Intelligent LED illumination energy-saving optimization method and system based on AI algorithm

Abstract

The invention discloses an intelligent LED illumination energy-saving optimization method and system based on an AI algorithm, and relates to the field of energy-saving optimization, and the method comprises the steps: inputting a multi-modal data set into a multi-task neural network model, carrying out the abnormal energy consumption detection, recognizing the abnormal energy consumption, generating a photon space distribution optimization instruction through fault root cause reasoning, and carrying out the optimization of the photon space distribution. Based on the photon space distribution optimization instruction, a U-Net generator in the generative adversarial network is adopted to obtain a photon space distribution probability graph, physical simulation is carried out through differentiable ray tracing, an optimal lens curvature parameter is obtained, and multi-target lighting effect optimization and thermal management cooperative control is carried out through a dynamic light field regulation and control algorithm; according to the method, through closed-loop linkage of data fusion, physical modeling and intelligent optimization, the energy efficiency utilization rate of an LED system is improved, the service life of a lamp is prolonged, and meanwhile the personalized illumination requirement and the thermal management requirement in a complex scene are met.

Description

Technical Field

[0001] The present invention relates to the field of energy-saving optimization, and in particular to an intelligent LED lighting energy-saving optimization method and system based on an AI algorithm. Background Art

[0002] With the continuous development of modern building technology and people's high attention to energy conservation and environmental protection, LED lighting has been widely used in commercial, industrial and home lighting fields due to its high efficiency, energy saving, long life and environmental friendliness. Traditional energy-saving control methods are mainly based on regular strategies such as light intensity threshold-triggered dimming, timed start and stop, and simple zoning control, which have limitations in dynamically adapting to complex environmental changes and multi-objective optimization needs. In recent years, with the rapid development of artificial intelligence technology, the application of energy conservation and environmental protection in smart lighting has gradually become a research hotspot. AI technology can realize intelligent control of LED lighting by real-time monitoring and analysis of environmental data, thereby further improving the energy efficiency and comfort of lighting.

[0003] However, although the existing technology has made certain progress in LED lighting units, there are still some shortcomings. The existing technology cannot effectively balance the relationship between multiple objectives such as light efficiency, energy consumption and thermal management in optimizing the spatial distribution of photons, resulting in the overall performance of the lighting unit to be improved. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an intelligent LED lighting energy-saving optimization method based on AI algorithm to solve the problem that the relationship between multiple objectives of photon spatial distribution optimization cannot be effectively balanced between light efficiency, energy consumption and thermal management.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides an intelligent LED lighting energy-saving optimization method based on an AI algorithm, comprising: collecting raw data and preprocessing it to generate a multimodal data set; the multimodal data set includes a heat map of personnel positions, light intensity values, operating temperature, vibration acceleration signals, and effective current values;

[0008] Input multimodal data sets into a multi-task neural network model to perform abnormal energy consumption detection, identify energy consumption anomalies, and generate photon spatial distribution optimization instructions through fault root cause reasoning;

[0009] Based on the photon spatial distribution optimization instruction, the U-Net generator in the generative adversarial network is used to obtain the photon spatial distribution probability map, which is then verified using the discriminator. Physical simulation is then performed through differentiable ray tracing to obtain the optimal lens curvature parameters.

[0010] According to the optimal lens curvature parameters, multi-objective light efficiency optimization and thermal management coordinated control are performed through the dynamic light field control algorithm to generate an LED lighting energy-saving optimization strategy.

[0011] As a preferred solution of the intelligent LED lighting energy-saving optimization method based on AI algorithm described in the present invention, wherein: the construction of a multi-task neural network model, including model architecture and training strategy;

[0012] The model architecture is divided into an anomaly detection branch and an instruction generation branch;

[0013] The anomaly detection branch performs abnormal energy consumption detection on multimodal data sets through a multi-scale temporal convolutional network and a cross-modal attention mechanism, and outputs the anomaly level;

[0014] The instruction generation branch is based on the abnormality level. It uses the causal graph neural network and the reinforcement learning strategy network to analyze the root cause of the fault and output the description of the root cause of the fault. In addition, it dynamically generates instructions for optimizing the photon spatial distribution through collaboration with the reinforcement learning strategy network.

[0015] Based on multimodal time series data, cross-modal feature extraction and anomaly detection training are performed on the anomaly detection branch, and the instruction generation branch is jointly trained through dynamic weight adjustment and differentiable physical simulation methods.

