An AI Perception-Based All-DC Lighting Method and System

Abstract

The present invention discloses an all-DC lighting method and system device based on AI perception, belonging to the field of lighting control. The method includes: collecting environmental data by deploying different types of sensors, and sending the environmental data collected by the different types of sensors to an edge gateway; preprocessing the received environmental data, and performing fusion processing on the preprocessed environmental data to obtain a fused environmental state vector, and sending the environmental state vector to the cloud; the cloud calculates an optimal dimming current through a dynamically optimized target dimming strategy deployed in the cloud; the optimal dimming current is sent down to the terminal execution layer, and a dimming instruction is generated according to the received optimal dimming current to adjust the state of the lighting fixture. The present invention solves the problem that existing lighting control strategies are difficult to simultaneously meet multiple requirements.

Description

Technical Field

[0001] The present invention relates to the field of lighting control, and particularly to an all-dc lighting method and system based on AI perception. Background Art

[0002] In the prior art, the following problems exist in the environmental monitoring and lighting control system: the environmental data collected by a single sensor has low accuracy and cannot comprehensively reflect the actual situation of a complex environment; the lighting control strategy usually only considers a single goal such as energy saving or comfort, and it is difficult to simultaneously meet multiple requirements such as comfort, energy saving, and stability; at the same time, there are also factors such as line voltage drop during actual execution, resulting in low execution accuracy. Summary of the Invention

[0003] One object of the present invention is to provide an all-dc lighting method based on AI perception, so as to solve the problem in the prior art that the lighting control strategy usually only considers a single goal such as energy saving or comfort, and it is difficult to simultaneously meet multiple requirements such as comfort, energy saving, and stability, and it is impossible to achieve precise control of lamp dimming.

[0004] The present invention is realized by the following technical solutions. An all-dc lighting method based on AI perception includes the following steps: S100, collecting environmental data by deploying different types of sensors, and sending the environmental data collected by the different types of sensors to an edge gateway; S200, the edge gateway preprocesses the received environmental data, and performs fusion processing on the preprocessed environmental data to obtain a fused environmental state vector, and sends the environmental state vector to the cloud; S300, the cloud calculates an optimal dimming current according to the received environmental state vector through a dynamically optimized target dimming strategy deployed in the cloud; S400, the cloud sends the optimal dimming current to a terminal execution layer, and the terminal execution layer generates a dimming instruction according to the received optimal dimming current to adjust the state of the lamp.

[0005] Further, the all-dc lighting method further includes step S500, by embedding an edge execution strategy in a PI controller of the terminal execution layer, and controlling the PI controller to compensate for line voltage drop through the edge execution strategy, and correcting the PWM duty cycle in actual execution, so that the current is as close as possible to the optimal current under the condition of fluctuations caused by factors such as line impedance and voltage drop.

[0006] Further, the different types of sensors include:

[0007] a light sensor, a presence sensor, a temperature sensor, and a color temperature / spectrum sensor.

[0008] Further, a light sensor is used to detect the ambient light intensity, distinguish the contributions of natural light and artificial light sources, and support dynamic dimming. One light sensor can be deployed per 10 - 20 square meters to prevent light shadow errors. A presence sensor is used to detect the activities of people in the target area, distinguish whether people are present, and the detection threshold can be set according to the specific scenario requirements of the target area. A temperature sensor is used to monitor the ambient temperature in real time to assist the lighting system in optimizing energy consumption. A color temperature / spectrum sensor is used to detect the color temperature and spectral distribution of the ambient light to guide the lighting system in optimizing the light color matching.

[0009] Further, when the deployment area is indoors, indoor positioning sensors can also be arranged to track the positions and movement trajectories of people through BLE / UWB technology, so as to achieve follow - up lighting or scene linkage.

[0010] Further, the pre - processing in step S200 includes: Denoising: Using a Kalman filter or a moving window mean to process the transient noise of the sensors, identifying the abnormal data of the sensors based on statistical methods, and removing the outliers; Timestamp synchronization: After denoising, synchronize the timestamps of the data collected by different types of sensors.

[0011] Further, the timestamp synchronization can adopt hardware - level synchronization or software - level interpolation synchronization.

[0012] Further, the fusion processing of the pre - processed environmental data includes:

[0013] Fusing the environmental data through a fusion algorithm based on Bayesian inference. The fusion algorithm based on Bayesian inference includes the following steps: S210. Based on the pre - processed sensor data, define an environmental state vector according to different types of sensors to describe the real environment collected by the sensors at a certain moment, and construct an observation model for each type of sensor; S220. Align the timestamp differences of the environmental data of different types of sensors through spatio - temporal alignment, and then perform weighted averaging according to the data of the light sensor group to obtain a comprehensive observation value. The spatio - temporal alignment aligns the environmental data with different sampling frequencies so that different environmental data have data values at the same time point. The spatio - temporal alignment can be achieved through linear interpolation methods. By given a sensor Z i , sample and record the data value Z k at different timestamps t i (t k ), and align these data to any time point t; S230. Fuse the environmental data from different types of sensors through Bayes' theorem, and combine the observation values of different sensor observation models and the previous environmental state vector X t-1 to calculate the current environmental state vector X t .

