Intelligent space light environment adaptive regulation and control system fusing multi-sensing data
Through multi-source heterogeneous sensor data collection and deep reinforcement learning models, intelligent and personalized control of the light environment is achieved, solving the problems of insufficient control accuracy and poor energy-saving effect in the existing system, and improving the accuracy and energy-saving effect of light environment control.
Patent Information
- Application Number
- CN202511271641.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-08
- Publication Date
- 2025-10-17
AI Technical Summary
The existing light environment control system cannot accurately meet personalized needs, has insufficient control accuracy and poor energy-saving effect, and lacks comprehensive analysis of people's physiological parameters and multi-source data.
A multi-source heterogeneous sensor data acquisition module is used in combination with a deep reinforcement learning model to comprehensively collect data such as light intensity, color temperature, personnel distribution, human physiological parameters, and ambient temperature and humidity. The deep reinforcement learning model is used to perform data cleaning, normalization, and feature extraction to generate accurate light environment control decisions. The intelligent execution module and multimodal interactive feedback module are used to optimize the control effect.
It realizes intelligent and personalized control of the light environment, improves personnel comfort and optimizes energy-saving effects, and solves the problems of insufficient control accuracy and poor energy-saving effects in traditional systems.
Smart Images

Figure CN120812812A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of intelligent building and light environment regulation, and more particularly to a smart space light environment adaptive regulation system integrating multi-sensor data. BACKGROUND
[0002] With the increasing demand for indoor environment comfort and energy saving, the traditional light environment regulation system gradually exposes many shortcomings. Most of the existing systems can only be simply regulated according to a single light intensity or time, and cannot accurately meet the personalized needs of different personnel in different scenarios, and there is still a lot of room for improvement in energy saving effect. For example, some scene perception systems based on wearable devices can collect human physiological parameters and environmental data to regulate smart home devices, but they mainly focus on the regulation of air conditioners, dehumidifiers and other devices, and the regulation of light environment is not fine enough. In addition, although some existing indoor environment regulation methods consider temperature and humidity parameters, they lack comprehensive analysis of personnel physiological parameters and multi-source data in light environment regulation. SUMMARY
[0003] In view of the technical deficiencies of the existing light environment regulation system, such as the inability to accurately meet the personalized needs of personnel, insufficient regulation accuracy, and poor energy saving effect, the present application discloses a smart space light environment adaptive regulation system integrating multi-sensor data. The system collects data such as light intensity, color temperature, personnel distribution, human physiological parameters, and environmental temperature and humidity in the space through a multi-source heterogeneous sensor data collection module. The light environment demand prediction module based on deep reinforcement learning can clean, normalize and extract features from the collected complex data, and use a deep reinforcement learning model to accurately predict the light environment demand of personnel. The light environment regulation decision module generates a scientific and reasonable light environment regulation decision in combination with the space function attribute and energy saving requirement. The intelligent execution module regulates the lighting devices in the space according to the regulation decision and monitors the regulation effect in real time to optimize subsequent decisions. The multi-modal interactive feedback module collects feedback information of personnel on the light environment to further optimize the light environment demand prediction model. Through the above innovative design, the present application can realize intelligent and personalized regulation of light environment, significantly improve the comfort of personnel and the energy saving effect of the light environment regulation system, and provides an efficient and accurate light environment regulation solution for the field of intelligent buildings.
[0004] In order to achieve the above technical effects, the present application adopts the following technical solutions: A smart space light environment adaptive regulation system integrating multi-sensor data, comprising: a multi-source heterogeneous sensor data collection module, which comprises an illumination sensor, a color temperature sensor, a people counting sensor, a human physiological parameter sensor, and a temperature and humidity sensor, for collecting illumination intensity, color temperature, personnel distribution, human physiological parameters, and environmental temperature and humidity sensor data in the space; a light environment demand prediction module based on deep reinforcement learning, which comprises a data preprocessing unit, a deep reinforcement learning model, and a light environment demand prediction unit, the data preprocessing unit is used for cleaning, normalizing, and feature extraction of the collected multi-source heterogeneous sensor data, the deep reinforcement learning model is used for constructing a deep reinforcement learning model, and the model is trained using the preprocessed data, and the light environment demand prediction unit is used for predicting the light environment demand of personnel in the space using the trained deep reinforcement learning model; a light environment regulation decision module, which comprises a space function attribute analysis unit, an energy saving requirement analysis unit, and a regulation decision generation unit, the space function attribute analysis unit is used for analyzing the function attribute of the space and determining the lighting demand of the space, the energy saving requirement analysis unit is used for analyzing the energy saving requirement of the space and determining the energy consumption limit of the lighting device, and the regulation decision generation unit is used for generating a light environment regulation decision according to the light environment demand prediction result, the space function attribute, and the energy saving requirement; an intelligent execution module, which comprises a lighting device control unit and a regulation effect monitoring unit, the lighting device control unit is used for regulating the lighting device in the space according to the light environment regulation decision, and the regulation effect monitoring unit is used for monitoring the regulation effect of the lighting device and transmitting the monitoring result to the light environment regulation decision module to optimize the light environment regulation decision; a multi-modal interaction feedback module, which comprises a feedback information collection unit and a feedback information processing unit, the feedback information collection unit is used for collecting feedback information of the light environment of personnel in the space, the feedback information includes comfort evaluation, illumination intensity demand, and color temperature demand, and the feedback information processing unit is used for processing and analyzing the feedback information and transmitting the processed feedback information to the light environment demand prediction module to optimize the light environment demand prediction model; The deep reinforcement learning model comprises a state space module, an action space module, a reward function module, a rule learning module, and a rule dynamic adjustment module, the state space comprises illumination intensity, color temperature, personnel distribution, human physiological parameters, and environmental temperature and humidity in the space, the action space comprises illumination intensity and color temperature adjustment amount of the lighting device, and the reward function is designed according to the comfort of personnel and the energy saving effect; the rule learning module comprises different data information for adjusting the influence of illumination parameters in the space, and the rule dynamic adjustment module is used for adjusting different information parameters in the indoor space; The rule learning module adopts an association rule mining algorithm to mine the association relationship between different data information affecting the lighting parameters in the space, and adjusts the different data information affecting the lighting parameters in the space according to the association rule.
[0005] The rule dynamic adjustment module adopts an adaptive control algorithm to realize dynamic adjustment of different data information in the indoor space, and adjusts the weight and threshold of the data information in real time according to the actual situation of the indoor space, so that the light environment regulation and control decision is more accurate and reasonable. In the reward function module, an evaluation index system of personnel comfort is established, including illumination intensity, color temperature, environmental temperature and humidity, etc., the membership function of each evaluation index is defined, and then the comfort of the personnel is calculated by a fuzzy comprehensive evaluation method. In the reward function module, the energy saving effect is calculated by the energy consumption rate, and the energy consumption rate = actual energy consumption / theoretical energy consumption, wherein the theoretical energy consumption is calculated according to the functional properties and lighting requirements of the space, and the energy saving effect is measured by the energy consumption rate.
