Intelligent bulb lamp control method based on target working mode

Abstract

This invention discloses an intelligent bulb lamp control method based on a target operating mode, belonging to the field of intelligent lighting technology. This invention solves the problem of existing methods requiring manual adjustment of lighting parameters and lacking flexibility in adjusting the operating mode according to actual usage. By setting a confidence threshold and minimum duration, it ensures the accuracy and stability of mode switching, avoiding erroneous switching. During the transition process, it continuously monitors environmental and user behavior data; if negative feedback or a sudden change in situation is detected, it immediately interrupts the transition and re-performs mode recognition. It automatically adjusts brightness according to ambient light intensity, making full use of natural light and saving energy. When the user leaves the light-covered area for an extended period, it automatically switches to energy-saving mode, further improving energy efficiency. This makes lighting control more tailored to the user's personalized needs, achieving the dual goals of energy saving and personalized control.

Description

Technical Field

[0001] This invention relates to the field of intelligent lighting technology, specifically to an intelligent bulb lamp control method based on a target operating mode. Background Technology

[0002] As people's demands for quality of life and energy efficiency continue to rise, intelligent lighting systems are gradually becoming an important development direction for home and commercial lighting.

[0003] While traditional smart bulbs can be remotely controlled via mobile apps or voice assistants, in practice, users need to frequently and manually adjust parameters such as brightness and color to suit different scenarios. This is not only cumbersome but also makes it difficult to achieve precise and personalized lighting effects. Furthermore, existing smart bulbs also have limitations in energy saving, as they cannot flexibly adjust their operating modes according to actual usage to achieve optimal energy efficiency.

[0004] Therefore, to meet current needs, a smart bulb control method based on the target operating mode is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent bulb light control method based on a target operating mode. By setting a confidence threshold and a minimum duration, the method ensures the accuracy and stability of mode switching and avoids erroneous switching. During the gradual change process, it continuously monitors environmental and user behavior data. Once negative feedback or a sudden change in situation is detected, the gradual change is immediately interrupted and the mode recognition is restarted. The method automatically adjusts the brightness according to the ambient light intensity to make full use of natural light and save energy. When the user leaves the light coverage area for a long time, the method automatically switches to an energy-saving mode to further improve energy efficiency. This makes the lighting control more in line with the user's personalized needs, achieving the dual goals of energy saving and personalized control, and solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] The intelligent bulb lamp control method based on the target operating mode includes the following steps:

[0008] S1. Based on historical usage data from IoT data acquisition technology, multiple target working modes are preset and parameterized based on historical usage data. Each mode consists of a mode feature vector and a corresponding multi-dimensional lighting parameter vector, including but not limited to reading mode, movie-watching mode, working mode, sleep mode, and party mode. The environmental and behavioral conditions that trigger the feature vector of each working mode are defined, as well as the corresponding lighting parameter output set.

[0009] S2. Through multiple sensors deployed on the smart bulb itself, raw environmental data and user behavior data are collected in real time. After the raw data is initially filtered and feature extracted at the local node, it is transmitted in real time to the smart home central gateway for data fusion through near-field wireless communication technology to form a real-time feature data stream.

[0010] S3. Based on the decision tree, perform coarse classification of the real-time feature data stream, analyze the coarse classification results using the time-series pattern recognition algorithm, combine the historical feature sequence of the specified time window to perform fine pattern recognition, and output the most likely pattern and its confidence level.

[0011] S4. Based on the identified target working mode, obtain the corresponding lighting parameter output set from the preset mode parameter library; automatically adjust the output of the smart bulb lamp according to the obtained lighting parameters to match the light with the target working mode.

[0012] S5. Provide user feedback options based on the control interface, allowing users to manually confirm or correct the current working mode identified by the system; record user feedback data into the system, and update the parameters of the decision tree and time-series pattern recognition model in real time based on user feedback; generate a personalized pattern parameter library for each user based on user feedback and behavioral habits.

[0013] Furthermore, in S3, the most likely current pattern and its confidence level are output, including the following steps:

[0014] Set a confidence threshold and a minimum duration to evaluate the confidence of the identified new patterns;

[0015] A mode switch is triggered when the confidence level of the new mode is greater than the confidence threshold and the prediction continues for more than the minimum duration.

[0016] During switching, a non-linear gradient is used to smoothly transition from the current value to the target value, ensuring a smooth user experience.

[0017] During the smoothing of lighting parameters, real-time environmental data and user behavior data are continuously monitored; if a clear negative feedback is detected or a sudden change in the situation occurs, the current gradual change process is immediately interrupted, and pattern recognition and decision-making are re-performed based on the latest data.

[0018] Furthermore, the smooth transition from the current value to the target value through a non-linear gradual change includes the following steps:

[0019] Calculate the perception deviation index based on the current value and the target value;

[0020] The type of scene to switch is determined based on the perception deviation index;

[0021] When the scene switching type is dark switching scene, a positive support type compensation signal is generated so that the actual brightness driving trajectory is temporarily maintained at a level higher than the target set brightness in the early stage of switching, and slowly decays to approach the target set brightness according to the duration characteristics of dark adaptation over time, so as to form a trailing transition.

[0022] When the scene switching type is bright switching scene, a reverse suppression type compensation signal is generated, so that the actual brightness driving trajectory is limited to a level lower than the target set brightness in the early stage of switching, and gradually rises to the target set brightness according to the brightness adaptation duration characteristics over time, so as to form a soft start transition.

[0023] The actual driving brightness value is the main control variable, and the first color temperature change curve is calculated in real time using the human visual psychological comfort model.

[0024] The total range of color temperature variation is divided into multiple color tolerance ranges;

[0025] Based on the McAdam ellipse principle, the sensitivity of the human eye to color changes in different color temperature ranges is evaluated.