[0016] As a preferred solution of the intelligent LED lighting energy-saving optimization method based on AI algorithm described in the present invention, wherein: the U-Net generator in the generative adversarial network is used to obtain the photon spatial distribution probability map, and the specific steps are as follows:

[0017] In the generative adversarial network, the encoder uses a cross-modal attention mechanism to dynamically fuse the thermal characteristics of the person, the natural lighting characteristics, and the vibration energy characteristics to form a high-dimensional semantic representation;

[0018] The decoder gradually upsamples through the deconvolution layer, combines the detailed features of each level of the encoder with the decoding features in combination with jump connections, and uses the residual gating mechanism to suppress noise interference and generate a photon spatial distribution probability map.

[0019] As a preferred solution of the intelligent LED lighting energy-saving optimization method based on AI algorithm described in the present invention, wherein: the discriminator is used to verify the photon spatial distribution probability map, and the specific steps are as follows:

[0020] Perform spatial registration between the photon spatial distribution probability map and the real scene light effect distribution map;

[0021] The discriminator adopts a joint mechanism of global discrimination and local discrimination. The global discrimination compares the overall distribution of the light field, and the local discrimination randomly samples the area to analyze the statistical characteristics of the light spot.

[0022] Overfitting is suppressed through dynamic label smoothing technology, while gradient penalty constrains the discriminator's sensitivity to the edge of the spot;

[0023] During the physical verification phase, the photon spatial distribution probability map is converted into a power allocation scheme, and the temperature field distribution is verified in combination with the thermal resistance model. The confidence score of the exceeding standard area is lowered, and the light intensity error, thermodynamic score and light spot consistency are integrated to generate the final judgment result.

[0024] As a preferred solution of the intelligent LED lighting energy-saving optimization method based on AI algorithm described in the present invention, wherein: the physical simulation is performed by differentiable ray tracing to obtain the optimal lens curvature parameters, and the specific steps are as follows:

[0025] Based on the photon spatial distribution probability map, the lens spherical curvature parameters are initialized, and a light path covering the light output half-angle is generated through Monte Carlo sampling. The refraction angle of the light is calculated using the differentiable Snell's law. The light energy attenuation is integrated by combining the lens material absorptivity and the surface scattering model to generate a simulated light field distribution.

[0026] The simulated light field is compared pixel by pixel with the probability map to calculate the mean square error of the light intensity. At the same time, the thermal resistance model is used to verify whether the temperature gradient exceeds the standard, and the lens curvature parameters that meet the light intensity distribution and thermodynamic constraints are output.

[0027] As a preferred solution of the intelligent LED lighting energy-saving optimization method based on AI algorithm described in the present invention, wherein: the multi-objective light efficiency optimization and thermal management collaborative control are performed by the dynamic light field control algorithm, and the specific steps are:

[0028] Initialize the lamp grouping based on the optimal lens curvature parameters, define total energy consumption, temperature gradient and illumination uniformity as optimization objectives, and use a multi-objective particle swarm algorithm to dynamically adjust the weights to generate a Pareto optimal solution;

[0029] After verifying that the total energy consumption, temperature gradient and illumination uniformity meet the standards through physical simulation, the optimization strategy is converted into lighting control instructions to achieve coordinated optimization of light field distribution and thermodynamic constraints.

[0030] As a preferred solution of the intelligent LED lighting energy-saving optimization method based on AI algorithm described in the present invention, wherein: the LED lighting energy-saving optimization strategy is generated, the specific steps are:

[0031] Based on the optimal lens curvature parameters, the dynamic light field control algorithm is used to group the lamps. Through the multi-objective particle swarm optimization algorithm, the dimming parameters and illumination angles of each group of lamps are optimized with the lowest energy consumption and the minimization of the lamp surface temperature gradient as the optimization goals, while meeting the illumination uniformity of the target area, to generate an LED lighting energy-saving optimization strategy.

[0032] In a second aspect, the present invention provides an intelligent LED lighting energy-saving optimization system based on AI algorithm, comprising: an acquisition module, a detection module, a simulation module and an optimization module.

[0033] An acquisition module is used to collect raw data and perform preprocessing to generate a multimodal data set; the multimodal data set includes a thermal map of personnel positions, light intensity values, operating temperature, vibration acceleration signals, and effective current values;

[0034] The detection module is used to input the multimodal data set into the multi-task neural network model, perform abnormal energy consumption detection, identify energy consumption anomalies, and generate photon spatial distribution optimization instructions through fault root cause reasoning;

[0035] The simulation module is used to optimize the photon spatial distribution instructions, use the U-Net generator in the generative adversarial network to obtain the photon spatial distribution probability map, use the discriminator to verify the photon spatial distribution probability map, and perform physical simulation through differentiable ray tracing to obtain the optimal lens curvature parameters;

[0036] The optimization module is used to perform multi-objective light efficiency optimization and thermal management collaborative control based on the optimal lens curvature parameters through a dynamic light field control algorithm to generate an LED lighting energy-saving optimization strategy.