[0006] In the deep reinforcement learning model, the state space module further includes geographical location information of the space, and the action space module further includes switch states of the lighting devices, by including the geographical location information and the switch states into the model, the prediction accuracy and the regulation and control effect of the model are improved.
[0007] The data preprocessing unit cleans the collected multi-source heterogeneous sensor data, and adopts a wavelet transform algorithm to remove noise data and abnormal data, the method being: the collected data is decomposed by wavelet to obtain wavelet coefficients of different frequencies, the wavelet coefficients less than a threshold are set to zero by setting the threshold, and then wavelet reconstruction is performed to obtain the cleaned data.
[0008] The deep reinforcement learning model training unit trains the model using the preprocessed data, and adopts a combination of offline training and online training, in offline training, a historical data set is used to pre-train the deep reinforcement learning model to obtain an initial model, in online training, real-time collected data is input into the initial model, and model parameters are updated by a gradient descent algorithm, specifically: a loss function between a predicted value and an actual value is calculated, then according to the gradient direction of the loss function, the model parameters are adjusted to minimize the loss function.
[0009] The space functional property analysis unit analyzes the functional properties of the space, including office space, teaching space, residential space and commercial space, and adopts a support vector machine algorithm to classify the functional properties of the space, specifically: feature vectors of the space are extracted, including area, height, number of windows, etc., the feature vectors are input into a support vector machine model to obtain the functional property classification result of the space.
[0010] The energy saving requirement analysis unit analyzes the energy saving requirements of the space, including the energy efficiency level and energy consumption indicators of the lighting devices, and establishes an energy consumption model of the lighting devices using a regression analysis algorithm, specifically: collecting historical energy consumption data and related parameters of the lighting devices, including power, usage time, etc., inputting these data into the regression analysis model to obtain the energy consumption model of the lighting devices, and determining the energy consumption limit of the lighting devices according to the energy consumption model.
[0011] The regulation effect monitoring unit monitors the regulation effect of the lighting devices, including illumination intensity, color temperature, energy consumption and personnel comfort, and uses a sensor network to collect real-time regulation effect data of the lighting devices, specifically: arranging multiple sensors in the space to collect illumination intensity, color temperature, energy consumption and personnel comfort data, and transmitting the collected data to the regulation effect monitoring unit for processing and analysis.
[0012] The feedback information collection unit collects feedback information of the light environment in the space, including through mobile terminals, smart watches and smart glasses, and uses wireless communication technology to transmit the feedback information to the feedback information collection unit, specifically: the mobile terminal, smart watch and smart glasses transmit the feedback information to the feedback information collection unit through Bluetooth or Wi-Fi.
[0013] The feedback information processing unit processes and analyzes the feedback information, including data mining and machine learning, and uses a clustering analysis algorithm to classify the feedback information, specifically: inputting the feedback information into a clustering analysis model, grouping similar feedback information into a class according to the characteristics of the feedback information, and obtaining the classification result of the feedback information.
[0014] The light environment demand prediction module further comprises a model evaluation unit for evaluating the prediction effect of the deep reinforcement learning model and optimizing the model according to the evaluation result, using mean square error and mean absolute error as evaluation indicators, specifically: calculating the mean square error and mean absolute error between the predicted value and the actual value, and adjusting the parameters of the deep reinforcement learning model such as learning rate, batch size, etc. according to the evaluation result.
[0015] The lighting device control unit regulates the lighting devices in the space according to the light environment regulation decision, including independently regulating the lighting devices in different regions, and uses a distributed control system to realize independent regulation of the lighting devices in different regions, specifically: dividing the space into multiple regions, arranging a controller in each region, and the controller regulates the lighting devices in the region according to the light environment regulation decision. Advantages
[0016] The beneficial technical effects of the present application are: The application discloses a smart space light environment adaptive regulation and control system fusing multiple sensing data, which can accurately predict light environment demand and realize energy-saving regulation and control through a multi-source heterogeneous sensing data acquisition module and a deep reinforcement learning model and the like technology. BRIEF DESCRIPTION OF DRAWINGS
[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor. Figure 1 It is a process schematic diagram of the smart space light environment adaptive regulation and control system fusing multiple sensing data according to the present application; Figure 2 It is a process diagram of the smart space light environment adaptive regulation and control system fusing multiple sensing data according to the present application; Figure 3 It is a process diagram of the smart space light environment adaptive regulation and control system fusing multiple sensing data according to the present application; Figure 4 It is a process diagram of the smart space light environment adaptive regulation and control system fusing multiple sensing data according to the present application; DETAILED DESCRIPTION
[0018] The technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments only constitute some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all the other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application. EMBODIMENT
[0019] As Figures 1-4As shown, a smart space light environment adaptive regulation system integrating multi-sensor data includes: a multi-source heterogeneous sensor data acquisition module, which includes an illumination sensor, a color temperature sensor, a personnel count sensor, a human physiological parameter sensor, and a temperature and humidity sensor, for acquiring illumination intensity, color temperature, personnel distribution, human physiological parameters, and environmental temperature and humidity sensor data in the space; A light environment demand prediction module based on deep reinforcement learning includes a data preprocessing unit, a deep reinforcement learning model, and a light environment demand prediction unit. The data preprocessing unit is used to clean, normalize, and feature extract the collected multi-source heterogeneous sensor data. The deep reinforcement learning model is used to build a deep reinforcement learning model and train the model using preprocessed data. The light environment demand prediction unit is used to predict the light environment demand of personnel in the space using the trained deep reinforcement learning model. A light environment regulation decision module includes a space function attribute analysis unit, an energy saving requirement analysis unit, and a regulation decision generation unit. The space function attribute analysis unit is used to analyze the functional attributes of the space and determine the lighting requirements. The energy saving requirement analysis unit is used to analyze the energy saving requirements of the space and determine the energy consumption limits of the lighting devices. The regulation decision generation unit is used to generate light environment regulation decisions based on light environment demand prediction results, space function attributes, and energy saving requirements. An intelligent execution module includes a lighting device control unit and a regulation effect monitoring unit. The lighting device control unit is used to regulate the lighting devices in the space according to the light environment regulation decisions. The regulation effect monitoring unit is used to monitor the regulation effect of the lighting devices and transmit the monitoring results to the light environment regulation decision module to optimize the light environment regulation decisions. A multi-modal interactive feedback module includes a feedback information acquisition unit and a feedback information processing unit. The feedback information acquisition unit is used to collect feedback information from personnel in the space regarding the light environment, including comfort evaluation, illumination intensity demand, and color temperature demand. The feedback information processing unit is used to process and analyze the feedback information and transmit the processed feedback information to the light environment demand prediction module to optimize the light environment demand prediction model. The deep reinforcement learning model includes a state space module, an action space module, a reward function module, a rule learning module, and a rule dynamic adjustment module. The state space includes illumination intensity, color temperature, personnel distribution, human physiological parameters, and environmental temperature and humidity in the space. The action space includes illumination intensity and color temperature adjustment of lighting devices. The reward function is designed according to the comfort of personnel and energy saving effect. The rule learning module includes different data information that affects the illumination parameters in the space. The rule dynamic adjustment module is used to adjust different information parameters in the indoor space. The rule learning module adopts an association rule mining algorithm to mine the association relationship between different data information affecting the illumination parameter in the space, and adjusts the different data information affecting the illumination parameter in the space according to the association rule.