[0026] Based on the sensitivity-rate of change comparison table, determine the rate of change for different color tolerance ranges and plot the second color temperature change curve;

[0027] The first color temperature change curve and the second color temperature change curve are weighted and fused to obtain the actual color temperature driving trajectory.

[0028] Furthermore, if a clear negative feedback is detected or a sudden change in the situation occurs, the current gradual change process is immediately interrupted, and pattern recognition and decision-making are re-performed based on the latest data, including:

[0029] Obtain the target lighting parameter vector adjusted by the user after receiving negative feedback;

[0030] Calculate the deviation vector between the target illumination parameter vector and the reference parameter vector of the current target operating mode;

[0031] The Euclidean distance is calculated based on the deviation vector. When the Euclidean distance falls within the preset target range, the corresponding target lighting parameter vector is used as a valid learning sample; otherwise, it is discarded as an outlier.

[0032] Based on the dual momentum mechanism, the baseline parameter vector of the current target working mode is updated according to the deviation vector, the user's historical adjustment trend, the fluctuation energy of the user's historical adjustment amplitude, and the baseline parameter vector of the current target working mode.

[0033] Pattern recognition and decision-making are based on the updated pattern parameter library.

[0034] Furthermore, continuous monitoring of real-time environmental data and user behavior data includes the following steps:

[0035] During the target working mode recognition process, if it is detected that the user has left the light coverage area for a long time, exceeding the set threshold time, the bulb will automatically switch to energy-saving mode, reduce the brightness to the minimum or turn off the light.

[0036] The bulb brightness is automatically adjusted according to the ambient light intensity. When there is sufficient natural light, the bulb brightness output is reduced.

[0037] Based on the incremental data of each manual adjustment by the user, the lighting parameter output set of the corresponding mode is updated online, and the preset mode parameter library is optimized to make the lighting control more in line with the user's personalized needs.

[0038] Historical contextual data is analyzed periodically using unsupervised clustering algorithms to identify user-defined behavioral patterns that are not covered by preset patterns, and new custom target working patterns are created accordingly.

[0039] Furthermore, in S2, preliminary filtering and feature extraction are performed on the raw data at the local node, including the following steps:

[0040] Edge processing is performed by a built-in lightweight microprocessor on various sensors deployed on the smart bulb body;

[0041] Perform moving average filtering, median filtering, or Kalman filtering on the raw sensor signal to eliminate random noise and transient interference pulses;

[0042] The illuminance change slope and current color temperature value are extracted from the ambient light data; the number of targets, the main target's movement speed, the main target's dwell time in the preset area, and the micro-motion frequency are extracted from the millimeter-wave radar data.

[0043] Extract confidence scores, ambient noise levels, and whether voice command keywords trigger specific sound sources from audio features; extract posture classification confidence and activity intensity index from visual features;

[0044] The extracted feature values, along with the data acquisition timestamp, device ID, and sensor type identifier, are encapsulated into a data packet; and the data packet is transmitted to the smart home central gateway via near-field wireless communication technology.

[0045] Furthermore, data packets are transmitted to the smart home central gateway using near-field communication technology, including the following steps:

[0046] Different transmission priorities are set for different types of feature data, including: level 1, level 2, level 3 and level 4;

[0047] Establish a feature-priority mapping table. When extracting real-time feature values, query the pattern feature vector and the priority mapping table to generate data packets with priority labels.

[0048] Data transmission is performed according to the transmission priority of the data packet, and wireless channel congestion is continuously monitored;

[0049] When the channel load is high, the transmission frequency of the third and fourth level data will be automatically reduced or the fourth level data will be temporarily suspended to ensure the smooth transmission of the first and second level critical data.

[0050] Furthermore, in S3, a coarse classification of the real-time feature data stream is performed based on a decision tree, and the coarse classification results are analyzed using a time-series pattern recognition algorithm, including the following steps:

[0051] Based on historical usage data, select features that have an impact on pattern recognition;

[0052] Label historical data to identify pattern instances, and then use the labeled historical data to train a decision tree model.

[0053] A decision tree is generated from the training data. Each node represents a judgment condition for a feature, each branch represents a range of feature values, and each leaf node represents a coarse classification result.

[0054] The real-time feature data stream is input into the decision tree model. Starting from the root node, the model moves down the branches of the decision tree according to the current feature value until it reaches the leaf node, and outputs the coarse classification result.

[0055] Furthermore, in S3, refined pattern recognition is performed by combining historical feature sequences within a specified time window, including the following steps:

[0056] Select a specified time window to extract the historical feature sequence of the specified time window from the real-time feature data stream;

[0057] The historical feature sequence is input into the time series pattern recognition model. Based on the input time series features and the trained pattern recognition logic, the model outputs the most likely pattern and its confidence level.

[0058] The coarse classification results of the decision tree and the fine classification results of the temporal pattern recognition are comprehensively judged. If the two results are consistent, the confidence score is increased; if the results are inconsistent, the final judgment is made based on the confidence score.

[0059] The system combines confidence level and minimum duration to determine whether to trigger mode switching; if the confidence level of the current mode exceeds the threshold and the duration exceeds the minimum duration, then mode switching is triggered.

[0060] Furthermore, a comprehensive judgment is made on the coarse classification results of the decision tree and the fine classification results of the temporal pattern recognition, including the following steps:

[0061] Obtain coarse classification results and fine classification results. The coarse classification results include: the initial pattern classification result and its confidence level; the fine classification results include: the fine classification result and its confidence level.

[0062] Compare the coarse classification results of the decision tree with the fine classification results of the temporal pattern recognition to determine whether the two are consistent;

[0063] If they match, they are the same pattern, and the confidence scores of the two are weighted and averaged to obtain a higher confidence score. The overall confidence score is then compared with a preset confidence score threshold. If the overall confidence score exceeds the threshold, the current pattern is confirmed as the pattern, and the final result is output.