[0014] Furthermore, the environmental state vector can be represented by the following formula:

[0015] , where is the environmental state vector, is the light intensity, Lux; is the temperature, °C; is the human presence, binary: 0 means no one, 1 means someone; is the light color interference coefficient, dimensionless, used to describe the deviation degree of the light source color temperature; the observation model of the light sensor is shown as the following formula:

[0016] , where is the observation value of the light sensor, is the true light intensity, is the light measurement error, indicates that the observation error of the light sensor follows a normal distribution with a mean of 0, is the noise variance of the light sensor; the observation model of the temperature sensor is shown as the following formula:

[0017] , where is the observation value of the temperature sensor, is the true temperature, is the temperature measurement error, indicates that the temperature measurement error follows a normal distribution with a mean of 0 and a variance of , is the noise variance of the temperature sensor; the observation model of the human presence sensor is shown as the following formula:

[0018] , where is the observation value of the human presence sensor, indicating whether a person is detected by the sensor, is the true human presence state, is the misdetection error, indicating that the sensor misdetects one state as another state, is the exclusive OR operation, used to represent the misdetection situation, that is, if and are inconsistent, an incorrect observation value will be returned; the observation model of the color temperature sensor is shown as the following formula:

[0019] , where is the observation value of the color temperature sensor, is the true light color interference coefficient, indicating the actual light color interference degree at time t, is the color temperature measurement error. ​​

[0020] Furthermore, the aligned data can be calculated by the following formula:

[0021] , where is the aligned data after interpolation, is the observation value of sensor i at timestamp t k , t k is the timestamp of the original sensor data, t k+1 and t k are two adjacent timestamps, is the weight of linear interpolation, representing the relative distance of time point t from timestamps t k and t k+1 . This weight is used for weighted averaging of sensor data.

[0022] Furthermore, the fusion algorithm based on Bayesian inference can also include a dynamic weight adjustment strategy for sensors. When the observation value of a sensor differs significantly from the previous environmental state vector estimate, it indicates that there may be a sudden anomaly or an abnormal measurement value. At this time, the weight will be reduced to reduce the influence of this sensor on the final state estimate, making the system more robust; conversely, if the difference between the sensor's observation value and the state estimate is small, it means that the sensor data is relatively reliable, and the weight will be increased to enhance the role of this sensor in state estimation. The weight can be calculated by the following formula:

[0023] , where is the updated weight at time t, is the variance related to the j-th variable, λ is the sensitivity coefficient, is the observation value measured by the j-th sensor or observation model at time t, representing the current measurement data regarding the j-th variable, is the previous estimate of the j-th variable at time t−1. This estimate is usually a prediction or estimation based on past data.

[0024] Furthermore, the dimming instruction maps the optimal dimming current to a duty cycle between 0 and 1, and controls the states such as brightness and color temperature of the lamp through this duty cycle.

[0025] Furthermore, the dimming instruction realizes the mapping of the optimal dimming current through the following formula:

[0026] ,

[0027] where is the pulse width modulation duty cycle, is the optimal dimming current.

[0028] Furthermore, the goal of edge execution strategy control is to keep the current of the lamp close to the optimal dimming current and compensate for the current errors caused by line voltage drop, hardware characteristics, and load changes. Through the proportional and integral control of the PI controller, it can quickly respond and eliminate long-term errors, and gradually adjust the actual current to the target current.

[0029] Furthermore, the edge execution strategy can be expressed by the following formula:

[0030] , where is the actual pulse width modulation duty cycle after being adjusted by the PI controller, is the target pulse width modulation duty cycle obtained by mapping, is the current error, is the proportional gain of the PI controller, is the accumulation of the error function over time, is the integral variable, is the integral gain of the PI controller.

[0031] Furthermore, the current error is calculated by the following formula: , where is the optimal dimming current, is the actual measured current.

[0032] Furthermore, the dynamic optimization target dimming strategy includes an objective function and dynamic constraint conditions.

[0033] Furthermore, the objective function of the dynamic optimization target dimming strategy includes: a comfort term, a stability term, and an energy-saving term. Among them, the comfort term is related to the actual illuminance and the target illuminance. The actual illuminance is controlled by the driving current and is related to the ambient light intensity. The energy-saving term is related to the system power consumption, and the power consumption is a function of the driving current. The stability term is related to the current change rate, and the goal is to limit the rapid change of the current. The comfort term, the stability term, and the energy-saving term together form a whole, and the comprehensive balance of comfort, energy saving, and stability is achieved by optimizing the current; the dynamic constraint conditions of the dynamic optimization target dimming strategy are used to ensure that the dimming strategy meets the physical limitations in actual applications while ensuring the stability of the system and the comfort of users, including: current constraint, illuminance deviation constraint, and stability derivative constraint. Among them, the current constraint is used to ensure that the current is between the minimum and maximum values of the current allowed by the lamp, so that the current value does not exceed the maximum limit that the device can withstand and does not fall below the minimum operating current of the device; the illuminance deviation constraint is used to ensure that the illuminance of the target area is as close as possible to the predetermined target illuminance to provide the best comfort experience; the stability derivative constraint ensures the smooth operation of the system by setting the maximum limit of the current change rate and limiting the current change rate.