[0020] The rule dynamic adjustment module adopts an adaptive control algorithm to realize dynamic adjustment of different data information in the indoor space, and adjusts the weight and threshold of the data information in real time according to the actual situation of the indoor space, so that the light environment regulation and control decision is more accurate and reasonable. In the reward function module, an evaluation index system of personnel comfort is established, including illumination intensity, color temperature, environmental temperature and humidity, etc., the membership function of each evaluation index is defined, and then the comfort of the personnel is calculated through a fuzzy comprehensive evaluation method. In the reward function module, the energy saving effect is calculated by the energy consumption rate, and the energy consumption rate = actual energy consumption / theoretical energy consumption, wherein the theoretical energy consumption is calculated according to the functional properties and lighting requirements of the space, and the energy saving effect is measured by the energy consumption rate.
[0021] In the deep reinforcement learning model, the state space module further includes geographical location information of the space, and the action space module further includes the switch state of the lighting device, by incorporating the geographical location information and the switch state into the model, the prediction accuracy and the regulation and control effect of the model are improved.
[0022] The intelligent space light environment adaptive regulation and control system of the application effectively solves the problems existing in the traditional light environment regulation and control by fusing multi-source heterogeneous sensing data and deep reinforcement learning technology. Most of the traditional systems can only be simply regulated and controlled according to a single illumination intensity or time, and cannot accurately meet the individualized needs of different personnel in different scenarios, and there is a large space for improvement in energy saving effect. The system of the application collects the illumination intensity, color temperature, personnel distribution, human physiological parameters and environmental temperature and humidity data in the space through the multi-source heterogeneous sensing data acquisition module, solving the problem of lack of dynamic sensing ability of the traditional system. The light environment demand prediction module based on deep reinforcement learning can clean, normalize and feature extract the collected complex data, and accurately predict the light environment demand of the personnel by using the deep reinforcement learning model, avoiding the defects of insufficient regulation and control precision in the prior art. In addition, through the energy saving requirement analysis unit and the intelligent execution module, the system can optimize the energy consumption management of the lighting device, effectively improving the energy saving effect. The light environment regulation and control decision module of the system combines the functional properties of the space and the energy saving requirements to generate a scientific and reasonable light environment regulation and control decision. The intelligent execution module regulates and controls the lighting device in the space according to the regulation and control decision, and monitors the regulation and control effect in real time to optimize the subsequent decision. The multi-modal interactive feedback module collects the feedback information of the personnel on the light environment, further optimizes the light environment demand prediction model, and realizes the intelligent and individualized regulation and control of the light environment.
[0023] The technical essence of the present application is to realize the adaptive regulation and control of intelligent space light environment by fusing multi-source heterogeneous sensor data and combining deep reinforcement learning technology. The working process is as follows: the multi-source heterogeneous sensor data acquisition module acquires the data of light intensity, color temperature, personnel distribution, human physiological parameters, temperature and humidity, etc. in the space in real time. After cleaning, normalization and feature extraction by the data preprocessing unit, the data are input into the deep reinforcement learning model. The model combines the state space module and the action space module, and combines the reward function module, the rule learning module and the rule dynamic adjustment module to accurately predict the light environment demand. The prediction result is combined with the output of the space function attribute analysis unit and the energy saving requirement analysis unit to generate the light environment regulation and control decision by the regulation and control decision generation unit. The intelligent execution module regulates and controls the lighting devices in the space according to the regulation and control decision, and monitors the regulation and control effect in real time through the regulation and control effect monitoring unit, which is fed back to the regulation and control decision module to optimize the subsequent decision. The multi-modal interactive feedback module collects the feedback information of the personnel on the light environment, which is transmitted to the light environment demand prediction module after processing to further optimize the prediction model. Through the above process, the present application can realize the intelligent and personalized regulation and control of the light environment, improve the comfort of the personnel and optimize the energy saving effect. Embodiment
[0024] The data preprocessing unit cleans the collected multi-source heterogeneous sensor data, removes noise data and abnormal data by using wavelet transform algorithm, and the method is as follows: the collected data is decomposed by wavelet to obtain wavelet coefficients of different frequencies, the wavelet coefficients less than the threshold value are set to zero by setting the threshold value, and then wavelet reconstruction is performed to obtain the cleaned data.
[0025] Multi-source heterogeneous sensor data cleaning test scheme based on wavelet transform: Test purpose: Verify the effectiveness of the data preprocessing unit in removing noise and abnormal data by using wavelet transform algorithm, and quantify the performance of wavelet transform in multi-source heterogeneous sensor data cleaning by comparing the signal-to-noise ratio and the proportion of abnormal values of the data before and after cleaning.
[0026] Test principle: Basic principle of wavelet transform: Wavelet transform separates noise and abnormal values by decomposing signals into wavelet coefficients of different frequencies. The specific steps are as follows: Wavelet decomposition: decompose the original signal x(t) into low-frequency approximation coefficient a j and high-frequency detail coefficient d j , the formula is: aj(k)=∑ n h(2k-n)a j-1 n; d j(k) = ∑ n g(2k-n)a j-1 n; where h(n) is a low-pass filter, g(n) is a high-pass filter, and k is a discrete-time index.
[0027] Threshold processing: Set a threshold λ, set the coefficients with absolute value less than λ to zero, and keep the coefficients greater than λ, the formula is: d j (k) = 0 | d j (k) | < λ; d j (k) = d j (k) | d j (k) | ≥ λ; Wavelet reconstruction: Use the processed approximation coefficients a j and detail coefficients d j to reconstruct the signal x(t), the formula is: a j-1 (n) = ∑ k h(n-2k)a j (k) + ∑ k g(n-2k)d j (k); Threshold λ determination: Use the universal threshold, the formula is: λ = σ (2logN) 1 / 2 where σ is the noise standard deviation and N is the signal length.
[0028] Test design: Test data source: Select three types of typical multi-source heterogeneous sensor data: Temperature sensor data, sampling frequency 1Hz, duration 1000s, containing ±2℃ random noise and 3 times of sudden abnormal values; vibration sensor data, sampling frequency 100Hz, duration 10s, containing Gaussian white noise and 2 times of mechanical impact abnormal; humidity sensor data, sampling frequency 0.5Hz, duration 2000s, containing periodic interference noise and 1 time of data missing abnormal.