[0064] If they are inconsistent, the confidence levels of the two are compared, and the result with the higher confidence level is selected as the final judgment. The selected confidence level is then compared with the preset confidence level threshold. If the confidence level exceeds the threshold, the current mode is confirmed as the current mode, and the final result is output.

[0065] Compared with the prior art, the beneficial effects of the present invention are:

[0066] In this invention, by setting a confidence threshold and a minimum duration, the accuracy and stability of mode switching are ensured, avoiding erroneous switching. Simultaneously, a non-linear gradation method is used to smoothly transition lighting parameters, ensuring a smooth user experience. During the gradation process, environmental and user behavior data are continuously monitored; if negative feedback or a sudden change in context is detected, the gradation is immediately interrupted and mode recognition is restarted. Brightness is automatically adjusted according to ambient light intensity to fully utilize natural light and save energy. When the user leaves the light-covered area for an extended period, the system automatically switches to energy-saving mode, further improving energy efficiency. This makes lighting control more tailored to the user's personalized needs, achieving the dual goals of energy saving and personalized control. Attached Figure Description

[0067] Figure 1 This is a flowchart of the intelligent bulb lamp control method based on the target working mode of the present invention. Detailed Implementation

[0068] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0069] To address the technical issue that existing methods require manual adjustment of lighting parameters and cannot flexibly adjust the operating mode according to actual usage, please refer to [link to relevant documentation]. Figure 1 This embodiment provides the following technical solution:

[0070] The intelligent bulb lamp control method based on the target operating mode includes the following steps:

[0071] S1. Based on historical usage data from IoT data acquisition technology, including user records of lighting parameter adjustments in different scenarios, sensor data, etc.; based on historical usage data, preset and parameterize multiple target working modes. Each mode consists of a mode feature vector and a corresponding multi-dimensional lighting parameter vector, including but not limited to reading mode, movie-watching mode, work mode, sleep mode, and party mode. Define the environmental and behavioral conditions that trigger the feature vector of each working mode, as well as the corresponding lighting parameter output set; for example, reading mode includes: time window such as 18:00-23:00, presence of stationary human body in the main activity area, ambient light brightness below the threshold, sound spectrum characteristics such as: no continuous human voice or TV sound, etc. The lighting parameters corresponding to reading mode are: brightness: 90%, color temperature: 4200K, color: RGB (255, 240, 220).

[0072] S2. Using multiple sensors deployed on the smart bulb itself, such as light sensors, infrared human body sensors, sound sensors, and cameras, raw environmental data and user behavior data are collected in real time. Specifically, the light sensor detects indoor light intensity; the infrared human body sensor detects the user's activity area and stationary state; the sound sensor identifies specific sound sources, such as television sound; and the camera further analyzes the user's posture and behavior. After preliminary filtering and feature extraction of the raw data at the local node, it is transmitted in real time to the smart home central gateway via short-range wireless communication technologies such as Bluetooth and ZigBee for data fusion, forming a real-time feature data stream. This ensures time synchronization of multi-sensor data, providing a consistent snapshot of the context for subsequent analysis. This includes the following steps:

[0073] The smart bulb incorporates multiple sensors, each with a built-in lightweight microprocessor for edge processing. It performs moving average filtering, median filtering, or Kalman filtering on the raw sensor signals to eliminate random noise and transient interference pulses. It extracts the illuminance change slope and current color temperature from ambient light data; the number of targets, the main target's speed, the main target's dwell time in a preset area, and micro-motion frequency from millimeter-wave radar data; specific sound sources such as television confidence scores, ambient noise levels, and the presence of voice command keywords from audio features; and posture classification confidence and activity intensity index from visual features. The extracted feature values, along with the data acquisition timestamp, device ID, and sensor type identifier, are encapsulated into a data packet. This data packet is then transmitted to the smart home hub gateway via near-field communication technology.

[0074] Different transmission priorities are set for different types of feature data, including: Level 1, Level 2, Level 3, and Level 4. For example, the presence / disappearance event of human body detection is the highest priority message, which must ensure extremely low latency and reliable transmission; while continuous ambient light readings can be uploaded periodically as a lower priority. A feature-priority mapping table is established. When extracting real-time feature values, the pattern feature vector and the priority mapping table are queried to generate data packets with priority tags. Data transmission is executed according to the transmission priority of the data packet, and the wireless channel congestion is continuously monitored. When the channel load is high, the transmission frequency of Level 3 and Level 4 data is automatically reduced or Level 4 data is temporarily suspended to ensure the smooth transmission of Level 1 and Level 2 critical data.

[0075] S3. Based on decision trees, perform coarse classification of real-time feature data streams. Analyze the coarse classification results using a time-series pattern recognition algorithm, and combine this with a specified time window, such as historical feature sequences from the past 30 seconds, to perform fine pattern recognition. Output the most likely pattern and its confidence level. For example, when the light sensor detects dim indoor lighting and the human infrared sensor detects that the user has been stationary at the desk for a period of time, it can be determined that the user is in reading mode; when the sound sensor detects television sound and the human infrared sensor detects that the user is moving around in the sofa area, it can be determined that the user is in movie-watching mode. This includes the following steps:

[0076] Set a confidence threshold and a minimum duration, such as 80% or 3 minutes, to evaluate the confidence of newly identified patterns. When the confidence of a new pattern exceeds the confidence threshold and the prediction continues for longer than the minimum duration, a pattern switch is triggered. During the switch, a smooth transition from the current value to the target value is achieved through a non-linear gradient, ensuring a smooth user experience. During the smoothing of lighting parameters, real-time environmental data and user behavior data are continuously monitored. If a clear negative feedback or a sudden change in the situation is detected, the current gradient process is immediately interrupted, and pattern recognition and decision-making are re-performed based on the latest data.

