Intelligent lighting control system based on Internet of Things

Abstract

The invention relates to the technical field of illumination control, provides an intelligent illumination control system based on the Internet of Things, and aims to improve the accuracy, stability and intelligent level of illumination control. The system comprises a sensor acquisition module, a data denoising module, a feature extraction module, a control module, a terminal execution module, a user interaction module, a lighting device and a communication module. Firstly, the system carries out denoising processing on environment data collected by a sensor through a Viterbi trend hidden horse model, secondly, a fuzzy Q learning method is adopted to carry out fuzzification processing on environment characteristic data, meanwhile, dual time difference optimization and a multi-target reward mechanism are combined, the user comfort degree, the energy saving performance and the prediction accuracy are comprehensively considered, and a control strategy is generated; the system can automatically adjust the illumination brightness, the color temperature and the on-off state according to the real-time environment state, improves the user experience, and optimizes the energy utilization efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of lighting control, and in particular to an intelligent lighting control system based on the Internet of Things. Background Art

[0002] With the rapid development of the Internet of Things (IoT), artificial intelligence (AI) and smart home technologies, intelligent lighting control systems have become an important part of smart cities, smart buildings and home automation. However, traditional intelligent lighting control systems still have many shortcomings: First, traditional systems are deficient in data processing and lack effective data denoising and environmental state modeling capabilities. The data collected by sensors often contain high noise components, which may lead to unstable system responses if used directly for lighting control. For example, in the case of large light fluctuations or short-term personnel movement, the system may have problems such as frequent switching of lights or brightness fluctuations, affecting user experience. In addition, existing systems mostly use fixed thresholds or simple filtering algorithms to process data, which makes it difficult to accurately extract environmental state features and build a stable and accurate lighting control model. Secondly, traditional systems lack adaptive adjustment capabilities and are difficult to take into account multi-faceted coordination capabilities. Machine learning methods based on a single optimization objective often only focus on lighting brightness adjustment, and fail to comprehensively consider user preferences, energy consumption optimization and future lighting demand predictions, resulting in light pollution and electricity waste. Summary of the invention

[0003] The present invention provides an intelligent lighting control system based on the Internet of Things, which aims to solve the deficiencies of traditional intelligent lighting systems in environmental perception, data processing, adaptive adjustment and energy-saving optimization, and improve the accuracy, stability and intelligence level of lighting control; the system first denoises the data through the Viterbi trend hidden Markov model, which combines environmental state modeling with trend weight calculation, and solves the optimal hidden state sequence to eliminate sensor noise and improve the stability and accuracy of environmental data; then, the system extracts features from the denoised data, comprehensively analyzes environmental states, user behaviors and energy consumption characteristics, and ensures that the lighting control strategy can accurately match the current environmental needs. In terms of control decision-making, this system adopts the fuzzy Q learning method to fuzzy the environmental feature data to reduce the uncertainty of environmental variables. At the same time, it combines dual time difference optimization to improve the stability and adaptability of the learning algorithm by integrating short-term and long-term rewards. In addition, the system introduces a multi-objective reward mechanism, which comprehensively considers user comfort, energy saving and prediction accuracy, so that the reinforcement learning model can balance different optimization goals and finally generate the optimal lighting control strategy. This strategy can dynamically adjust the lighting brightness, color temperature and switch status according to the real-time environmental status, and realize precise and adaptive intelligent lighting control, thereby improving user experience and optimizing energy efficiency.

[0004] The present invention provides an intelligent lighting control system based on the Internet of Things, which includes a sensor acquisition module, a data denoising module, a feature extraction module, a control module, a terminal execution module, a user interaction module, a lighting device, a communication module 1 and a communication module 2;

[0005] The sensor acquisition module collects environmental data through ambient light intensity sensors, human presence sensors, temperature and humidity sensors, and current power consumption sensors;

[0006] Environmental data includes ambient light intensity - judging external lighting conditions, human presence - detecting whether there is someone in the room, temperature and humidity information - providing supplementary comfort information, current power consumption - monitoring current energy consumption, and user brightness preference - recording user habits based on historical data;