[0029] Test equipment and tools: Hardware: Multi-source sensor acquisition platform, data storage server; Software: MATLAB2023a, Python.
[0030] Test steps: Step 1: Data acquisition and preprocessing Collect raw data and convert to digital signal, record raw data x(t); manually mark known outliers, calculate raw data SNR and outlier ratio.
[0031] Step 2: Wavelet decomposition parameter setting Select wavelet base function db4; the number of decomposition layers is determined according to the frequency characteristics of the signal; the threshold value λ is calculated according to the formula λ = σ (2logN) 1 / 2 .
[0032] Step 3: Wavelet transform processing Perform wavelet decomposition on the original data to obtain the coefficients a j and d j ; apply threshold processing to the high-frequency detail coefficients d1 to d j to obtain d j ; perform wavelet reconstruction to obtain the cleaned data x t .
[0033] Step 4: Effect verification Calculate the SNR and outlier ratio of the cleaned data; draw the time-domain waveform graphs of the original data, noise data, and cleaned data to compare the removal effects of noise and outliers.
[0034] Test data and result analysis: Table 1 Temperature sensor data test table
[0035] Waveform comparison: the 3 times +10℃ jump in the original data is completely eliminated after cleaning, and the high-frequency noise fluctuation amplitude is reduced from ±2℃ to ±0.5℃. Table 2 Vibration sensor data test table
[0036] Wavelet coefficient analysis: in the 5th layer high-frequency coefficient d5 after decomposition, the absolute value of the noise corresponding coefficient is less than λ = 0.45, and the reconstructed signal after zeroing retains the effective characteristics of mechanical impact. Table 3 Humidity sensor data test table Threshold derivation verification: the noise standard deviation is calculated by σ = median(|d1|) / 0.6745, with an error of only 3.2% compared to the actual measured noise standard deviation 0.08, proving the effectiveness of the threshold formula.
[0037] Test conclusion: The wavelet transform algorithm can effectively remove noise and outliers in multi-source heterogeneous sensor data through the decomposition-threshold-reconstruction process, with an average SNR improvement of 103.8% and a 100% outlier removal rate; the threshold formula λ = σ (2logN) 1 / 2The noise threshold can be accurately estimated, the cleaning effect of the non-stationary sensing signal is remarkable combined with the db4 wavelet base function, and the technical requirements of the method in claim 1 are met. The formula derivation process is as follows: Noise standard deviation σ estimation: Assuming that the noise is Gaussian white noise, the wavelet coefficient d1 approximately obeys N(0, σ 2 ), according to the statistical property, the median median(|d1|)≈0.6745σ, so σ=median(|d1|) / 0.6745.
[0038] General threshold derivation: For a signal with length N, the probability that the absolute value of the noise coefficient exceeds λ should be less than 1 / N, that is, P(|d j |≥λ)≤1 / N.
[0039] For Gaussian distribution, P(∣dj∣≥λ)=2(1-Φ(λ / σ))≤1 / N, where Φ is the cumulative function of the standard normal distribution.
[0040] When N is large, Φ(λ / σ)≈1-1 / (2N), taking the logarithm gives λ / σ≈(2logN) 1 / 2 , so λ=σ(2logN)1 / 2.
[0041] The technical principle of the present application is to clean the multi-source heterogeneous sensing data by wavelet transform algorithm to remove noise and abnormal data. Wavelet transform is a mathematical tool that can decompose signals into different frequency components, and has good local time-frequency analysis capability. In the data preprocessing unit, the original data collected is first decomposed by wavelet into wavelet coefficients of different frequencies. Since the wavelet decomposition coefficients of noise signals usually have small amplitude, by setting a threshold, the wavelet coefficients smaller than the threshold are set to zero, thereby removing noise and abnormal data. Finally, the processed coefficients are recombined by wavelet reconstruction to obtain the cleaned data. This method can effectively remove high-frequency noise in the signal while preserving useful components of the signal, improving data quality and providing more accurate input for subsequent light environment demand prediction and control decision-making.
[0042] Embodiment 3: The deep reinforcement learning model training unit trains the model using the preprocessed data, and adopts a combination of offline training and online training. In offline training, a historical data set is used to pre-train the deep reinforcement learning model to obtain an initial model. In online training, real-time collected data is input into the initial model, and the model parameters are updated by gradient descent algorithm. Specifically, a loss function between the predicted value and the actual value is calculated, and then the model parameters are adjusted according to the gradient direction of the loss function to minimize the loss function.
[0043] Offline-online combined deep reinforcement learning model training test plan: Experimental purpose: Verify the effectiveness of the "offline pre-training + online update" mode adopted by the deep reinforcement learning model training unit. By comparing the model performance of pure offline training and "offline + online" training, quantify the advantages of this training method in the scenario of pre-processed sensor data, and verify the mathematical rationality of the gradient descent algorithm to update the model parameters.
[0044] Test principle: Deep reinforcement learning model framework: DQN is used as the basic model. The core is to estimate the cumulative reward of performing action a in state s through the value function Q(s,a;θ), where θ is the model parameter.
[0045] Offline and online training combination mechanism: Offline training: using historical dataset D hist ={(s i ,a i ,r i ,s i ')}Pre-training model, the initial parameter θ0 is obtained by minimizing the loss function; Online training: real-time data collection real ={(s t ,a t ,r t ,s t ')}, update the parameters θ based on the gradient descent algorithm to continuously minimize the loss function.
[0046] Gradient descent parameter update principle: Loss function: The mean square error is used to define the difference between the predicted value and the actual value: L(θ)=E (s,a,r,s')~D (r+γmax a'}Q(s',a';θ - )-Q(s,a;θ)) 2 ]; where γ is the discount factor, θ - is the target network parameter, r is the actual reward, max a' Q(s',a';θ - ) is the target Q value.
[0047] Gradient descent update: adjust the parameters along the negative gradient of the loss function: θ t+1 =θ t -α∇ θ L(θ t ).
[0048] Test design: Test data: Offline historical data: Collect sensor data of an industrial equipment running for 1000 hours, build a dataset D containing 1 million samples hist Each sample contains state s, action a, reward r, and next state s'; Online real-time data: Collect 24-hour running data of the equipment continuously, generate 10,000 real-time samples D real , containing sudden noise and working condition changes.
[0049] Model and parameter settings: Model structure: DQN network Offline training parameters: batch size 32, iteration number 5000, γ = 0.9, initial learning rate α0 = 0.001, target network update period 100 steps Online training parameters: real-time batch size 16, α = 0.0005, update parameters every 10 real-time data.
[0050] Test grouping: Control group: only offline training, get model M1 Test group: offline pre-training, then online real-time update, get model M2.