[0034] Further, the objective function is expressed by the following formula:

[0035] , where J is the objective function, representing the weighted sum of trade - offs among three objectives from time t0 to t. is the weight coefficient of the comfort term. is the weight coefficient of the energy - saving term. is the weight coefficient of the stability term. is the actual illuminance. is the target illuminance. is the comfort term. is the energy - saving term. is the power consumption. is the stability term, used to describe the smoothness of the current change rate.

[0036] Further, the actual illuminance can be expressed by the following formula: , where in the formula, is the light efficiency of the lighting system. is the ambient light intensity, that is, the light intensity in the probability distribution of the ambient state obtained by Bayesian inference.

[0037] Further, the power consumption is proportional to the square of the current and can be expressed by the following formula: , where R LED is the resistance of the lamp, which together with the square of the current determines the power consumption. is the stability term, which is used to describe the smoothness of the current change rate. To avoid sudden dimming instructions, the change rate of the current needs to be kept smooth. Especially during the dimming process, sudden current changes may cause the lights to flicker or become unstable.

[0038] Further, the current constraint is expressed by the following formula: , is the minimum current allowed for the lamp. is the maximum current allowed for the lamp; the illuminance deviation constraint is expressed by the following formula: , is the maximum allowable illuminance deviation; the stability derivative constraint is expressed by the following formula: , is the maximum limit of the current change rate.

[0039] On the other hand, the present invention provides an AI - aware all - DC lighting system, which includes a processor and a memory. A computer program is stored in the memory. When the computer program is executed by the processor, the above - described AI - aware all - DC lighting method is implemented.

[0040] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0041] 1. The present invention can effectively improve the accuracy and comprehensiveness of environmental monitoring by arranging multiple heterogeneous sensors in the target area and fusing the heterogeneous sensor data using a two-layer fusion framework based on Bayesian inference and dynamic weights, thereby overcoming the shortcoming of low accuracy of data collected by a single sensor.

[0042] 2. The present invention deploys a dynamic optimization target dimming strategy in the cloud, and calculates the optimal dimming current based on the fused environmental state probability distribution. This strategy comprehensively balances the comfort, energy saving and stability of the lighting system, solving the problem that existing lighting control strategies are difficult to meet multiple needs at the same time.

[0043] 3. This solution achieves high-precision and high-efficiency environmental monitoring and lighting control through the synergy of multi-sensor fusion, cloud-based dynamic dimming optimization, and terminal edge execution strategies. This meets the multiple needs of lighting systems in complex environments and demonstrates strong practical value. Furthermore, by integrating edge execution strategies into the terminal execution layer and using a PI controller to compensate for interference factors such as line voltage drop and hardware nonlinearity, the actual executed current is kept as close to the optimal current as possible, improving the accuracy of dimming control and overcoming the shortcomings of existing technologies that often result in lower execution accuracy due to factors such as line voltage drop. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0045] Figure 1 This is a flow chart of the method provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0046] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0047] Example 1

[0048] In the existing process of area dynamic dimming, environmental data collected by a single sensor is usually adopted, and dimming is carried out according to the data of a single sensor, resulting in low dimming accuracy, being unable to comprehensively reflect the actual situation of a complex environment, and being difficult to provide accurate environmental state information for lighting control. Moreover, the data of a single sensor will lead to the output dimming control strategy usually only considering a single goal such as energy saving or comfort, being difficult to simultaneously meet multiple requirements such as comfort, energy saving and stability, and being unable to achieve the comprehensive optimization of the lighting system.

[0049] In view of the above problems, this embodiment discloses an all-dc lighting method based on AI perception. By adopting a variety of heterogeneous sensors to collect environmental data and processing the heterogeneous data through a data fusion algorithm, the accuracy of environmental monitoring is improved. In this embodiment, a dynamic optimization target dimming strategy is deployed in the cloud, and the optimal dimming current is calculated according to the probability distribution of the fused environmental state, realizing the comprehensive balance of the comfort, energy saving and stability of the lighting system. An edge execution strategy is built in at the terminal execution layer. By compensating factors such as line voltage drop through a PI controller, the actually executed current is made as close as possible to the optimal current, improving the accuracy of dimming control. The solution disclosed in this embodiment realizes high-precision and high-efficiency environmental monitoring and lighting control through the cooperation of multi-sensor fusion, cloud dynamic optimization dimming strategy and terminal edge execution strategy, and meets the multiple requirements of the lighting system in a complex environment.

[0050] Figure 1 The overall flowchart of the all-dc lighting method based on AI perception in this embodiment is shown. It can be seen from Figure 1 that this embodiment includes the following steps:

[0051] Step 1: Arrange a variety of sensors in the target area for collecting environmental data, and send the environmental data collected by various types of sensors to the edge gateway.

[0052] After receiving the sensor data, the edge gateway first filters the noise of the sensor data. After noise reduction, the heterogeneous data collected by a variety of sensors is fused and processed to obtain fused data and send the fused data to the cloud.

[0053] Specifically, the Kalman filter or the moving window mean can be used to process the transient noise of the light and temperature sensors for noise filtering of the sensor data. At the same time, abnormal data (such as mis-triggering of the human body sensor) is identified based on statistical methods to eliminate outliers. After noise reduction is completed, timestamp synchronization is performed on the data collected by different sensors. Timestamp synchronization can adopt hardware-level synchronization (such as using GPS / PPS pulses) or software-level interpolation synchronization (linear interpolation for low-sampling-rate sensors).

