Research On User-personalized Intelligent Lighting Control Strategy Based On Deep Reinforcement Learning

In recent years,the state has strongly supported the development of smart lighting,and the lighting industry has transitioned customization to light customization.Lighting industry will undergo a transition from customization of lamps to customization of light.But current intelligent lighting only stays at a level of replacing traditional switches with mobile phones or remotes.It still needs human participation,can only realize operation by sending a control command,has not realized real intelligence,lacks independent learning,and cannot realize non-inductive operation.To solve different needs for light environment of different groups of people,based on deep learning and intensive learning algorithms,research on intelligent lighting control strategies applied in family scenes is done in this paper,thus achieving the goal of intelligent lighting.Research is conducted from the following two aspects:(1)In the first part of this paper,a personalised predictive lighting system is built.Firstly,according to the user’s lighting habits,BP neural network model is used to learn,train and predict different users’ lamp use habits.The neural network model is optimized with two methods of weight optimization and parameter adjustment to find the optimal generalization ability of model.Secondly,in order to give full play to the advantages of the Long Short Term Memory Network(LSTM)model,the system will mine the information characteristics of different users’ light data to predict the next light use time.Propose a cascaded model of LSTM and BP networks in the system,the cascaded network architecture studied can further improve prediction ability.By constructing an experimental system to validate the proposed model,the experimental results show that the two models have stable performance and strong generalization ability.Compared with the BP network model,the prediction ability of the cascaded model has increased from 58.59% to 89.71%.(2)The second part of this article studies a desk lamp lighting energy-saving control strategy based on deep reinforcement learning to address the limitations of current lighting equipment control strategies.By utilizing deep reinforcement learning to interact with the environment and continuously iteratively adjust the control strategy,the optimal environmental lighting strategy can be obtained.In order to reduce the algorithm’s demand for data volume,the Deep Deterministic Policy Gradient(DDPG)algorithm was used to further optimize the table lamp control illumination prediction strategy in this study,thereby achieving energy-saving lighting.The internal structure of the energy-saving optimization control determination algorithm includes an action network and an evaluation network,designed to meet the energy consumption requirements with the lowest possible reward signal.The experiment shows that the table lamp illuminance control strategy system proposed here can reach suitable illuminance value in a short time,the standard illuminance values set for table lamps are between 350 lx and 400 lx,the absolute error reaches 0.0155 finally,and the control effect is good.The predicted pulse-width modulation(PWM)value and the output power of table lamp are recorded,significant decrease of current,voltage and power is observed,and the max decrease rate of output power reaches 42.5%.Compared with traditional table lamp,the illuminance prediction and energy-saving control strategy based on deep and intensive learning proposed here can achieve the effect of energy consumption reduction.In conclusion,with the development of AI technology and the upgrade of hardware facilities,the deep reinforcement learning model proposed by this system has certain research and application value in lighting prediction and related fields.