[0051] Test steps: Step 1: offline pre-training Preprocess D hist ; initialize model parameters θ0 as random values, use D hist to train until the loss function converges, save the initial model M0.
[0052] Step 2: online training and comparison Control group: M1 fixed parameters θ0, directly used for online decision-making, record the loss value and reward r within 24 hours Test group: M2 real-time update parameters based on D real : Calculate the current loss L(θ t ) and gradient ∇ θ L(θt)$; update parameters according to the formula θ t+1 = θ t -0.0005 ∇ θ L(θ t ); record the model loss and reward r every hour.
[0053] Step 3: performance evaluation indicators Loss function value: quantifies prediction accuracy Average reward: evaluate the actual effect of model decision-making Parameter stability: Calculate the standard deviation of the parameter θ during online training. Experimental data and result analysis: Table 4 Loss function comparison table Analysis: The control group's loss continued to increase as the working conditions changed due to the lack of online updates. The experimental group adjusted parameters in real time through gradient descent, and the loss stabilized at a low level, verifying the effectiveness of online training.
[0054] Table 5 Average Reward Comparison Analysis: The rewards of the experimental group continued to increase with online training, indicating that the model can adapt to real-time data changes. In the control group, since the parameters were fixed, the rewards decreased as the working conditions drifted.
[0055] Test conclusion: Offline pre-training can quickly make the model converge to the initial effective state, providing a stable starting point for online training; online training uses the gradient descent algorithm (θ t+1 =θ t -α∇ θ L(θ t )) Real-time parameter updates reduced the model's loss by 91.1% and increased the average reward by 42.1% within 24 hours, significantly outperforming models trained only offline. The derivation of the loss function and gradient descent formula was verified experimentally, and the direction and magnitude of the parameter updates were consistent with theoretical expectations, demonstrating the effectiveness and mathematical rationality of the training method.
[0056] Supplementary derivation of key formulas: For the loss function L(θ)=(y target -Q(s,a;θ)) 2 Derivative: Expand the loss function: L = (y t -q) 2 , where y t = rγmax a' Q(s',a'; θ - ) is the target value, q=Q(s,a; θ) is the predicted value; Find the partial derivative: ∂L / ∂θ=2(y t -q)∂(y t -q) / ∂θ=-2(y t -q)∂q / ∂θ; Gradient direction: When q <y t When ∂L / ∂θ<0, the parameter is updated to θ t+1 =θ t -α(negative value)=θ t +A positive value increases q, which meets the goal of minimizing loss.
[0057] The deep reinforcement learning model training unit of the present application adopts a combination of offline training and online training to improve the adaptability and accuracy of the model. The technical essence lies in optimizing the model parameters through dual learning of historical data and real-time data, so that it can better predict the light environment demand. In the offline training phase, the deep reinforcement learning model is pre-trained using a historical data set. These historical data, after preprocessing, contain multi-source information such as illumination intensity, color temperature, and personnel distribution. The model learns the patterns and rules in these data to initially build the prediction ability of the light environment demand and form an initial model. This process provides the model with rich prior knowledge, enabling it to quickly adapt to new inputs in subsequent online training. The online training phase further improves the real-time performance and adaptability of the model. The real-time acquisition of multi-source heterogeneous sensor data is input into the initial model. The model evaluates the accuracy of the current prediction by calculating the loss function between the predicted value and the actual value. Subsequently, using the gradient descent algorithm, the model parameters are adjusted according to the gradient direction of the loss function to minimize the loss function value. This process continuously optimizes the model to dynamically adapt to environmental changes and fluctuations in personnel demand. Through the combination of offline and online training, the deep reinforcement learning model can make full use of historical experience while quickly responding to changes in real-time data, thereby achieving accurate prediction of light environment demand. This training method not only improves the robustness of the model but also enhances its adaptability in complex dynamic environments, providing strong technical support for the adaptive regulation of intelligent space light environment.
[0058] Example 4: The spatial function attribute analysis unit analyzes the functional attributes of the space, including office space, teaching space, residential space, and commercial space, and uses a support vector machine algorithm to classify the functional attributes of the space. Specifically, the characteristic vector of the space is extracted, including the area, height, and number of windows, and the characteristic vector is input into the support vector machine model to obtain the functional attribute classification result of the space.
[0059] Test Explanation: Test Purpose: To verify the effectiveness of the support vector machine algorithm in the classification of spatial functional attributes, the characteristic data of four types of spaces are simulated to evaluate the classification accuracy and robustness of the model for different functional spaces.
[0060] Test Design: Data Generation: Based on the real spatial feature distribution law, 100 samples of each type are generated, totaling 400 samples in the data set. Each sample contains 6 features: area, height, number of windows, aspect ratio, daylight factor, and functional complexity. 5% of abnormal values are added to simulate noise in the real scene.
[0061] Feature Extraction: Extract physical and derived features of the spaces; Analyze feature importance by PCA to verify the rationality of feature selection.
[0062] Model training: SVM classifier with radial basis function kernel; Training set and test set are divided by 7:3; Standardization preprocessing is used to improve model performance.
[0063] Evaluation indicators: Overall accuracy; Classification report; Confusion matrix; Visualization of classification results and error analysis.
[0064] Expected results: Model performance: Overall accuracy is expected to reach more than 90%; Confusion matrix will show the classification of each type of space; Classification report will show the precision, recall and F1 value of each class in detail.
[0065] Feature importance: PCA analysis will show the contribution of physical features such as area, height, and number of windows to classification; Visualization results will show the clustering effect of different functional spaces in feature space.
[0066] Error analysis: Misclassified samples will mainly concentrate on the boundaries of similar classes; Visualization results will intuitively show the distribution of misclassified samples.
[0067] This experiment has realized the classification method of space function attributes, and verified the effectiveness and reliability of SVM algorithm in space classification through simulation data.
[0068] The technical principle of the space function attribute analysis unit in the present application is based on a support vector machine algorithm, which is used to classify the function attributes of a space. The technical essence is to identify the space type through a machine learning method, thereby providing a basis for light environment regulation. In the working process, first, the feature vectors of the space are extracted, including the area, height, number of windows, etc. of the space. These feature vectors can reflect the physical characteristics and use scenarios of the space, providing a data basis for subsequent classification. Then, these feature vectors are input into the support vector machine model. SVM distinguishes spaces with different function attributes by finding an optimal classification hyperplane. In dealing with linearly inseparable problems, SVM uses kernel functions to map data to high-dimensional space, making data linearly separable in that space, thereby simplifying the classification problem. The classification process of SVM is based on the principle of structural risk minimization, which improves the generalization ability of the model by maximizing the classification interval. For multi-classification problems, such as the classification of office spaces, teaching spaces, residential spaces, and commercial spaces in the present application, multiple binary SVM models can be constructed to achieve this. For example, the "one-versus-all" or "one-versus-one" method is used to decompose the multi-class problem into multiple binary classification problems. Through the above process, the space function attribute analysis unit can accurately classify the function attributes of the space, providing accurate space type information for the light environment regulation decision module, thereby realizing personalized light environment regulation according to different space function requirements.