[0054] Specifically, in this embodiment, a light sensor, a human sensor, a temperature sensor, and a color temperature / spectrum sensor are arranged in the target area.

[0055] Among them, the light sensor is used to detect the ambient light intensity (unit: Lux), distinguish the contribution of natural light and artificial light, and support dynamic dimming; one can be deployed every 10 to 20 square meters to prevent light and shadow errors.

[0056] The human presence sensor is used to detect the activities of people (moving, stationary) in the target area and distinguish whether people are present. The detection threshold can be set according to the specific scenario requirements of the target area (for example, high sensitivity is required in corridors to prevent missed detections).

[0057] Temperature sensors are used to monitor ambient temperature in real time and assist lighting systems in optimizing energy consumption.

[0058] Color temperature / spectrum sensors are used to detect the color temperature (unit: K) and spectral distribution of ambient light, guiding the lighting system to optimize light color matching (for example, art exhibition halls, medical operating tables, and other areas require specific ambient light color temperatures).

[0059] In addition, if the target area is indoors, indoor positioning sensors can also be deployed to track the location and movement trajectory of people through BLE / UWB technology, thereby realizing follow-up lighting or scene linkage.

[0060] Specifically, heterogeneous data can be fused using a fusion algorithm based on Bayesian inference. In this embodiment, an FPGA-based multi-type heterogeneous sensor data fusion algorithm is built into the edge gateway. This algorithm is a two-layer fusion framework based on Bayesian inference and dynamic weighting, which is used to solve the spatiotemporal alignment problem of heterogeneous sensor data while balancing the accuracy and real-time performance of heterogeneous sensor data.

[0061] The heterogeneous sensor data fusion algorithm includes the following:

[0062] 1) First, based on the noise reduction preprocessed sensor data, we define the environment state vector, which describes the real environment captured by the sensor at a certain time t. We then construct observation models for different sensor types.

[0063] Specifically, in this embodiment, the environment state vector can be expressed by the following formula:

[0064] ,

[0065] in, is the environment state vector, which is the aforementioned fusion data, is the light intensity, Lux; is the temperature, °C; It is for the presence of personnel, binary value: 0 (no person) / 1 (person present); It is the light color interference coefficient, dimensionless, and is used to describe the deviation degree of the color temperature of the light source.

[0066] In this embodiment, the observation models of four types of sensors are exemplarily given:

[0067] The observation model of the light sensor is shown in the following formula:

[0068] ,

[0069] where, is the observed value of the light sensor, is the true light intensity, is the light measurement error, indicates that the observation error of the light sensor follows a normal distribution, with a mean of 0 and a variance of , which means that the error of the sensor is random, and most of the errors will be concentrated near 0 (i.e., the measured value is close to the true value), but there may also be some larger errors. is the noise variance of the light sensor.

[0070] The observation model of the temperature sensor is shown in the following formula:

[0071] ,

[0072] where, is the observed value of the temperature sensor, is the true temperature, is the temperature measurement error, indicates that the temperature measurement error follows a normal distribution, with a mean of 0 and a variance of , is the noise variance of the temperature sensor.

[0073] The observation model of the person-in sensor is shown in the following formula:

[0074] ,

[0075] where, is the observed value of the person-in sensor, indicating whether a person is detected by the sensor. This value can be 0 or 1, where 0 means no person and 1 means person present. is the true personnel presence status, indicating whether there is a person in the environment at time t, where 1 means yes and 0 means no. is the false detection error, indicating that the sensor wrongly detects one state (such as no person) as another state (such as person present). This error is related to the detection accuracy of the sensor and is usually expressed as the false detection rate. is an exclusive OR operation, used to represent the situation of false detection, that is, if and are inconsistent (false detection occurs), an incorrect observation value will be returned. is the false detection rate, that is, the probability of the sensor misjudging. For example, the false detection rate means that the sensor has a 10% probability of misjudging the unmanned state as occupied.

[0076] The observation model of the color temperature sensor is shown as follows:

[0077] ,

[0078] where is the observation value of the color temperature sensor, that is, the light color interference coefficient measured by the sensor, which reflects the color temperature deviation of the light source and is determined by the true light color interference coefficient and the measurement error. is the true light color interference coefficient, representing the actual light color interference degree at time t. It describes the deviation of the light source color. is the color temperature measurement error, representing the noise of the color temperature sensor. This noise follows a uniform distribution represents that the color temperature measurement error is uniformly distributed in the interval inside. is the maximum amplitude of the color temperature error, used to control the error range. A larger a represents a larger measurement error of the color temperature sensor, and a smaller a represents a smaller error.

[0079] It should be noted that in the exemplary observation model disclosed in this embodiment, both the light sensor and the temperature sensor adopt the Gaussian noise model to reflect their measurement error distributions. represents a Gaussian distribution, that is, a normal distribution. and in, 0 is the mean of this Gaussian distribution, representing the central value of this distribution. The mean being 0 means that the expected value of the error between the observation value and the true value is 0, that is, the error does not deviate in the positive or negative direction. and are the variances of the Gaussian distribution, representing the volatility or uncertainty of the error. The larger the variance, the more unstable the error and the larger the fluctuation range; the smaller the variance, the smaller the error and the smaller the fluctuation range.