[0077] During the target working mode recognition process, if a user is detected to have left the light-covered area for an extended period, exceeding a set threshold time (e.g., 10 minutes), the bulb automatically switches to energy-saving mode, reducing brightness to the minimum or turning off the light to minimize energy waste. The bulb's brightness is automatically adjusted based on ambient light intensity; when natural light is sufficient, the bulb's brightness output is reduced to fully utilize natural light and further improve energy efficiency. Based on incremental data from each manual adjustment by the user, the corresponding lighting parameter output set is updated online, optimizing the preset mode parameter library to better suit the user's personalized lighting needs. The learning rate decays with sample accumulation to ensure system convergence and stability. For example, if a user repeatedly adjusts the color temperature to 4500K in reading mode, the system automatically adjusts the default color temperature of reading mode to 4500K. Unsupervised clustering algorithms are periodically used to analyze historical contextual data, identifying user-defined behavioral patterns not covered by preset modes, and creating new custom target working modes accordingly.

[0078] Based on historical usage data, select features that influence pattern recognition; for example, for the reading mode, key features include light intensity, the stillness of the human body, and the absence of continuous sound. Label pattern instances in the historical usage data, such as reading, watching movies, working, etc., and train the decision tree model using the labeled historical data. Generate a decision tree from the training data, where each node represents a judgment condition for a feature, each branch represents a range of feature values, and each leaf node represents a coarse classification result, such as possibly reading mode or non-reading mode. Input the real-time feature data stream into the decision tree model, starting from the root node, and move down the branches of the decision tree according to the current feature value until reaching the leaf node, and output the coarse classification result, such as possibly reading mode or possibly watching movies mode.

[0079] Select a specified time window, such as the historical feature sequence of the past 30 seconds. Extract the historical feature sequence of the specified time window from the real-time feature data stream, including changes in light intensity, changes in human activity state, changes in sound features, and changes in user posture. Input the historical feature sequence into the temporal pattern recognition model. Based on the input time series features and combined with the trained pattern recognition logic, output the most likely current pattern and its confidence level. For example, if the current pattern is reading mode, the confidence level is 90%. Perform a comprehensive judgment on the coarse classification result of the decision tree and the fine classification result of the temporal pattern recognition. If the two results are consistent, increase the confidence level; if the results are inconsistent, make a final judgment based on the confidence level. Combine the confidence level and the minimum duration, such as 3 minutes, to determine whether to trigger a mode switch. If the confidence level of the current pattern exceeds the threshold and the duration exceeds the minimum duration, trigger a mode switch.

[0080] Obtain coarse and fine classification results. Coarse classification results include: the initial pattern classification result and its confidence level; for example, if the decision tree outputs "reading mode," a confidence level of 70% means the decision tree is 70% confident that the current pattern is reading. Fine classification results include: the fine classification result and its confidence level; for example, if temporal pattern recognition outputs "reading mode," a confidence level of 85% means the temporal pattern recognition is 85% confident that the current pattern is reading. Compare the coarse classification results of the decision tree. The system compares the results of the fine classification of temporal pattern recognition with those of the temporal pattern recognition. If they match, they are considered to be the same pattern (e.g., reading pattern). The system then performs a weighted average of the confidence scores of the two patterns to obtain a higher confidence score. The overall confidence score is then compared with a preset confidence threshold (e.g., 80%). If the overall confidence score exceeds the threshold, the current pattern is confirmed as the reading pattern, and the final result is output. For example, if the overall confidence score is 77.5%, which is lower than the 80% threshold, the pattern is not confirmed. If the overall confidence score is 85%, the current pattern is confirmed as the reading pattern.

[0081] If there is a discrepancy, such as the decision tree classifying it as reading mode while the temporal pattern recognition classifies it as watching movie mode, the confidence levels of the two are compared, and the result with the higher confidence level is selected as the final judgment. For example, if the decision tree has a confidence level of 70% and the temporal pattern recognition has a confidence level of 85%, then the result of the temporal pattern recognition (i.e., reading mode) is selected as the final judgment. The selected confidence level is compared with a preset confidence threshold (e.g., 80%). If the confidence level exceeds the threshold, the current mode is confirmed as that mode, and the final result is output. For example, if the confidence level of the temporal pattern recognition is 85%, exceeding the 80% threshold, then the current mode is confirmed as reading mode.

[0082] S4. Based on the identified target working mode, retrieve the corresponding lighting parameter output set from the preset mode parameter library, including: brightness, color temperature, color, flicker frequency, etc.; automatically adjust the output of the smart bulb according to the retrieved lighting parameters to match the light with the target working mode; for example: in reading mode, adjust the brightness to a higher level, such as 80%-100%, and adjust the color temperature to a warm white light suitable for reading, about 4000K-5000K; in sleep mode, reduce the brightness to the lowest level, such as 10%-20%, adjust the color temperature to a warm yellow light, about 2700K-3000K, and can be set to flicker slowly to help the user fall asleep.

[0083] S5. Provide user feedback options based on the control interface, allowing users to manually confirm or correct the current working mode identified by the system; for example, users can click to confirm reading mode or correct it to movie viewing mode; record user feedback data in the system, and update the parameters of the decision tree and temporal pattern recognition model in real time based on user feedback; for example, if the user corrects the system's misclassification of movie viewing mode as reading mode multiple times, the model will automatically adjust the weights of relevant features to improve the accuracy of future recognition; generate a personalized mode parameter library for each user based on user feedback and behavioral habits; for example, if a user prefers lower brightness and higher color temperature in reading mode, the system will automatically adjust the user's reading mode parameters to better meet their personalized needs.

[0084] The beneficial effects achieved by the above are as follows: By setting confidence thresholds and minimum durations, the accuracy and stability of mode switching are ensured, avoiding accidental switching; at the same time, a non-linear gradation method is used to smoothly transition lighting parameters, ensuring a smooth user experience; during the gradation process, environmental and user behavior data are continuously monitored, and once negative feedback or a sudden change in situation is detected, the gradation is immediately interrupted and mode recognition is restarted; brightness is automatically adjusted according to ambient light intensity to make full use of natural light and save energy; when users leave the light coverage area for a long time, the system automatically switches to energy-saving mode to further improve energy efficiency, making lighting control more in line with users' personalized needs, thus achieving the dual goals of energy saving and personalized control.