[0069] Embodiment 5: The energy saving requirement analysis unit analyzes the energy saving requirements of the space, including the energy efficiency level and energy consumption indicators of the lighting devices, and establishes an energy consumption model of the lighting devices using a regression analysis algorithm. Specifically, historical energy consumption data and related parameters of the lighting devices are collected, including power, usage time, etc. These data are input into the regression analysis model to obtain the energy consumption model of the lighting devices, and the energy consumption limit of the lighting devices is determined according to the energy consumption model.
[0070] The technical principle of the energy-saving requirement analysis unit in the present application is based on a regression analysis algorithm, aiming to determine the energy consumption limit of the lighting device by establishing an energy consumption model. The technical essence is to use historical data and related parameters to predict the energy consumption performance of the lighting device under different operating conditions through mathematical modeling and statistical analysis, thereby providing energy-saving basis for light environment regulation. In the working process, the energy-saving requirement analysis unit first collects the historical energy consumption data of the lighting device and related parameters, including the power, usage time, switch state, etc. of the device. These data reflect the energy consumption characteristics of the lighting device in actual operation and are the basis for establishing the energy consumption model. Then, these data are input into the regression analysis model. Regression analysis is a statistical method used to study the relationship between variables. In the present application, through regression analysis, a mathematical model can be established that can predict the energy consumption of the lighting device according to the input parameters. Specifically, the regression analysis model will find a best function relationship by fitting the historical data, so that the error between the predicted value of the model and the actual energy consumption data is minimized. This function relationship can be linear or nonlinear, depending on the characteristics and complexity of the data. Once the model is established, the expected energy consumption of the lighting device can be calculated according to the input current or predicted operating parameters. By analyzing the output of the energy consumption model, the energy-saving requirement analysis unit can determine the energy consumption limit of the lighting device and provide energy-saving constraint conditions for the light environment regulation decision-making module. Through this regression analysis-based energy consumption model, the energy-saving requirement analysis unit can scientifically evaluate the energy consumption performance of the lighting device and develop reasonable energy-saving strategies accordingly. This not only helps to reduce the energy consumption of the lighting system, but also realizes efficient use of energy while meeting the light environment demand.
[0071] Embodiment 6: The regulation effect monitoring unit monitors the regulation effect of the lighting device, including illumination intensity, color temperature, energy consumption and personnel comfort, and uses a sensor network to collect the regulation effect data of the lighting device in real time. Specifically, a plurality of sensors are arranged in the space to collect illumination intensity, color temperature, energy consumption and personnel comfort data, and the collected data are transmitted to the regulation effect monitoring unit for processing and analysis.
[0072] The technical principle of the regulation effect monitoring unit in the present application is to collect the regulation effect data of the lighting device in real time through the sensor network to evaluate the actual performance of the light environment regulation. The technical essence lies in using multiple sensors to obtain key indicators such as illumination intensity, color temperature, energy consumption, and personnel comfort, thereby providing real-time feedback for the optimization of the light environment regulation system. In the working process, the regulation effect monitoring unit first arranges multiple sensors in the space, which are distributed at different positions to ensure the comprehensiveness and accuracy of the data. Specifically, the illumination sensor is used to measure the illumination intensity in the space, the color temperature sensor is used to detect the color temperature of the light, the energy consumption sensor is used to monitor the energy consumption of the lighting device, and the personnel comfort sensor evaluates the comfort of the light environment by collecting environmental parameters or directly obtaining personnel feedback. These sensors collect data in real time and transmit the collected illumination intensity, color temperature, energy consumption, and personnel comfort data to the regulation effect monitoring unit. In the monitoring unit, the data is processed and analyzed to evaluate the actual effect of the light environment regulation. For example, by comparing the illumination intensity with the preset target value, it can be judged whether the lighting device has reached the expected brightness; by analyzing the color temperature data, it can be evaluated whether the comfort of the light environment meets the needs of the personnel; the energy consumption data is used to evaluate the operating efficiency and energy saving effect of the lighting device; the personnel comfort data directly reflects the subjective feeling of the personnel to the light environment. The real-time data collection and analysis function of the regulation effect monitoring unit enables the system to dynamically evaluate the effect of the light environment regulation and feed back the monitoring results to the light environment regulation decision module. This provides an important basis for the adaptive regulation of the system, enabling the system to adjust the regulation strategy in a timely manner according to the actual effect, further optimize the light environment, improve the comfort of the personnel, and at the same time ensure the realization of the energy saving goal.
[0073] Embodiment 7; The feedback information collection unit collects the feedback information of the personnel to the light environment in the space, including collection through mobile terminals, smart watches and smart glasses, and transmission of the feedback information to the feedback information collection unit using wireless communication technology, specifically: the mobile terminal, smart watch and smart glasses transmit the feedback information to the feedback information collection unit through Bluetooth or Wi-Fi.
[0074] The technical principle of the feedback information collection unit in the present application is to collect subjective feedback information of personnel in the space on the light environment in real time through various wearable devices and mobile terminals, and to transmit these feedback information to the system using wireless communication technology. The technical essence lies in realizing the interaction between personnel and the light environment regulation system, thereby providing a direct basis for the adaptive regulation of the system. In the working process, the feedback information collection unit collects the subjective evaluation of personnel on the light environment with the help of mobile terminals, smart watches and smart glasses and other devices. These feedback information may include evaluation on light intensity, color temperature, comfort and other aspects, as well as specific improvement suggestions. These devices can conveniently collect real-time feedback of users due to their portability and popularity, and provide rich user data for the system. In order to realize efficient transmission of data, the feedback information collection unit adopts wireless communication technology such as Bluetooth or Wi-Fi. These technologies have low power consumption, high stability and wide compatibility, which can ensure timely and accurate transmission of feedback information. Specifically, the mobile terminal, smart watch and smart glasses send the collected feedback information to the feedback information collection unit through the built-in Bluetooth or Wi-Fi module. After the collection unit receives these information, it performs preliminary processing and analysis, and then transmits it to the light environment demand prediction module. In this way, the feedback information collection unit can not only obtain the subjective feedback of users in real time, but also quickly deliver these feedback information to the system, so as to realize dynamic optimization of light environment regulation. This interactive mechanism based on user feedback enables the system to better meet the individual needs of users and improve the comfort and satisfaction of the light environment.
[0075] Example 8: The feedback information processing unit processes and analyzes the feedback information, including data mining and machine learning, and uses a clustering analysis algorithm to classify the feedback information. Specifically, the feedback information is input into a clustering analysis model, similar feedback information is classified into a category according to the characteristics of the feedback information, and the classification result of the feedback information is obtained.