[0037] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the intelligent LED lighting energy-saving optimization method based on the AI ​​algorithm as described in the first aspect of the present invention is implemented.

[0038] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the intelligent LED lighting energy-saving optimization method based on the AI ​​algorithm as described in the first aspect of the present invention.

[0039] The beneficial effects of the present invention are: by adopting a multi-task neural network combined with a cross-modal attention mechanism and a reinforcement learning strategy, it can dynamically detect energy consumption anomalies and infer the root cause of the fault, generate targeted photon distribution instructions, improve the abnormal response speed and processing accuracy, use the U-Net generative adversarial network to generate a photon spatial distribution probability map, and optimize the lens curvature parameters through differentiable ray tracing physical simulation to ensure that the light efficiency distribution conforms to physical laws while reducing thermodynamic risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0041] Figure 1 This is the core flow chart of the intelligent LED lighting energy-saving optimization method based on AI algorithm;

[0042] Figure 2 This is a diagram of the multi-task neural network architecture;

[0043] Figure 3 To generate an interactive graph between adversarial networks and ray tracing;

[0044] Figure 4 Execution graph of the dynamic light field control algorithm; DETAILED DESCRIPTION

[0045] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0046] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0047] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0048] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides an intelligent LED lighting energy-saving optimization method based on an AI algorithm, comprising the following steps:

[0049] S1, collect raw data and preprocess them to generate a multimodal data set;

[0050] Infrared sensor nodes are deployed on the ceiling surface in a rectangular grid topology. The node spacing is dynamically adjusted according to the spatial area, so that there is partial overlap in the coverage areas of adjacent nodes. The infrared sensor has a built-in Fresnel lens group that detects the heat source of the human body vertically downward to obtain personnel position distribution data and output a personnel position heat map represented by two-dimensional coordinates. All infrared sensor nodes are connected via the RS-485 bus and configured with a unified sampling frequency to avoid signal conflicts. A group of horizontally spaced light sensors are installed at a fixed distance from the window edge, with the vertical height being the same as the LED lamp. Each light sensor covers the horizontal field of view, measures the ambient light intensity and outputs a normalized light intensity value, which is transmitted to the central controller in real time using a wireless protocol. A temperature sensor with a range covering the operating range of the lamp is mounted on the surface of the lamp radiator to measure the operating temperature and obtain the operating temperature data. A vibration sensor with a frequency response range covering the mechanical vibration frequency band is installed at the connection between the lamp housing and the bracket to measure the vibration acceleration signal and obtain the vibration acceleration signal. A current sensor with a sampling rate that meets the real-time monitoring requirements is connected in series to the output end of the driving power supply to measure the effective value of the driving current.

[0051] All sensor nodes have built-in hardware clock chips and are synchronized with the central controller through a time synchronization protocol. A local timestamp is added to each data sampling. The personnel position heat map output by the infrared sensor is spatially filtered to remove discrete points with abnormal signal-to-noise ratios. A sliding window mean filter is used for the light intensity value to suppress instantaneous light fluctuations. The working temperature data and the effective current value are synchronized to the same time axis through cubic spline interpolation, taking the vibration acceleration signal with the highest sampling rate as the benchmark. A three-dimensional spatiotemporal correlation matrix is ​​constructed based on the timestamp, sensor type, and spatial coordinates. The grayscale value of the personnel position heat map, the lux value of the light intensity value, the Celsius value of the working temperature, the gravity acceleration value of the vibration acceleration signal, and the ampere value of the effective current value are linearly mapped to a unified interval to complete the normalization processing. The normalized data are integrated with the timestamp as the index to form a multimodal data set with structured storage.

[0052] S2. Input the multimodal data set into the multi-task neural network model to perform abnormal energy consumption detection, identify energy consumption anomalies, and generate photon spatial distribution optimization instructions through fault root cause reasoning;

[0053] Kernel density estimation is performed on the personnel location heat map to generate a spatial probability density distribution, and sliding window mean filtering is used to eliminate positioning jitter noise. The spatial distribution feature tensor is output. Frequency domain decomposition is performed on the light intensity value to separate the natural light trend term from the lamp light intensity fluctuation term. The difference between the lamp light intensity fluctuation term and the natural light trend term is calculated as the dynamic compensation feature vector. The time series of the operating temperature data and the vibration acceleration signal are aligned through cubic spline interpolation to construct a temperature gradient matrix. In the vibration acceleration signal, wavelet packet decomposition is used to extract frequency bands preset based on the mechanical fault characteristics of the equipment, such as the energy of the high-frequency resonance and low-frequency modulation frequency bands. The vibration energy feature vector is generated and Hadamard product operation is performed with the temperature gradient matrix to output the joint feature matrix of temperature gradient and vibration energy. Current surge events are detected by real-time monitoring of the current effective value. Sliding window root mean square processing is used, combined with the instantaneous current value surge exceeding the statistical baseline, and time domain marking is performed as a high-level anomaly candidate feature.