[0085] Working principle: The system collects user lighting preferences and sensor data in different scenarios through IoT technology, presets multiple target working modes and corresponding specific lighting parameters; uses multiple sensors to collect environmental and user behavior data in real time, performs coarse classification through decision trees, and then performs fine classification by combining time-series pattern recognition algorithms, outputting the most likely pattern and its confidence level; when the confidence level of a new pattern exceeds the threshold and the duration meets the requirements, the system smoothly switches the lighting parameters and continuously monitors environmental and user behavior data so that the switching can be interrupted and the pattern re-identified when negative feedback or sudden changes in the situation are detected; and automatically adjusts the brightness according to the ambient light intensity to optimize energy utilization efficiency.

[0086] In one embodiment, a smooth transition from the current value to the target value using a non-linear gradual change includes the following steps:

[0087] Calculate the perception deviation index based on the current value and the target value;

[0088] In this embodiment, the perceptual deviation index is the difference between the current value and the target value, such as the difference in brightness and color temperature between the current working mode and the target working mode.

[0089] The type of scene to switch is determined based on the perception deviation index;

[0090] In this embodiment, the switching scene types include dark switching scene (working mode switching scene with reduced brightness) and bright switching scene (working mode switching scene with increased brightness).

[0091] When the scene switching type is dark switching scene, a positive support type compensation signal is generated so that the actual brightness driving trajectory is temporarily maintained at a level higher than the target set brightness in the early stage of switching, and slowly decays to approach the target set brightness according to the duration characteristics of dark adaptation over time, so as to form a trailing transition.

[0092] In this embodiment, the positive support type compensation signal is used to initially provide brightness support higher than the target set brightness during the brightness reduction process;

[0093] In this embodiment, the duration of dark adaptation refers to the adaptation time of the human eye from a bright environment to a dark environment, for example: 60 seconds;

[0094] In this embodiment, the trailing transition refers to the brightness change curve exhibiting an initial high value followed by a slow decay.

[0095] When the scene switching type is bright switching scene, a reverse suppression type compensation signal is generated, so that the actual brightness driving trajectory is limited to a level lower than the target set brightness in the early stage of switching, and gradually rises to the target set brightness according to the brightness adaptation duration characteristics over time, so as to form a soft start transition.

[0096] In this embodiment, the reverse suppression type compensation signal is used to initially limit the brightness to be lower than the target set brightness during the brightness increase process;

[0097] In this embodiment, the duration of light adaptation refers to the adaptation time of the human eye from a dark environment to a bright environment. For example, 10 seconds is shorter than the duration of dark adaptation.

[0098] In this embodiment, the soft-start transition refers to the brightness change curve showing an initial low value followed by a gradual increase.

[0099] The actual driving brightness value is the main control variable, and the first color temperature change curve is calculated in real time using the human visual psychological comfort model.

[0100] In this embodiment, the human visual psychological comfort model refers to a mathematical model that comprehensively considers the physiological characteristics and psychological perception of the human eye;

[0101] In this embodiment, the first color temperature change curve is a comfortable color temperature change path calculated based on the human eye visual psychological comfort model and the actual driving brightness value.

[0102] The total range of color temperature variation is divided into multiple color tolerance ranges;

[0103] In this embodiment, the color tolerance range is a color temperature region divided in standard units on the CIE 1931 chromaticity diagram;

[0104] In this embodiment, the McAdam ellipse is an elliptical region on the CIE chromaticity diagram that represents the human eye's perception of color differences.

[0105] Based on the McAdam ellipse principle, the sensitivity of the human eye to color changes in different color temperature ranges is evaluated.

[0106] In this embodiment, sensitivity refers to the level of sensitivity of the human eye to different color temperature regions, for example:

[0107] Low color temperature range (2700K-3500K, warm white light): The human eye is less sensitive to changes in yellow-red tones and allows for faster changes;

[0108] Medium color temperature range (4000K-5000K, neutral white): The human eye is most sensitive to changes in this range, and a slow transition is necessary;

[0109] High color temperature range (5500K-6500K, cool white light): The human eye has moderate sensitivity to changes in blue-white tones.

[0110] Based on the sensitivity-rate of change comparison table, determine the rate of change for different color tolerance ranges and plot the second color temperature change curve;

[0111] In this embodiment, the sensitivity-rate of change comparison table is a preset table that specifies the allowable rate of color temperature change for different sensitivity levels;

[0112] In this embodiment, the second color temperature change curve is a color temperature change path adjusted based on human eye sensitivity.

[0113] The first color temperature change curve and the second color temperature change curve are weighted and fused to obtain the actual color temperature driving trajectory.

[0114] In this embodiment, during weighted fusion, the weights are adjusted based on the current time (e.g., weights are biased towards circadian rhythm protection at night), the magnitude of brightness changes (e.g., weights are biased towards visual comfort when there are large changes), and user history feedback (weights are adjusted based on correction operations).

[0115] The working principle and beneficial effects of the above technical solution are as follows:

[0116] This invention first calculates the perceptual deviation index between the current brightness and color temperature and the target value to determine whether it is a dark transition scenario or a bright transition scenario. During a dark transition, a positive support compensation signal is used to make the initial brightness higher than the target value, followed by a slow decay according to the dark adaptation time characteristic, forming a trailing transition. During a bright transition, a negative suppression compensation signal is used to make the initial brightness lower than the target value, followed by a gradual increase according to the bright adaptation time characteristic, forming a soft-start transition. Simultaneously, the system uses the actual driving brightness as the main control variable, calculates the first color temperature change curve using a human visual psychological comfort model, and generates a second color temperature change curve by combining the sensitivity characteristics of different color temperature ranges analyzed by the McAdam ellipse principle. Finally, the actual color temperature driving trajectory is obtained by adaptively weighted fusion of the two curves. This invention significantly reduces visual discomfort during mode switching. By distinguishing the different time characteristics of dark adaptation and bright adaptation, it avoids the visual impact caused by traditional linear transitions; while the zonal control strategy for color temperature changes ensures a smooth transition in the human eye's sensitive areas, greatly improving the user experience.

