation of the prediction error is more stable.The research combines the fuzzy control algorithm with the LSTM prediction model to construct the fuzzy predictive control of the indoor environment to realize the control of the indoor environment comfort.The simulation results show that the indoor thermal comfort and air comfort have good control capabilities.The research proposes to use the deep reinforcement learning algorithm DQN to construct a comfort control model for smart homes,and use the Dueling network structure and Target network structure to improve it.The test results of the DQN model show that the model design has a certain effectiveness,and the control for comfort is more effective and stable than the fuzzy predictive control.