[0080] 2) Process the timestamp differences of the environmental data collected by different types of sensors through spatio-temporal alignment, and then perform weighted averaging based on the data of the light sensor group to obtain a comprehensive observation value.

[0081] Specifically, spatio-temporal alignment aligns environmental data with different sampling frequencies so that different environmental data have data values at the same time point. In this embodiment, spatio-temporal alignment of sensor data can be exemplarily achieved by linear interpolation. By giving a sensor Z i , sampling and recording the data value Z k at different timestamps t i (t k ), and aligning this data to an arbitrary time point t. Specifically, it can be expressed by the following formula:

[0082] ,

[0083] where, is the aligned data after interpolation, representing the estimated value at time t. is the observed value of sensor i at timestamp t k , t k is the timestamp of the original sensor data, representing the sampling moment. t k+1 and t k are two adjacent timestamps, is the weight of linear interpolation, representing the relative distance of time point t from timestamps t k and t k+1 , and this weight is used for weighted averaging of sensor data.

[0084] Specifically, in this embodiment, weighted averaging can also be performed according to the data of the light sensor group by fusing the observed values of multiple light sensors into a comprehensive value. Let the observed value of each sensor be given a weight , and this weight reflects the position or importance of the sensor. Sensors closer to the working area will have a higher weight, which can be expressed by the following formula:

[0085] ,

[0086] where, is the fused light value, representing the weighted average light intensity at time t. is the weight of the j-th sensor, and this weight is usually related to the position or importance of the sensor. is the observed value of the j-th light sensor at time t. is the number of light sensors.

[0087] 3) Finally, fuse the data from different sensors through Bayes' theorem, and estimate the current environmental state vector X t-1 by combining the observed values calculated from the observation models of different sensors and the previous environmental state vector X t。

[0088] Specifically, the multi-modal Bayesian fusion of this embodiment can be expressed by the following formula:

[0089]

[0090] ,

[0091] where is the posterior probability of estimating the current environmental state vector given all the observed data (illumination, temperature, human presence, color temperature).

[0092] is the likelihood function of the light sensor, representing the probability of the observed value of the light sensor given the true illumination. is the likelihood function of the temperature sensor, representing the probability of the observed value of the temperature sensor given the true temperature. is the likelihood function of the human presence sensor, representing the probability observed by the human presence sensor given the true human presence. is the likelihood function of the color temperature sensor, representing the probability observed by the color temperature sensor given the true color interference coefficient. is the state transition probability, which describes the relationship between the current environmental state vector X t and the environmental state vector X t-1 at the previous moment. In the actual calculation process, the state transition probability can be modeled by a Markov process.

[0093] It should be noted that these likelihood functions are used to describe the relationship between each sensor data and the true environmental state. The specific form of the likelihood function depends on the type of the observed data and the measurement characteristics of the sensor. In this embodiment, for sensors with Gaussian noise, the Gaussian distribution can be used as the likelihood function; for count data, the Poisson distribution can be used, etc. The specific form of the likelihood function can also be determined by methods such as data fitting, literature experience, or Bayesian inference.

[0094] This embodiment also includes a dynamic weight adjustment strategy for sensors. When the observed value of a sensor is significantly different from the estimated value of the previous environmental state vector X t-1 it indicates that there may be a sudden anomaly or an abnormal measured value. At this time, the weight will be reduced to reduce the influence of this sensor on the final state estimation, so as to make the system more robust. On the contrary, if the difference between the observed value of the sensor and the state estimation is small, it means that the sensor data is relatively reliable, and the weight will be increased to enhance the role of this sensor in the state estimation. The weight can be calculated by the following formula:

[0095] ,

[0096] wherein, is the updated weight at time t, representing the weight updated according to the current observation data and the previous estimated value. is the variance associated with the j-th variable, which represents the measurement noise or uncertainty of the variable. Generally, a larger variance means a larger noise in the signal and a lower reliability of the signal. λ is the sensitivity coefficient, which is used to control the relative influence of the error term and the noise term in the model. By adjusting the sensitivity coefficient, the weight of the difference between the observation error and the previous estimated value in the formula can be updated. A larger sensitivity coefficient will make the weight more affected by this difference, meaning that the difference between the previous estimated value and the current observed value has a greater impact on the weight update. is the observed value measured by the j-th sensor or observation model at time t, representing the current measurement data regarding the j-th variable, is the previous estimated value of the j-th variable at time t - 1, which is usually a prediction or estimation based on past data.

[0097] is the absolute difference between the current observed value and the previous estimated value of the j-th variable, which is used to measure the deviation between the observed value and the predicted value. If this difference is large, it means that the gap between the current observed value and the previous estimated value is large, which may indicate a large measurement error or a large environmental change, thus affecting the update of the weight.

[0098] It should be noted that the dynamic weight adjustment strategy provides an adaptive mechanism for the multi-type sensor data fusion algorithm, enabling it to adjust the weight in real time according to the performance of the sensors and the changes in the observation data, thereby improving the accuracy and robustness of the state estimation. By introducing the sensitivity coefficient, the response intensity to sudden anomalies can be flexibly controlled to ensure that the system can make reasonable state estimations in the face of different sensor qualities and abnormal situations.