[0117] In one embodiment, if a clear negative feedback is detected or a sudden change in the situation occurs, the current gradual change process is immediately interrupted, and pattern recognition and decision-making are re-performed based on the latest data, including:

[0118] Obtain the target lighting parameter vector adjusted by the user after receiving negative feedback;

[0119] Calculate the deviation vector between the target illumination parameter vector and the reference parameter vector of the current target operating mode;

[0120] The Euclidean distance is calculated based on the deviation vector. When the Euclidean distance falls within the preset target range, the corresponding target lighting parameter vector is used as a valid learning sample; otherwise, it is discarded as an outlier.

[0121] Based on the dual momentum mechanism, the baseline parameter vector of the current target operating mode is updated according to the deviation vector, the user's historical adjustment trend, the fluctuation energy of the user's historical adjustment amplitude, and the baseline parameter vector of the current target operating mode. Specifically, this includes:

[0122] Update the user-adjusted trend vector:

[0123] ;

[0124] in, Adjust the trend vector for the current iteration of users. Adjust the trend vector for users from the previous iteration. For the deviation vector, First-order momentum decay factor ;

[0125] Update the volatility assessment vector:

[0126] ;

[0127] in, This is the volatility evaluation vector for the user adjustment amplitude in the current iteration. This is the volatility evaluation vector of the user adjustment amplitude from the previous iteration. It is the second-order moment attenuation factor. This represents the Hadamard product (element-by-element multiplication).

[0128] Calculate the adaptive learning rate matrix:

[0129] ;

[0130] in, This is the adaptive learning rate matrix for the current iteration; The initial learning rate, To prevent smooth terms with a denominator of zero, For the number of iterations Increased time decay factor, Represents a vector Extracting the square root of each element;

[0131] Update the baseline parameter vector for the current target operating mode:

[0132] ;

[0133] in, The baseline parameter vector for updating the current target operating mode. This is the baseline parameter vector for the current target operating mode.

[0134] Pattern recognition and decision-making are based on the updated pattern parameter library.

[0135] The working principle and beneficial effects of the above technical solution are as follows:

[0136] Negative feedback refers to a user's expression of dissatisfaction with the current lighting effect, including explicit feedback (clicking "I don't like it" in the app, or a voice command saying "It's too dark") and implicit feedback (frequent manual adjustment of the lights, specific gestures). Negative feedback is acquired through multimodal fusion perception: voice commands are obtained through a built-in microphone array and offline keyword recognition; gesture operations are detected through millimeter-wave radar to detect specific gesture patterns; and app interaction events are pushed through the smart home hub. The target lighting parameter vector refers to the final set of lighting parameters after the user manually adjusts the lights, including parameters such as brightness, color temperature, RGB color components, or saturation. The target lighting parameters are directly read from the registers of the LED driver controller. After detecting negative feedback from the user, the system records the lighting parameters that the user has finally adjusted, serving as the basic data for the system to learn user preferences.

[0137] The reference parameter vector for the current target working mode is the standard lighting parameters stored in the mode parameter library that correspond to the current recognition working mode; the deviation vector is a vector composed of the differences between the target parameters and the reference parameters in each dimension. In the bulb microcontroller, each parameter dimension is normalized (e.g., the brightness value is normalized to the range of [0,1]), and then vector subtraction is performed.

[0138] The preset target range is a reasonable adjustment interval determined based on human visual perception characteristics and user behavior research. The target range was determined through extensive user experiments: the lower threshold (usually 5% of the diagonal length of the parameter space) filters out minor adjustments; the upper threshold (usually 60% of the diagonal length) filters out extreme adjustments. These thresholds are stored in the device firmware.

[0139] The dual momentum mechanism refers to a parameter update mechanism that simultaneously considers the first moment (trend) and the second moment (fluctuation). The user's historical adjustment trend refers to the directional preference exhibited by the user in historical adjustments; the fluctuation energy of the user's historical adjustment amplitude refers to the degree of change in the user's adjustment amplitude, reflecting the stability of the adjustment behavior. Through the dual momentum mechanism, the system can smoothly learn user preferences, considering both the adjustment trend and the volatility of the adjustment amplitude, avoiding parameter mutations caused by a single abnormal operation. Specifically, the long-term trend of user adjustments is calculated using exponential moving averages, giving higher weight to recent adjustments, enabling the system to adapt to gradual changes in user preferences. The volatility of user adjustment behavior is assessed to provide a basis for the adaptive learning rate. Parameter dimensions with high volatility will receive a smaller learning rate, enhancing stability. The learning rate is dynamically adjusted according to parameter volatility and time; a larger learning rate is used in the initial stage for rapid adaptation, gradually decreasing over time to improve stability; simultaneously, a smaller learning rate is used for parameter dimensions with high volatility. Combining the trend direction and the adaptive learning rate, the baseline parameters are smoothly updated, allowing the system to gradually adapt to user preferences without abrupt changes. The updated parameter library is then applied to subsequent decisions, enabling the system to output lighting effects that better match user preferences in the same context.

[0140] For example, when a user is watching a movie, the system automatically switches to movie viewing mode (brightness 40%, color temperature 3000K, RGB(255,210,180)), but the user thinks the light is too bright and manually reduces the brightness to 25%.