[0054] The construction of a multi-task neural network model based on multimodal data fusion and hierarchical task processing mechanism is mainly divided into two branches working together. The anomaly detection branch extracts local spatiotemporal features through the 3×3 convolution kernel of the multi-scale temporal convolutional network and uses the dilated convolution layer to capture long-term patterns. Long-term patterns refer to trend changes with a large span in the time dimension, such as diurnal light fluctuations and daily / weekly patterns of personnel flow. A cross-attention mechanism is used to dynamically allocate weights between the light intensity and personnel location heat maps. The anomaly detection branch adopts a hard parameter sharing strategy, sharing the underlying convolutional layer to extract common features, and then outputs the anomaly level through the task-specific layer to trigger different response strategies.

[0055] The instruction generation branch constructs a fault reasoning chain based on a causal graph neural network, associates multimodal features with the anomaly level as the root node, uses Bayesian reasoning to dynamically update edge weights, and generates confidence in the root cause of the fault. The reinforcement learning strategy network integrates a differentiable ray tracing algorithm, optimizes the photon distribution parameters through the reward function, and uses a particle swarm algorithm to solve the optimal solution for light field control in real time. The instruction generation branch combines soft parameter sharing, optimizes the lamp layout through spherical radial basis functions at the global level, and dynamically adjusts the parameters of single lamps through U-Net at the local level. At the same time, the Lagrange multiplier method is introduced to constrain the heat dissipation requirements and the physical feasibility of the lens size. The reward function calculates the weighted sum of the energy consumption reduction rate, the temperature gradient change, and the light uniformity score through dynamic weight coefficients to generate a reward value R. The formula is:

[0056] R = α·ΔE + β·ΔT + γ·S;

[0057] Among them, α represents the energy saving weight (value range: 0.3-0.6), β represents the temperature difference weight (value range: 0.2-0.5), γ represents the light uniformity weight (value range: 0.1-0.4), ΔE represents the energy consumption reduction rate, ΔT represents the temperature gradient change (such as ΔT>5℃ / m triggers heat dissipation), and S represents the illumination uniformity score;

[0058] The energy distribution of lamp photons in space is simulated by using a differentiable ray tracing algorithm. The photon distribution is defined as consisting of the total luminous flux, beam diffusion coefficient, and spatial distance. The actual lighting intensity is calculated as:

[0059]

[0060] Where Φ is the total luminous flux, d is the spatial distance, σ is the beam diffusion coefficient (0.1≤σ≤5.0), I represents the degree of influence of photon distribution on actual illumination, and e is a natural constant;

[0061] Based on the impact of photon distribution on actual illumination, the luminous flux and diffusion coefficient are optimized in real time, and hierarchical control is achieved through a multi-scale optimization strategy. At the global level, spherical radial basis functions are used to optimize the power distribution of lamp groups to improve light uniformity. At the local level, the U-Net generator is used to dynamically adjust the illumination angle and luminous flux of single lamps for high-energy consumption grids identified by kernel density estimation to respond to dynamic changes in the distribution of personnel monitored in real time. For example, when an abnormal change in the temperature gradient on the surface of the lamp is detected, a gradient power reduction strategy is triggered, and the lens size and heat dissipation requirements are constrained by the Lagrange multiplier method. When there is a dynamic deviation between the real-time monitored personnel distribution and the coverage range of the lamps, the power distribution weight coefficient of the lamps is dynamically adjusted according to the priority rule that safety instructions are greater than light efficiency instructions preset in the reinforcement learning strategy network. The particle swarm algorithm is combined to balance energy consumption, temperature gradient and light uniformity goals.

[0062] A differentiable ray tracing algorithm is used to perform physical simulations on the optimized photon distribution, verifying the consistency of the spot shape with the target irradiance. The illumination uniformity score and temperature gradient distribution map are output. Dynamic light field control uses a particle swarm optimization algorithm to find the optimal solution in real time on the edge computing device. This algorithm is combined with online Bayesian reasoning to update the edge weights of the causal graph neural network, dynamically balancing light uniformity and temperature gradient constraints.