[0099] Step 2: The cloud receives the fused data sent by the edge gateway. In this embodiment, the fused data is the processed environmental state probability distribution calculated by the fusion algorithm arranged in the edge gateway.

[0100] A dynamic optimization target dimming strategy is deployed in the cloud, and this strategy calculates the optimal dimming current according to the received environmental state probability distribution.

[0101] Specifically, in this embodiment, the dynamic optimization target dimming strategy includes an objective function and dynamic constraint conditions.

[0102] The objective function is used to comprehensively balance the comfort, energy efficiency, and stability of the lighting system. By summing the trade-offs between comfort, energy efficiency, and stability over a period of time, the optimal dimming strategy is achieved by combining these three requirements.

[0103] In this embodiment, the objective function can be expressed by the following formula:

[0104] ,

[0105] Among them, J is the objective function, which represents the sum of the trade-offs of the three objectives in the time period t0 to t. is the weight coefficient of the comfort term, is the weight coefficient of the energy-saving item, is the weight coefficient of the stability term.

[0106] is the actual illumination, which is controlled by the driving current and is related to the ambient light intensity. The target illumination is set according to the actual application requirements and can be set by the user across fields or automatically adjusted according to the environment. The comfort term represents the difference between the actual illuminance and the target illuminance. The goal is to minimize this difference and ensure that the indoor lighting meets the predetermined comfort standard as much as possible. The squared difference is used to emphasize that large deviations are more unfavorable than small deviations, thereby more strongly penalizing large deviations between the actual and target illuminance.

[0107] In this embodiment, the actual illumination can be expressed by the following formula:

[0108] ,

[0109] Where, is the light efficiency of the lighting system, is the ambient light intensity, that is, the light intensity in the probability distribution of the environmental state obtained by Bayesian inference.

[0110] For energy saving, The purpose of this item is to minimize the power consumption and thus achieve energy saving. With current It is proportional to the square of , which can be expressed by the following formula:

[0111] , where R LED is the resistance of the lamp, which together with the square of the current determines the power consumption.

[0112] is the stability term, which is used to describe the smoothness of the current change rate. To avoid sudden dimming instructions, the change rate of the current needs to be kept smooth. Especially during the dimming process, sudden current changes may cause the lights to flicker or become unstable.

[0113] It should be noted that in the objective function, the comfort term is related to the actual illuminance and the target illuminance. The actual illuminance is controlled by the driving current and is related to the ambient light intensity. The energy-saving term is related to the system power consumption, and the power consumption is a function of the driving current. The stability term is related to the current change rate, and the goal is to limit the rapid change of the current. These three parts together form a whole, and the comprehensive balance of comfort, energy saving, and stability is achieved by optimizing the current.

[0114] The dynamic constraint conditions of this embodiment are used to ensure that the dimming strategy not only conforms to physical limitations in actual applications but also guarantees the stability of the system and the comfort of users. It can include: current constraint, illuminance deviation constraint, and stability derivative constraint.

[0115] Among them, the current constraint can be expressed by the following formula:

[0116] ,

[0117] is the minimum current allowed for the lamp, is the maximum current allowed for the lamp. This constraint is used to ensure that the current is between the minimum and maximum values allowed for the lamp. Thus, it ensures that the current value will not exceed the maximum limit that the device can withstand and will not be lower than the minimum operating current of the device, thereby guaranteeing the safety and stability of the hardware.

[0118] The illuminance deviation constraint can be expressed by the following formula:

[0119] ,

[0120] is the maximum allowable illuminance deviation, which can be determined according to the actual usage situation. This constraint is used to ensure that the illuminance in the target area is as close as possible to the predetermined target illuminance to provide the best comfort experience.

[0121] The stability derivative constraint can be expressed by the following formula:

[0122] ,

[0123] is the maximum limit of the current change rate. During the dimming process, if the current changes too fast, it may cause the lights to flicker or become unstable, affecting the comfort of users and even damaging the lighting equipment. This constraint ensures the smooth operation of the system by limiting the current change rate.

[0124] It should be noted that the dynamic constraint conditions play a crucial role in the entire dimming strategy, ensuring the reliability, stability, and user comfort of the system. By precisely controlling these constraint conditions, the dimming strategy can neither violate the hardware limitations of the system nor provide the best lighting effect in the actual environment.

[0125] Step 3: The cloud sends the calculated optimal dimming current to the terminal execution layer. The terminal execution layer generates a dimming instruction based on the received optimal dimming current and adjusts the states of the lamp, such as brightness and color temperature.

[0126] Specifically, the dimming instruction maps the optimal dimming current to a duty cycle between 0 and 1, and adjusts the states of the lamp, such as brightness and color temperature, through this duty cycle.

[0127] Specifically, the mapping can be performed through the following formula:

[0128] In this embodiment, the optimal dimming current can be mapped through the following formula:

[0129] ,

[0130] where is the duty cycle of PWM (pulse width modulation duty cycle), is the optimal dimming current calculated through the objective function. Through this formula, the instruction of the optimal current can be converted into a PWM signal, and the brightness and color temperature of the lamp can be adjusted through the PWM drive circuit.