[0141] The system detects that the user manually adjusts the brightness (implicit negative feedback) and records the adjusted parameter vector [25%, 3000K, 255, 210, 180];

[0142] The original baseline parameters are [40%, 3000K, 255, 210, 180]. The normalized luminance difference is (0.25-0.40) / 1.0 = -0.15. Other parameters remain unchanged. The deviation vector is ∆V = [-0.15, 0, 0, 0, 0].

[0143] The Euclidean distance is calculated to be 0.15, which is within the preset range [0.05, 0.6], and is therefore determined to be a valid learning sample;

[0144] Update trend vector: , , ;

[0145] Update the fluctuation vector: , ,calculate ;

[0146] Calculate the learning rate: , , , The learning rate for the brightness dimension is 1.95.

[0147] The updated brightness baseline parameter is: 40% + 1.95 × (-0.087) = 39.83%.

[0148] The next time the system recognizes the movie-watching mode, it will use the updated baseline parameters (brightness 39.83%), which are closer to the user's preferences.

[0149] If the user continues to keep the brightness at around 25% during subsequent viewings, the system will gradually reduce the baseline brightness to 28-30% through multiple iterations, achieving adaptive learning based on user preferences.

[0150] This invention achieves adaptive learning of intelligent light bulbs in response to negative user feedback through a dual momentum mechanism. Specifically, when a user is detected as dissatisfied with the current lighting effect and manually adjusts it, the adjusted parameters are first recorded. Then, valid learning samples are filtered using Euclidean distance to avoid abnormal operations interfering with the system. Subsequently, a dual momentum mechanism is employed. On the one hand, first-order momentum captures the long-term trend of user adjustments; on the other hand, second-order momentum assesses parameter volatility, dynamically calculating the adaptive learning rate for each dimension. This allows the system to smoothly and stably adapt to the user's personalized preferences. This invention continuously learns user preferences and can automatically adjust the parameters of preset working modes, reducing the frequency of manual user intervention. Simultaneously, the dual momentum mechanism ensures the stability and convergence of the learning process, preventing system performance degradation due to a single abnormal operation, thus improving the intelligence and adaptability of intelligent light bulb control.

[0151] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or high-voltage switchgear that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or high-voltage switchgear.

[0152] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for controlling an intelligent light bulb based on a target operating mode, characterized in that, Includes the following steps: S1. Based on historical usage data from IoT data acquisition technology, multiple target working modes are preset and parameterized based on historical usage data. Each mode consists of a mode feature vector and a corresponding multi-dimensional lighting parameter vector, including but not limited to reading mode, movie-watching mode, working mode, sleep mode, and party mode. The environmental and behavioral conditions that trigger the feature vector of each working mode are defined, as well as the corresponding lighting parameter output set. S2. Through multiple sensors deployed on the smart bulb itself, raw environmental data and user behavior data are collected in real time. After the raw data is initially filtered and feature extracted at the local node, it is transmitted in real time to the smart home central gateway for data fusion through short-range wireless communication technology to form a real-time feature data stream. S3. Based on the decision tree, perform coarse classification of the real-time feature data stream, use the time-series pattern recognition algorithm to analyze the coarse classification results, combine the historical feature sequence of the specified time window to perform fine pattern recognition, and output the current most likely pattern and its confidence level. S4. Based on the identified target working mode, obtain the corresponding lighting parameter output set from the preset mode parameter library; automatically adjust the output of the smart bulb according to the obtained lighting parameters to match the light with the target working mode. S5. Provide user feedback options based on the control interface, allowing users to manually confirm or correct the current working mode identified by the system; record user feedback data into the system, and update the parameters of the decision tree and time-series pattern recognition model in real time based on user feedback; generate a personalized pattern parameter library for each user based on user feedback and behavioral habits.

2. The intelligent bulb lamp control method based on the target working mode according to claim 1, characterized in that, In S3, the most likely pattern and its confidence level are output, including the following steps: Set a confidence threshold and a minimum duration to evaluate the confidence of the identified new patterns; A mode switch is triggered when the confidence level of the new mode is greater than the confidence threshold and the prediction continues for more than the minimum duration. During switching, a non-linear gradient is used to smoothly transition from the current value to the target value, ensuring a smooth user experience. During the smoothing of lighting parameters, real-time environmental data and user behavior data are continuously monitored; if a clear negative feedback is detected or a sudden change in the situation occurs, the current gradual change process is immediately interrupted, and pattern recognition and decision-making are re-performed based on the latest data.

3. The intelligent bulb lamp control method based on the target working mode according to claim 2, characterized in that, The process of smoothly transitioning from the current value to the target value using a non-linear gradual change includes the following steps: Calculate the perception deviation index based on the current value and the target value; The type of scene to switch is determined based on the perception deviation index; When the scene switching type is dark switching scene, a positive support type compensation signal is generated so that the actual brightness driving trajectory is temporarily maintained at a level higher than the target set brightness in the early stage of switching, and slowly decays to approach the target set brightness as time goes by according to the duration characteristics of dark adaptation, so as to form a trailing transition. When the scene switching type is bright switching scene, a reverse suppression type compensation signal is generated, so that the actual brightness driving trajectory is limited to a level lower than the target set brightness in the early stage of switching, and gradually rises to the target set brightness according to the brightness adaptation duration characteristics over time, so as to form a soft start transition. Using the actual driving brightness value as the main control variable, the first color temperature change curve is calculated in real time using the human visual psychological comfort model; The total range of color temperature variation is divided into multiple color tolerance ranges; Based on the McAdam ellipse principle, the sensitivity of the human eye to color changes in different color temperature ranges is evaluated. Based on the sensitivity-rate of change comparison table, determine the rate of change for different color tolerance ranges and plot the second color temperature change curve; The first color temperature change curve and the second color temperature change curve are weighted and fused to obtain the actual color temperature driving trajectory.