[0076] The technical principle of the feedback information processing unit in the present application is to process the collected personnel feedback information through data mining and machine learning, to realize the classification and analysis of the feedback information. The technical essence lies in using clustering analysis algorithm to efficiently classify the feedback information, thereby providing structured data support for the light environment demand prediction module. In the working process, the feedback information processing unit first receives the raw data from the feedback information collection unit, which may include personnel's subjective evaluation of light intensity, color temperature, comfort, etc. These feedback information usually has diversity and complexity, so effective processing and analysis are needed to extract valuable information. In order to achieve this goal, the feedback information processing unit uses clustering analysis algorithm to classify the feedback information. Clustering analysis is an unsupervised learning method that can classify similar data points into a class according to the intrinsic characteristics of the data. Specifically, the feedback information is input into the clustering analysis model, which automatically identifies the similarity between data according to the characteristics of the feedback information, and classifies similar feedback information into a class. Through clustering analysis, the feedback information processing unit can divide a large amount of feedback information into several categories with similar characteristics. For example, feedback with high evaluation of light intensity can be classified into one category, and feedback with low evaluation of color temperature can be classified into another category. This classification result not only simplifies the data structure, but also provides more targeted information for subsequent light environment demand prediction. Finally, the feedback information processing unit transmits the classified results to the light environment demand prediction module, providing basis for the optimization and adjustment of the model. In this way, the feedback information processing unit can convert the user's subjective feedback into structured data that the system can recognize and process, thereby realizing the dynamic optimization and personalized service of the light environment regulation system.
[0077] Embodiment 9: The light environment demand prediction module further comprises a model evaluation unit for evaluating the prediction effect of the deep reinforcement learning model and optimizing the model according to the evaluation result, using mean square error and mean absolute error as evaluation indexes, specifically: calculating the mean square error and mean absolute error between the predicted value and the actual value, and adjusting the parameters of the deep reinforcement learning model such as learning rate, batch size, etc. according to the evaluation result.
[0078] The technical principle of the model evaluation unit of the light environment demand prediction module in the present application is to quantitatively evaluate the prediction effect of the deep reinforcement learning model and optimize the model parameters according to the evaluation results. The technical essence is to use the mean square error and the mean absolute error as evaluation indexes to measure the difference between the predicted value and the actual value of the model, so as to realize the accurate evaluation of the performance of the model. In the working process, the model evaluation unit first calculates the mean square error between the predicted value and the actual value. MSE measures the accuracy of the prediction by the average value of the squared error, which is sensitive to outliers and can effectively reflect the overall deviation of the model prediction. At the same time, the evaluation unit also calculates the mean absolute error, which is the average value of the absolute difference between the predicted value and the actual value. MAE provides an intuitive average error size, which is easier to explain and understand. According to the evaluation results of MSE and MAE, the model evaluation unit adjusts the parameters of the deep reinforcement learning model. For example, if the MSE or MAE value is high, it means that there is a large deviation between the model prediction and the actual value, and at this time, the learning rate, batch size and other parameters of the model need to be adjusted. The learning rate determines the step size of the model parameter update, and appropriate adjustment of the learning rate can speed up the convergence speed of the model or avoid overfitting; the batch size affects the stability and efficiency of the model training, and reasonable adjustment of the batch size helps to improve the generalization ability of the model. Through this evaluation and optimization mechanism based on MSE and MAE, the light environment demand prediction module can continuously improve the prediction accuracy and stability of the deep reinforcement learning model, so as to provide more accurate decision basis for the adaptive regulation of the intelligent space light environment.
[0079] Embodiment 10: The lighting device control unit regulates the lighting devices in the space according to the light environment regulation decision, including independently regulating the lighting devices in different regions, and using a distributed control system to realize independent regulation of the lighting devices in different regions. Specifically, the space is divided into multiple regions, and a controller is arranged in each region. The controller regulates the lighting devices in the region according to the light environment regulation decision.
[0080] The technical principle of the lighting device control unit in the present application is to realize independent regulation and control of lighting devices in different areas in the space based on a distributed control system. The technical essence lies in realizing precise control of lighting devices in each area through area division and independent deployment of controllers, thereby meeting the light environment needs of different areas. In the working process, first, the space is divided into multiple independent areas, and each area is arranged with corresponding lighting devices according to its function and needs. Each area is equipped with an independent controller, and the controllers are connected to the central control system through wireless communication technology. The central control system sends regulation and control instructions to the controllers of each area according to the instructions of the light environment regulation and control decision module. After receiving the instructions, the controller of each area regulates and controls the lighting devices in the area according to the instruction content, including adjusting the illumination intensity, color temperature and other parameters. The advantage of this distributed control mode is that it can realize fine management of lighting devices in different areas. For example, in the office area, high-brightness and cold-color-temperature lighting can be provided according to the work needs; while in the rest area, soft and warm-color-temperature lighting can be provided. At the same time, through independent regulation and control, the lighting strategy can also be flexibly adjusted according to the use of different areas, realizing energy-saving effect.
[0081] Although the specific embodiments of the present application are described above, those skilled in the art should understand that these specific embodiments are only illustrative, and those skilled in the art can make various omissions, substitutions and changes to the details of the above-mentioned methods and systems without departing from the principles and essence of the present application. For example, combining the above method steps, thereby performing substantially the same function in substantially the same manner to achieve substantially the same result, is within the scope of the present application. Therefore, the scope of the present application is only limited by the appended claims.