[0007] The data denoising module constructs a Viterbi trend hidden Markov model through environmental state modeling and trend weight enhancement, and denoises the environmental data through the Viterbi trend hidden Markov model to obtain denoised environmental data;

[0008] A feature extraction module extracts environmental state feature data, user behavior feature data, and energy consumption feature data from the denoised environmental data, and integrates them to obtain comprehensive environmental feature data;

[0009] The control module builds a multi-objective lighting reinforcement learning model through fuzzy Q learning, dual time difference optimization and multi-objective reward mechanism. The multi-objective lighting reinforcement learning model analyzes the comprehensive environmental feature data and generates the optimal lighting control strategy.

[0010] The terminal execution module generates control instructions and controls the lighting equipment according to the optimal lighting control strategy;

[0011] User interaction module provides a remote control portal, where users can view and control lighting equipment in real time through a mobile phone APP;

[0012] Communication module 1, realizes data interaction between the control module and the terminal execution module through the Internet of Things communication protocol;

[0013] Communication module 2 realizes data interaction between the terminal execution module and the lighting equipment through the Internet of Things communication protocol.

[0014] Furthermore, the data denoising module obtains the process of generating environmental data, which specifically includes the following contents:

[0015] Step B1: Define the double-layer hidden state and observation variables, take the environmental data as the observation variables, and build the environmental state DHMM model;

[0016] Step B2: Combined with the environment state DHMM model, the optimal hidden state sequence is solved by the Viterbi algorithm to generate denoised environment data.

[0017] Furthermore, step B2 specifically includes the following contents:

[0018] Step B21: Introduce environmental trend weights and recursively calculate the probability of the optimal path;

[0019] Step B22: Backtrack and calculate the optimal hidden state sequence according to the optimal path probability;

[0020] Step B23: Calculate the denoised environment data based on the optimal hidden state sequence.

[0021] Furthermore, the control module generates the optimal lighting control strategy, which specifically includes the following contents:

[0022] Step S1: define the state space and action space, initialize the environmental fuzzy Q value and lighting decision strategy; map the comprehensive environmental feature data to the state space; the action space includes: brightness adjustment, color temperature adjustment, mode switching and switch control;

[0023] Step S2: converting the comprehensive environmental feature data into environmental fuzzy Q value through fuzzy Q learning, the conversion process includes fuzzifying the environmental feature data, calculating the activation strength of the fuzzy rules, and calculating the environmental fuzzy Q value;

[0024] Fuzzy environmental characteristic data: define fuzzy sets, map environmental state characteristic data, user behavior characteristic data and energy consumption characteristic data to corresponding fuzzy sets, and calculate the fuzzy membership;

[0025] Calculate the activation strength of fuzzy rules: define fuzzy rules, calculate the weights of different fuzzy rules based on fuzzy membership, and determine the impact of the current environment state on different rules;

[0026] Calculate the environment fuzzy Q value: Use fuzzy rules to weight the environment fuzzy Q value so that the environment fuzzy Q value changes smoothly under continuous environment conditions;

[0027] Step S3: define three optimization goals, which are user comfort, energy saving and prediction accuracy, calculate the target rewards of the three optimization goals, and generate three types of target reward values;

[0028] User comfort refers to the difference between the current environment brightness and the user's preferred brightness;

[0029] Energy efficiency refers to the energy consumption of current lighting, with the goal of reducing unnecessary electricity consumption;

[0030] Prediction accuracy indicates the prediction error of future brightness requirements; it improves user experience while saving energy, and reduces invalid adjustments through accurate predictions;

[0031] Step S4: based on the three types of target reward values, update the environmental fuzzy Q value to obtain three types of fuzzy Q values;

[0032] Step S5: Based on the three types of fuzzy Q values, the hypervolume values ​​of different actions in the action space are calculated through the three-dimensional hypervolume, the performance of each action in the three optimization objectives is measured, and the hypervolume value is normalized; based on the normalized hypervolume value, action selection is performed through Softmax to optimize the lighting decision strategy;

[0033] Step S6: Set the maximum number of iterations, iterate steps S2 to S5, and generate the optimal lighting control strategy until the maximum number of iterations is reached.