4. The intelligent bulb lamp control method based on the target working mode according to claim 2, characterized in that, If a clear negative feedback is detected or a sudden change in the situation occurs, the current gradual change process is immediately interrupted, and pattern recognition and decision-making are re-performed based on the latest data, including: Obtain the target lighting parameter vector adjusted by the user after negative feedback; Calculate the deviation vector between the target illumination parameter vector and the reference parameter vector of the current target operating mode; The Euclidean distance is calculated based on the deviation vector. When the Euclidean distance falls within the preset target range, the corresponding target lighting parameter vector is used as a valid learning sample; otherwise, it is discarded as an outlier. Based on the dual momentum mechanism, the baseline parameter vector of the current target working mode is updated according to the deviation vector, the user's historical adjustment trend, the fluctuation energy of the user's historical adjustment amplitude, and the baseline parameter vector of the current target working mode. Pattern recognition and decision-making are based on the updated pattern parameter library.

5. The intelligent bulb lamp control method based on the target working mode according to claim 2, characterized in that, Continuously monitor real-time environmental data and user behavior data, including the following steps: During the target working mode recognition process, if it is detected that the user has left the light coverage area for a long time, exceeding the set threshold time, the bulb will automatically switch to energy-saving mode, reduce the brightness to the minimum or turn off the light. The bulb brightness is automatically adjusted according to the ambient light intensity. When there is sufficient natural light, the bulb brightness output is reduced. Based on the incremental data of each manual adjustment by the user, the corresponding lighting parameter output set is updated online, and the preset mode parameter library is optimized to make the lighting control more in line with the user's personalized needs. Historical contextual data is analyzed periodically using unsupervised clustering algorithms to identify user-defined behavioral patterns that are not covered by preset patterns, and new custom target working patterns are created accordingly.

6. The intelligent bulb lamp control method based on the target working mode according to claim 1, characterized in that, In S2, preliminary filtering and feature extraction are performed on the raw data at the local node, including the following steps: Edge processing is performed by a built-in lightweight microprocessor on various sensors deployed on the smart bulb body; Perform moving average filtering, median filtering, or Kalman filtering on the raw sensor signal to eliminate random noise and transient interference pulses; The illuminance change slope and current color temperature value are extracted from the ambient light data; the number of targets, the main target's movement speed, the main target's dwell time in the preset area, and the micro-motion frequency are extracted from the millimeter-wave radar data. Extract confidence scores, ambient noise levels, and whether voice command keywords trigger specific sound sources from audio features; extract posture classification confidence and activity intensity index from visual features; The extracted feature values, along with the data acquisition timestamp, device ID, and sensor type identifier, are encapsulated into a data packet; and the data packet is transmitted to the smart home central gateway via near-field wireless communication technology.

7. The intelligent bulb lamp control method based on the target working mode according to claim 6, characterized in that, Transmitting data packets to the smart home hub gateway using short-range wireless communication technology includes the following steps: Different transmission priorities are set for different types of feature data, including: level 1, level 2, level 3 and level 4; Establish a feature-priority mapping table. When extracting real-time feature values, query the pattern feature vector and the priority mapping table to generate data packets with priority labels. Data transmission is performed according to the transmission priority of the data packet, and wireless channel congestion is continuously monitored; When the channel load is high, the transmission frequency of the third and fourth level data will be automatically reduced or the fourth level data will be temporarily suspended to ensure the smooth transmission of the first and second level critical data.

8. The intelligent bulb lamp control method based on the target working mode according to claim 1, characterized in that, In S3, a coarse classification of the real-time feature data stream is performed based on a decision tree, and the coarse classification results are analyzed using a time-series pattern recognition algorithm, including the following steps: Based on historical usage data, select features that have an impact on pattern recognition; Label historical data to identify pattern instances, and then use the labeled historical data to train a decision tree model. A decision tree is generated from the training data. Each node represents a judgment condition for a feature, each branch represents a range of feature values, and each leaf node represents a coarse classification result. The real-time feature data stream is input into the decision tree model. Starting from the root node, the model moves down the branches of the decision tree according to the current feature value until it reaches the leaf node, and outputs the coarse classification result.

9. The intelligent bulb lamp control method based on the target working mode according to claim 8, characterized in that, In S3, fine-grained pattern recognition is performed by combining historical feature sequences within a specified time window, including the following steps: Select a specified time window to extract the historical feature sequence of the specified time window from the real-time feature data stream; The historical feature sequence is input into the time series pattern recognition model. Based on the input time series features and the trained pattern recognition logic, the model outputs the most likely pattern and its confidence level. The coarse classification results of the decision tree and the fine classification results of the temporal pattern recognition are comprehensively judged. If the two results are consistent, the confidence score is increased; if the results are inconsistent, the final judgment is made based on the confidence score. The system combines confidence level and minimum duration to determine whether to trigger mode switching; if the confidence level of the current mode exceeds the threshold and the duration exceeds the minimum duration, then mode switching is triggered.

10. The intelligent bulb lamp control method based on the target working mode according to claim 9, characterized in that, The comprehensive evaluation of the coarse classification results from the decision tree and the fine classification results from temporal pattern recognition includes the following steps: Obtain coarse classification results and fine classification results. The coarse classification results include: the initial pattern classification result and its confidence level; the fine classification results include: the fine classification result and its confidence level. Compare the coarse classification results of the decision tree with the fine classification results of the temporal pattern recognition to determine whether the two are consistent; If they match, they are the same pattern, and the confidence scores of the two are weighted and averaged to obtain a higher confidence score. The overall confidence score is then compared with a preset confidence score threshold. If the overall confidence score exceeds the threshold, the current pattern is confirmed as the pattern, and the final result is output. If they are inconsistent, the confidence levels of the two are compared, and the result with the higher confidence level is selected as the final judgment. The selected confidence level is then compared with the preset confidence level threshold. If the confidence level exceeds the threshold, the current mode is confirmed as the mode, and the final result is output.

Images