Claims
1. An intelligent spatial light environment adaptive control system integrating multi-sensor data, characterized by: include: A multi-source heterogeneous sensor data acquisition module, which includes a light sensor, a color temperature sensor, a people counting sensor, a human physiological parameter sensor, and a temperature and humidity sensor, is used to collect light intensity, color temperature, occupant distribution, human physiological parameters, and ambient temperature and humidity sensor data within the space; A light environment demand prediction module based on deep reinforcement learning includes a data preprocessing unit, a deep reinforcement learning model, and a light environment demand prediction unit. The data preprocessing unit is used to clean, normalize, and extract features from the collected multi-source heterogeneous sensor data. The deep reinforcement learning model is used to construct a deep reinforcement learning model and train the model using the preprocessed data. The light environment demand prediction unit is used to use the trained deep reinforcement learning model to predict the light environment demand of people in the space. A light environment control decision module includes a space function attribute analysis unit, an energy-saving requirement analysis unit, and a control decision generation unit. The space function attribute analysis unit is used to analyze the functional attributes of the space and determine the lighting requirements of the space. The energy-saving requirement analysis unit is used to analyze the energy-saving requirements of the space and determine the energy consumption limit of the lighting equipment. The control decision generation unit is used to generate a light environment control decision based on the light environment demand prediction results, the space function attributes, and the energy-saving requirements. An intelligent execution module, comprising a lighting device control unit and a control effect monitoring unit. The lighting device control unit is used to control the lighting devices in the space according to the light environment control decision. The control effect monitoring unit is used to monitor the control effect of the lighting devices and transmit the monitoring results to the light environment control decision module to optimize the light environment control decision. A multimodal interactive feedback module, comprising a feedback information collection unit and a feedback information processing unit. The feedback information collection unit is used to collect feedback information on the light environment from people in the space, including comfort evaluation, light intensity requirements, and color temperature requirements. The feedback information processing unit is used to process and analyze the feedback information and transmit the processed feedback information to the light environment demand prediction module to optimize the light environment demand prediction model. The deep reinforcement learning model includes a state space module, an action space module, a reward function module, a rule learning module, and a rule dynamic adjustment module. The state space includes the light intensity, color temperature, occupant distribution, human physiological parameters, and ambient temperature and humidity in the space. The action space includes the light intensity and color temperature adjustment amount of the lighting equipment. The reward function is designed based on the comfort of the occupants and the energy saving effect. The rule learning module includes different data information that affects the lighting parameters in the adjustment space. The rule dynamic adjustment module is used to adjust different information parameters of the indoor space. Wherein: the rule learning module uses an association rule mining algorithm to mine the association relationship between different data information affecting the lighting parameters in the space, and adjusts the different data information affecting the lighting parameters in the space according to the association rules; The rule dynamic adjustment module uses an adaptive control algorithm to achieve dynamic adjustment of different data information in the indoor space. According to the actual situation of the indoor space, the weight and threshold of the data information are adjusted in real time, making the light environment control decision more accurate and reasonable. In the reward function module, by establishing an evaluation index system for personnel comfort, including light intensity, color temperature, ambient temperature and humidity, the membership function of each evaluation index is defined, and then the personnel comfort is calculated using the fuzzy comprehensive evaluation method; In the reward function module, the energy saving effect is calculated by the energy consumption rate, which is calculated as actual energy consumption / theoretical energy consumption. The theoretical energy consumption is calculated based on the functional attributes and lighting requirements of the space, and the energy saving effect is measured by the energy consumption rate. In the deep reinforcement learning model, the state space module also includes the geographic location information of the space, and the action space module also includes the on / off status of the lighting equipment. By incorporating the geographic location information and the on / off status into the model, the prediction accuracy and control effect of the model are improved.
2. The intelligent spatial light environment adaptive control system integrating multi-sensor data according to claim 1 is characterized in that: The data preprocessing unit cleans the collected multi-source heterogeneous sensor data and uses a wavelet transform algorithm to remove noise data and abnormal data. The method is as follows: The collected data is decomposed by wavelet to obtain wavelet coefficients of different frequencies. By setting a threshold, the wavelet coefficients smaller than the threshold are set to zero, and then wavelet reconstruction is performed to obtain the cleaned data.
3. The intelligent spatial light environment adaptive control system integrating multi-sensor data according to claim 1 is characterized in that: The deep reinforcement learning model training unit uses the preprocessed data to train the model, adopting a combination of offline training and online training. During offline training, the deep reinforcement learning model is pre-trained using a historical data set to obtain an initial model; during online training, the real-time collected data is input into the initial model, and the model parameters are updated through the gradient descent algorithm, specifically: the loss function between the predicted value and the actual value is calculated, and then the model parameters are adjusted according to the gradient direction of the loss function to minimize the loss function.
4. The intelligent spatial light environment adaptive control system integrating multi-sensor data according to claim 1 is characterized in that: The spatial functional attribute analysis unit analyzes the functional attributes of the space, including office space, teaching space, residential space and commercial space, and classifies the functional attributes of the space using a support vector machine algorithm. Specifically, the feature vector of the space is extracted, including the area, height, number of windows, etc. of the space, and the feature vector is input into the support vector machine model to obtain the functional attribute classification result of the space.
5. The intelligent spatial light environment adaptive control system integrating multi-sensor data according to claim 1 is characterized in that: The energy-saving requirement analysis unit analyzes the energy-saving requirements of the space, including the energy efficiency level and energy consumption indicators of the lighting equipment, and uses a regression analysis algorithm to establish an energy consumption model of the lighting equipment. Specifically, the historical energy consumption data and related parameters of the lighting equipment, including power, usage time, etc., are collected, and these data are input into the regression analysis model to obtain the energy consumption model of the lighting equipment, and the energy consumption limit of the lighting equipment is determined based on the energy consumption model.
6. The intelligent spatial light environment adaptive control system integrating multi-sensor data according to claim 1 is characterized in that: The control effect monitoring unit monitors the control effect of the lighting equipment, including light intensity, color temperature, energy consumption and personnel comfort, and uses a sensor network to collect the control effect data of the lighting equipment in real time. Specifically, multiple sensors are arranged in the space to collect light intensity, color temperature, energy consumption and personnel comfort data respectively, and the collected data is transmitted to the control effect monitoring unit for processing and analysis.
7. The intelligent spatial light environment adaptive control system integrating multi-sensor data according to claim 1 is characterized in that: The feedback information collection unit collects feedback information about the light environment from people in the space, including collecting information through mobile terminals, smart watches and smart glasses, and transmitting the feedback information to the feedback information collection unit using wireless communication technology. Specifically, the mobile terminal, smart watch and smart glasses transmit the feedback information to the feedback information collection unit via Bluetooth or Wi-Fi.
8. The intelligent spatial light environment adaptive control system integrating multi-sensor data according to claim 1 is characterized in that: The feedback information processing unit processes and analyzes the feedback information, including data mining and machine learning, and classifies the feedback information using a cluster analysis algorithm. Specifically, the feedback information is input into a cluster analysis model, and similar feedback information is classified into one category based on the characteristics of the feedback information to obtain a classification result of the feedback information.
9. The intelligent spatial light environment adaptive control system integrating multi-sensor data according to claim 1 is characterized in that: The light environment demand prediction module also includes a model evaluation unit, which is used to evaluate the prediction effect of the deep reinforcement learning model and optimize the model based on the evaluation results, using mean square error and mean absolute error as evaluation indicators. Specifically, the mean square error and mean absolute error between the predicted value and the actual value are calculated, and the parameters of the deep reinforcement learning model, such as learning rate, batch size, etc., are adjusted according to the evaluation results.
10. The intelligent spatial light environment adaptive control system integrating multi-sensor data according to claim 1 is characterized in that: The lighting equipment control unit controls the lighting equipment in the space according to the light environment control decision, including independent control of the lighting equipment in different areas. A distributed control system is used to achieve independent control of the lighting equipment in different areas. Specifically, the space is divided into multiple areas, and a controller is arranged in each area. The controller controls the lighting equipment in the area according to the light environment control decision.
Citation Information
Cited By
Construction site illumination intelligent energy-saving system and control method
CN121126633A
Intelligent collaborative regulation and control method and system for photo-thermal equipment giving consideration to comfort, health and energy conservation
CN121596763A