[0034] Furthermore, step S4 specifically includes: based on the three types of target reward values, a dual-time difference method is used to calculate the Q-value error to simultaneously consider short-term and long-term returns; combined with the Q-value error, a historical Q-value backtracking term, an adaptive learning rate adjustment term and an adjustment coefficient are introduced to construct an environmental fuzzy Q-value method, and the environmental fuzzy Q-value is updated to obtain three types of fuzzy Q values.

[0035] By adopting the above scheme, the beneficial effects achieved by the present invention are as follows:

[0036] The present invention provides an intelligent lighting control system based on the Internet of Things. The system performs denoising processing on data collected by sensors through a Viterbi trend hidden Markov model, realizes high-precision environmental perception, effectively eliminates data noise caused by factors such as external light fluctuations and short-term personnel movement, and enables the lighting control system to more accurately identify the actual environmental state. Compared with the traditional lighting system that relies on fixed thresholds or simple filtering algorithms to process data, the denoising method of the present invention greatly improves data stability and reliability, avoids the problem of lighting misjudgment caused by noise interference, thereby ensuring the accuracy of lighting adjustment, improving user experience, and reducing the frequent switching or brightness fluctuation caused by light changes.

[0037] In addition, the present invention adopts a feature extraction method to extract environmental state features, user behavior features and energy consumption features from the denoised data, so that lighting control can better meet the needs of different scenarios; compared with the traditional intelligent lighting system that only relies on a single environmental variable (such as brightness) for adjustment, the present invention integrates multi-dimensional features to enable lighting strategies to more accurately match users' actual needs;

[0038] Finally, the present invention introduces fuzzy Q-learning, dual time difference optimization and multi-objective reward mechanism to achieve more intelligent lighting control decision-making; through fuzzy Q-learning, the system can still make accurate decisions when the environmental state is uncertain, avoiding the sensitivity of the traditional rule control mode to the fluctuation of environmental variables; dual time difference optimization enhances the system's ability to balance short-term and long-term energy consumption, making lighting control more stable; the multi-objective reward mechanism ensures that the system minimizes energy consumption while meeting user comfort, and improves the prediction accuracy of future lighting needs; finally, the system can automatically optimize the lighting strategy according to the real-time environmental status, avoid excessive or insufficient lighting, thereby improving energy utilization efficiency, reducing unnecessary electricity consumption, and improving the overall intelligence level. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a module schematic diagram of an intelligent lighting control system based on the Internet of Things proposed by the present invention;

[0040] Figure 2 This is a schematic diagram of the process of generating the optimal lighting control strategy proposed in Example 5. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0042] Embodiment 1, according to Figure 1 , the present invention provides an intelligent lighting control system based on the Internet of Things, the system includes a sensor acquisition module, a data denoising module, a feature extraction module, a control module, a terminal execution module, a user interaction module, a lighting device, a communication module 1 and a communication module 2;

[0043] The sensor acquisition module collects environmental data through ambient light intensity sensors, human presence sensors, temperature and humidity sensors, and current power consumption sensors;

[0044] Environmental data includes ambient light intensity - judging external lighting conditions, human presence - detecting whether there is someone in the room, temperature and humidity information - providing supplementary comfort information, current power consumption - monitoring current energy consumption, and user brightness preference - recording user habits based on historical data;

[0045] The data denoising module constructs a Viterbi trend hidden Markov model through environmental state modeling and trend weight enhancement, and denoises the environmental data through the Viterbi trend hidden Markov model to obtain denoised environmental data;

[0046] A feature extraction module extracts environmental state feature data, user behavior feature data, and energy consumption feature data from the denoised environmental data, and integrates them to obtain comprehensive environmental feature data;

[0047] The control module builds a multi-objective lighting reinforcement learning model through fuzzy Q learning, dual time difference optimization and multi-objective reward mechanism. The multi-objective lighting reinforcement learning model analyzes the comprehensive environmental feature data and generates the optimal lighting control strategy.