[0131] The above formula maps the dimming instruction to the duty cycle and adjusts the brightness of the lamp through the duty cycle. In addition, in this embodiment, the brightness of the lamp can also be adjusted by mapping the light instruction to a voltage of 0 - 10V. 0 - 10V dimming is a common dimming method, usually used to control the brightness and color temperature of LED lamps. By adjusting the voltage signal (between 0V and 10V), the brightness change of the lamp is controlled. 0V represents the lowest brightness (turning off the lamp), and 10V represents the highest brightness. In this embodiment, the mapping relationship between the optimal dimming current and the 0 - 10V voltage is:

[0132] ,

[0133] where is the 0 - 10V dimming voltage signal.

[0134] In this embodiment, considering that in actual engineering, although the PWM duty cycle D(t) is obtained through the above drive current mapping formula, there are still several factors that may cause the brightness of the lamp not to accurately follow the optimal current:

[0135] 1) Line voltage drop. In an actual circuit, especially when the current is large, the resistance of the line and the characteristics of the wire may cause current loss or voltage reduction. This means that the optimal current command cannot always be directly converted into the actual current value. If there is a voltage drop, the actual current may be less than the theoretical current.

[0136] 2) Hardware characteristics and non-linearity. The characteristics of the current source, drive circuit, and lamp may be non-linear and may behave differently at different current or voltage levels, which can lead to inconsistencies between the PWM signal and the actual current.

[0137] 3) Load change. If the load of the lamp changes, or there are other circuits in the system that affect the current of the lamp (such as changes in other loads), it will cause errors in current tracking.

[0138] Based on the above reasons, an edge execution strategy can be built into the PI controller at the terminal execution layer. By means of the edge execution strategy, the PI controller is controlled to compensate for the line voltage drop and correct the PWM duty cycle in actual execution, so that the current is as close as possible to the optimal current under the fluctuations caused by factors such as line impedance and voltage drop.

[0139] In this embodiment, the edge execution strategy can be expressed by the following formula:

[0140] ,

[0141] Where, is the actual PWM duty cycle, which is the actual PWM duty cycle after being adjusted by the PI controller.

[0142] is the target PWM duty cycle obtained by mapping, that is, the PWM duty cycle D(t) obtained through the drive current mapping formula, which can be understood as the ideal brightness of the lamp.

[0143] is the current error, which is the optimal current minus the difference between the actual measured current, which can be expressed by the following formula:

[0144] , where, is the actual measured current, which may be inconsistent with the optimal current due to various factors (such as line voltage drop, load change, etc.). is the proportional gain of the PI controller, which controls the response strength of the proportional part to the current error. If the error is large, the proportional control will provide a large correction amount and quickly adjust the PWM duty cycle to make the actual current quickly approach the expected value. A larger proportional gain will make the system respond faster to the error, but it may also introduce overshoot or oscillation; a smaller proportional gain will make the system respond slower.

[0145] is the integral part, which is the accumulation of the current error over a period of time. The integral term represents the influence of past errors on the control signal. is the accumulation of the error function over time and reflects the error over a past period of time in actual control. is the integral variable, which is a virtual time variable used to represent the accumulation of past errors during the integration process; is the integral gain of the PI controller, which determines the response strength of the integral part to the error accumulation. The integral gain controls the action of the integral part, and its function is to eliminate the steady-state error. That is, if the error persists for a long time, the integrator will gradually accumulate the error, thereby generating a correction in the control quantity to drive the error to gradually decrease.

[0146] It should be noted that the function of the integral part is to eliminate long-existing errors. Especially when there is a steady-state deviation in the system (such as because there are often small voltage drops or drifts in the system), the integral term will accumulate the error over time. The PI controller adjusts the duty cycle according to the time-accumulated error information to reduce the steady-state deviation caused by long-term or continuous errors (such as voltage drops or other system disturbances), and gradually corrects the control signal until the error disappears.