[0048] The terminal execution module generates control instructions and controls the lighting equipment according to the optimal lighting control strategy;

[0049] User interaction module provides a remote control portal, where users can view and control lighting equipment in real time through a mobile phone APP;

[0050] Communication module 1, realizes data interaction between the control module and the terminal execution module through the Internet of Things communication protocol;

[0051] Communication module 2 realizes data interaction between the terminal execution module and the lighting equipment through the Internet of Things communication protocol.

[0052] Embodiment 2: This embodiment is based on embodiment 1. In this embodiment, the data denoising module obtains a process of generating environmental data, which specifically includes the following contents:

[0053] Step B1: Define the double-layer hidden state and observation variables, take the environmental data as the observation variables, and build the environmental state DHMM model;

[0054] Step B2: Combined with the environmental state DHMM model, the optimal hidden state sequence is solved through the environmental trend-Viterbi algorithm to generate denoised environmental data.

[0055] Embodiment 3: This embodiment is based on embodiment 1. In this embodiment, the data denoising module obtains the process of generating environmental data, which specifically includes the following contents:

[0056] Step H1: Define the double-layer hidden state and observation variables, take the environmental data as the observation variables, and build the environmental state DHMM model;

[0057] Step H2: Combined with the environmental state DHMM model, the optimal hidden state sequence is solved by the Viterbi algorithm to generate denoised environmental data.

[0058] Embodiment 4: This embodiment is based on embodiment 2. In this embodiment, step B2 specifically includes the following contents:

[0059] Step B21: Introduce the environmental trend weight and recursively calculate the optimal path probability. The formula used is as follows:

[0060] ;

[0061] in, and represents the hidden state index, represents the time step index, Indicates time In state The optimal path probability is Indicates the selection of the optimal path at the previous moment; Indicates the status at the last moment The optimal path probability is Indicates status Transfer to state The probability of Indicates at time Observational data, Represents the observed value In Status The probability of represents the trend weight coefficient, represents the short-term window size, Short-term window index, Indicates a point in time in the past The observation data at the time, Indicates the past The observation probability accumulation at each moment;

[0062] Step B22: Backtrack and calculate the optimal hidden state sequence according to the optimal path probability;

[0063] Step B23: Calculate the denoised environment data based on the optimal hidden state sequence.

[0064] Embodiment 5, according to Figure 2 This embodiment is based on the fourth embodiment. In this embodiment, the process of the control module generating the optimal lighting control strategy specifically includes the following contents:

[0065] Step S1: Initialize state and action space: define state space and action space, initialize environmental fuzzy Q value and lighting decision strategy; map comprehensive environmental feature data to state space; action space includes: brightness adjustment, color temperature adjustment, mode switching and switch control;

[0066] The environmental fuzzy Q value is the Q value in the multi-objective lighting reinforcement learning model, which measures the long-term benefits of performing a lighting control action under different environmental states;

[0067] Lighting decision strategy refers to the strategy of controlling how the lighting system adjusts lighting parameters, dynamically optimizing lighting control to achieve energy saving, comfort and intelligent effects; including the adjustment of brightness, color temperature, switch status, dimming mode and energy saving mode;

[0068] Step S2: Calculating the environment fuzzy Q value: converting the comprehensive environment feature data into the environment fuzzy Q value through fuzzy Q learning, the conversion process includes fuzzifying the environment feature data, calculating the activation strength of the fuzzy rules, and calculating the environment fuzzy Q value;

[0069] Fuzzy environmental characteristic data: define fuzzy sets, map environmental state characteristic data, user behavior characteristic data and energy consumption characteristic data to corresponding fuzzy sets, and calculate the fuzzy membership;

[0070] Calculate the activation strength of fuzzy rules: define fuzzy rules, calculate the weights of different fuzzy rules based on fuzzy membership, and determine the impact of the current environment state on different rules;

[0071] Calculate the environment fuzzy Q value: Use fuzzy rules to weight the environment fuzzy Q value so that the environment fuzzy Q value changes smoothly under continuous environment conditions;

[0072] Step S3: Generate target rewards: define three optimization goals, which are user comfort, energy saving and prediction accuracy, calculate the target rewards of the three optimization goals, and generate three types of target reward values. The formula used is as follows:

[0073] ;

[0074] in, represents the weighted total reward, represents the user comfort reward value, represents the energy saving bonus value, represents the prediction accuracy reward value, , and represents the weight coefficient;

[0075] User comfort refers to the difference between the current environment brightness and the user's preferred brightness;

[0076] Energy efficiency refers to the energy consumption of current lighting, with the goal of reducing unnecessary electricity consumption;

[0077] Prediction accuracy indicates the prediction error of future brightness requirements; it improves user experience while saving energy, and reduces invalid adjustments through accurate predictions;

[0078] Step S4: Update the environment fuzzy Q value: based on the three types of target reward values, update the environment fuzzy Q value to obtain three types of fuzzy Q values;

[0079] Step S5: Optimization strategy: Based on the three types of fuzzy Q values, the hypervolume values ​​of different actions in the action space are calculated through the three-dimensional hypervolume, the performance of each action in the three optimization objectives is measured, and the hypervolume value is normalized; based on the normalized hypervolume value, action selection is performed through Softmax to optimize the lighting decision strategy. The formula used is as follows:

[0080] ;

[0081] in, Indicates action, Represents the time step The environmental status, represents the super volume value, represents the target index, Indicates The Q value of the optimization objective; Indicates the multiplication of the Q values ​​of all targets;

[0082] Step S6: Iteration loop: set the maximum number of iterations, iterate steps S2 to S5, and generate the optimal lighting control strategy until the maximum number of iterations is reached;

[0083] The maximum number of iterations is set to 2000.

[0084] Embodiment 6: This embodiment is based on embodiment 4. In this embodiment, the process of the control module generating the optimal lighting control strategy specifically includes the following contents:

[0085] Step T1: define the state space and action space, initialize the environmental fuzzy Q value and lighting decision strategy; map the comprehensive environmental feature data to the state space; the action space includes: brightness adjustment, color temperature adjustment, mode switching and switch control;

[0086] Step T2: converting the comprehensive environmental feature data into environmental fuzzy Q value through fuzzy Q learning, the conversion process includes fuzzifying the environmental feature data, calculating the activation strength of the fuzzy rules, and calculating the environmental fuzzy Q value;

[0087] Fuzzy environmental characteristic data: define fuzzy sets, map environmental state characteristic data, user behavior characteristic data and energy consumption characteristic data to corresponding fuzzy sets, and calculate the fuzzy membership;

[0088] Calculate the activation strength of fuzzy rules: define fuzzy rules, calculate the weights of different fuzzy rules based on fuzzy membership, and determine the impact of the current environment state on different rules;

[0089] Calculate the environment fuzzy Q value: Use fuzzy rules to weight the environment fuzzy Q value so that the environment fuzzy Q value changes smoothly under continuous environment conditions;

[0090] Step T3: define the optimization goal and generate the target reward value;

[0091] Step T4: Based on the target reward value, update the environmental fuzzy Q value and optimize the lighting decision strategy;

[0092] Step T5: Set the maximum number of iterations, iterate steps T2 to T4, and generate the optimal lighting control strategy until the maximum number of iterations is reached;

[0093] The maximum number of iterations is set to 2000.

[0094] Embodiment 7, this embodiment is based on embodiment 5. In this embodiment, step S4 specifically includes: based on the three types of target reward values, a dual-time difference method is used to calculate the Q value error to consider both short-term and long-term returns; combined with the Q value error, a historical Q value backtracking term, an adaptive learning rate adjustment term and an adjustment coefficient are introduced to construct an environmental fuzzy Q value method, and the environmental fuzzy Q value is updated to obtain three types of fuzzy Q values. The formula used is as follows:

[0095] Double-time difference method formula:

[0096] ;

[0097] in, represents the double TD error, represents the long-term error weight, represents the short-term error weight, represents the discount factor, Represents the time step The current action, Represents the time step The environmental status, Represents the time step The environmental status, Represents the time step The state-action Q value, Represents the time step The optimal Q value of Represents the time step The optimal Q value of represents one-step TD error, represents the two-step TD error;

[0098] Update the ambient blur Q value formula:

[0099] ;

[0100] in, represents the adaptive learning rate, represents the adjustment coefficient, represents the backtracking step length, Backtracking index, represents the backtracking weight, represents the historical Q value, Represents the historical Q value backtracking item.