[0147] The specific implementation manners described above further elaborate on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above description is only the specific implementation manners of the present invention and is not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A fully DC lighting method based on AI perception, characterized in that, The full DC lighting method includes: S100. Collect environmental data through the deployment of different types of sensors, and send the environmental data collected by the different types of sensors to the edge gateway. The different types of sensors include: light sensors, presence sensors, temperature sensors, and color temperature / spectrum sensors S200. The edge gateway preprocesses the received environmental data, and performs fusion processing on the preprocessed environmental data to obtain a fused environmental state vector, and sends the environmental state vector to the cloud. S300. The cloud calculates the optimal dimming current through a dynamically optimized target dimming strategy deployed in the cloud according to the received environmental state vector. S400. The cloud sends the optimal dimming current to the terminal execution layer. The terminal execution layer generates a dimming instruction according to the received optimal dimming current to adjust the state of the lamp. The fusion processing of the preprocessed environmental data includes: Fusing the environmental data through a fusion algorithm based on Bayesian inference. The fusion algorithm based on Bayesian inference includes the following steps: S210. Based on the preprocessed sensor data, define an environmental state vector according to different types of sensors to describe the real environment collected by the sensors at a certain moment, and construct an observation model for each type of sensor. S220. Process the timestamp differences of the environmental data of different types of sensors through spatio-temporal alignment, and then perform weighted averaging according to the data of the light sensor group to obtain a comprehensive observation value. Spatio-temporal alignment aligns environmental data with different sampling frequencies so that different environmental data have data values at the same time point. The spatio-temporal alignment is achieved by a linear interpolation method. By giving a sensor Z i , sampling and recording the data value Z k at different timestamps t i (t k ), these data are aligned to any time point t; S230. Fuse environmental data from different types of sensors through Bayes' theorem, combine the observations of different sensor observation models and the previous environmental state vector X t-1 to calculate the current environmental state vector X t ; The environmental state vector is represented by the following formula: ， Among them, is the environmental state vector, is the light intensity, Lux; is the temperature, °C; is the human presence, binary: 0 means no one, 1 means someone; is the light color interference coefficient, dimensionless, used to describe the deviation degree of the light source color temperature; The observation model of the light sensor is shown as the following formula: ， Among them, is the observed value of the light sensor, is the true light intensity, is the light measurement error, indicates that the observation error of the light sensor follows a normal distribution with a mean of 0, is the noise variance of the light sensor; The observation model of the temperature sensor is shown as the following formula: ， wherein, is the observed value of the temperature sensor, is the true temperature, is the temperature measurement error, indicates that the temperature measurement error follows a normal distribution with a mean of 0 and a variance of , is the noise variance of the temperature sensor; The observation model of the presence sensor is shown as the following formula: ， Among them, is the observed value of the human presence sensor, indicating whether the person measured by the sensor exists, is the true human presence status, is the false detection error, indicating that the sensor misdetects a status as another status, is the exclusive OR operation, used to represent the false detection situation, that is, if and are inconsistent, an incorrect observed value will be returned; The observation model of the color temperature / spectrum sensor is shown as the following formula: ， Among them, is the observed value of the color temperature / spectrum sensor, is the true light color interference coefficient, representing the actual light color interference degree at time t, is the color temperature / spectrum measurement error; The fusion algorithm based on Bayesian inference further includes: A sensor dynamic weight adjustment strategy, which determines the reliability of sensor data and its impact on the final state estimation according to the difference between the sensor observation value and the previous environmental state vector estimation. When the difference is large, the weight of the sensor is reduced. When the difference is small, the weight of the sensor is increased. The weight is calculated through the following formula: ， Among them, is the updated weight at time t, is the variance related to the j-th variable, and λ is the sensitivity coefficient. is the observed value measured by the j-th sensor or observation model at time t, representing the current measurement data regarding the j-th variable. is the previous estimated value of the j-th variable at time t - 1.

2. The full DC lighting method based on AI perception according to claim 1, wherein The full DC lighting method also includes step S500. By embedding an edge execution strategy in the PI controller of the terminal execution layer, the edge execution strategy is used to control the PI controller to compensate for the line voltage drop and correct the duty cycle of the pulse width modulation in actual execution.

3. The full DC lighting method based on AI perception according to claim 1, wherein The preprocessing in step S200 includes: Noise reduction: Use a Kalman filter or a moving window mean to process the transient noise of the sensors, and identify the abnormal data of the sensors based on statistical methods to eliminate outliers. Timestamp synchronization: After noise reduction is completed, synchronize the timestamps of the data collected by different types of sensors.

4. The full DC lighting method based on AI perception according to claim 1, wherein The dynamically optimized target dimming strategy includes an objective function and dynamic constraint conditions.

5. The full DC lighting method based on AI perception according to claim 1, wherein The dimming instruction controls the state of the lamp by mapping the optimal dimming current to a duty cycle between 0 and 1, and controlling the lamp with this duty cycle.

6. The full DC lighting method based on AI perception according to claim 2, wherein The goal of the edge execution strategy control is to keep the current of the lamp close to the optimal dimming current, and compensate for the current errors caused by line voltage drop, hardware characteristics, and load changes. Through the proportional and integral control of the PI controller, it can quickly respond and eliminate long-term errors, and gradually adjust the actual current to the target current.

7. The full DC lighting method based on AI perception according to claim 4, characterized in that The objective function of the dynamic optimization target dimming strategy includes: A comfort term, a stability term, and an energy-saving term. Among them, The comfort term is related to the actual illuminance and the target illuminance. The actual illuminance is controlled by the driving current and is related to the ambient light intensity. The energy-saving term is related to the system power consumption, and the power consumption is a function of the driving current. The stability term is related to the rate of change of the current, and the goal is to limit the rapid change of the current. The comfort term, the stability term, and the energy-saving term together form a whole, and achieve a comprehensive balance of comfort, energy saving, and stability by optimizing the current. The dynamic constraint conditions of the dynamic optimization target dimming strategy are used to ensure that the dimming strategy meets the physical limitations in actual applications while ensuring the stability of the system and the comfort of users, including: Current constraint, illuminance deviation constraint, and stability derivative constraint. Among them, The current constraint is used to ensure that the current is between the minimum and maximum values of the current allowed by the lamp, so that the current value will not exceed the maximum limit that the device can withstand, nor will it be lower than the minimum operating current of the device. The illuminance deviation constraint is used to ensure that the illuminance in the target area is close to the predetermined target illuminance to provide the best comfort experience. The stability derivative constraint ensures the smooth operation of the system by setting the maximum limit of the rate of change of the current and restricting the rate of change of the current.

8. A fully DC lighting system based on AI perception, characterized in that, The all-DC lighting system includes: A processor; A memory storing a computer program, which when executed by the processor, implements the all-DC lighting system method based on AI perception as described in any one of claims 1 to 7.

Images