[0101] The present invention and its embodiments are described above, which is not restrictive. What is shown in the accompanying drawings is only one of the embodiments of the present invention, and the actual structure is not limited to this. In short, if ordinary technicians in this field are inspired by it and do not deviate from the purpose of the invention, they can creatively design structural methods and embodiments similar to the technical solution, which should all fall within the scope of protection of the present invention.

Claims

1. An intelligent lighting control system based on the Internet of Things, comprising a sensor acquisition module and lighting equipment, wherein the sensor acquisition module collects environmental data; characterized in that: The system also includes a data denoising module, a feature extraction module, a control module and a terminal execution module; The data denoising module constructs a Viterbi trend hidden Markov model through environmental state modeling and trend weight enhancement, and denoises the environmental data through the Viterbi trend hidden Markov model to obtain denoised environmental data; A feature extraction module extracts environmental state feature data, user behavior feature data, and energy consumption feature data from the denoised environmental data, and integrates them to obtain comprehensive environmental feature data; The control module builds a multi-objective lighting reinforcement learning model through fuzzy Q learning, dual time difference optimization and multi-objective reward mechanism. The multi-objective lighting reinforcement learning model analyzes the comprehensive environmental feature data and generates the optimal lighting control strategy. The terminal execution module controls the lighting equipment according to the optimal lighting control strategy.

2. According to claim 1, the intelligent lighting control system based on the Internet of Things is characterized by: The data denoising module generates denoised environmental data, which specifically includes the following: Step B1: Construct the environmental state DHMM model based on environmental data; Step B2: Combined with the environmental state DHMM model, the optimal hidden state sequence is solved through the environmental trend-Viterbi algorithm to generate denoised environmental data.

3. The intelligent lighting control system based on the Internet of Things according to claim 2 is characterized in that: Step B2 specifically includes the following contents: Step B21: Introduce environmental trend weights and recursively calculate the probability of the optimal path; Step B22: Backtrack and calculate the optimal hidden state sequence according to the optimal path probability; Step B23: Calculate the denoised environment data based on the optimal hidden state sequence.

4. The intelligent lighting control system based on the Internet of Things according to claim 1, characterized in that: The control module generates the optimal lighting control strategy, which specifically includes the following: Step S1: define the state space and action space, initialize the environmental fuzzy Q value and lighting decision strategy; map the comprehensive environmental feature data to the state space; Step S2: converting the comprehensive environmental feature data into environmental fuzzy Q value through fuzzy Q learning; Step S3: define three optimization objectives, calculate the target rewards of the three optimization objectives, and generate three types of target reward values; Step S4: based on the three types of target reward values, update the environmental fuzzy Q value to obtain three types of fuzzy Q values; Step S5: Based on the three types of fuzzy Q values, the hypervolume values ​​of different actions in the action space are calculated through the three-dimensional hypervolume, the hypervolume values ​​are normalized, and the action is selected through Softmax to optimize the lighting decision strategy; Step S6: Iterate step S2 to step S5 to generate an optimal lighting control strategy.

5. The intelligent lighting control system based on the Internet of Things according to claim 4 is characterized in that: In step S3, the three optimization objectives are user comfort, energy saving and prediction accuracy.

6. The intelligent lighting control system based on the Internet of Things according to claim 4 is characterized in that: Step S4 specifically includes: based on the three types of target reward values, using the double-time difference method to calculate the Q value error; combining the Q value error, introducing the historical Q value backtracking term, the adaptive learning rate adjustment term and the adjustment coefficient to construct the environmental fuzzy Q value method, updating the environmental fuzzy Q value, and obtaining three types of fuzzy Q values.

7. The intelligent lighting control system based on the Internet of Things according to claim 4 is characterized in that: The action space in step S1 includes: brightness adjustment, color temperature adjustment, mode switching and switch control.

Images