[0063] When it is detected that the population density in a certain spatial grid exceeds the historical average, a local brightening instruction is generated, and the brightness of the lamps in the spatial grid area is increased and the brightness of adjacent grids is adaptively limited to prioritize the illumination of the active area. When the confidence level of the root cause of the fault exceeds the designed physical fault tolerance boundary, the power of the corresponding lamp is reduced and a maintenance work order is generated. The output includes a photon spatial distribution optimization instruction including the instruction type, the active grid, the parameter settings and the physical constraints.

[0064] S3. Based on the photon spatial distribution optimization instruction, the U-Net generator in the generative adversarial network is used to obtain the photon spatial distribution probability map, the discriminator is used to verify the photon spatial distribution probability map, and physical simulation is performed through differentiable ray tracing to obtain the optimal lens curvature parameters;

[0065] The personnel position heat map is generated by fusing the infrared sensor and camera positioning data. The kernel density estimation algorithm is used to calculate the spatial probability density of the original coordinate data and output the heat distribution map. The natural light intensity data is collected by the light sensor and multi-scale decomposition is performed using wavelet transform to separate the low-frequency background light and high-frequency dynamic change components. The natural light intensity map is generated by weighted fusion. The personnel position heat map and the natural light intensity map are aligned in time and space. The vibration acceleration is mapped to the lamp group coordinates through spatial clustering to generate a vibration energy distribution map, which shares the same spatial coordinate system with the heat map. A three-dimensional grid spatial coordinate system is constructed and encoded. The encoder uses multi-scale convolutional layers to extract the spatial features of thermal distribution maps, light intensity maps, and vibration energy distribution maps. It dynamically weights and fuses the thermal distribution features of personnel, natural light intensity features, and vibration energy features through a cross-modal attention mechanism to generate a cross-modal joint feature map. The cross-modal joint feature map output by the encoder is input into the decoder part of the U-Net generator. The decoder gradually improves the resolution of the cross-modal joint feature map through deconvolution layers. After each layer of deconvolution, the cross-modal joint feature map corresponding to the encoder level is spliced ​​through jump connections, and the residual gating mechanism is used to suppress noise interference to generate a photon spatial distribution probability map.

[0066] The photon spatial distribution probability map output by the U-Net generator is spatially aligned with the real scene light effect distribution map measured by a high-precision light intensity meter. The positions of the two need to correspond one to one in the three-dimensional grid space coordinate system. The numerical ranges of the photon spatial distribution probability map and the real scene light effect distribution map are normalized to eliminate the influence of sensor dimension differences on the discrimination process. The real scene light effect distribution map can be measured point by point in the target area using a light intensity meter, and the illuminance value of each point is recorded to form a two-dimensional light intensity matrix. The CCD image sensor is used to shoot the diffuse reflection plate light spot, and the light intensity distribution is converted by the pixel grayscale value. At the same time, the integrating sphere full-space measurement technology is combined to obtain three-dimensional light intensity data, and the standard The standardization process performs multi-angle rotation scanning on the lamp, and finally forms a data normalization and spatial registration. The luminous flux-illuminance conversion model is used to map the probability map value to the physical luminous flux value. In combination with international lighting standards, the non-compliant areas are identified and luminous flux compensation instructions are generated. The thermal resistance model of the lamp is simultaneously integrated to verify whether the power distribution corresponding to the probability map causes the local temperature gradient to exceed the standard. The lighting density of the high-heating area is dynamically adjusted. The temperature field distribution is output through the thermal resistance model of the lamp. The ratio of the area where the temperature gradient exceeds the material threshold to the total area is statistically calculated to generate a thermodynamic constraint score. At the same time, the normalized mean square error of the light intensity is used as the light intensity distribution score to quantify the consistency between the simulated light field and the photon probability map.

[0067] In adversarial training, the registered probability map and the real light effect map are input to the discriminator. Dynamic label smoothing technology is used to inject Gaussian noise into the real data labels to suppress the generator's overfitting to local noise. At the same time, a gradient penalty mechanism is used to constrain the discriminator's sensitivity to local light spots, improving the stability of adversarial training.

[0068] Based on the photon spatial distribution probability map, the light source position is defined as the geometric center of the LED chip of the lamp. The initial lens curvature parameters are generated according to the spherical curvature radius. The emission direction of the light source is randomly generated by the Monte Carlo sampling method, covering all possible paths within the half-angle range of the lamp. The refraction of light is calculated using the differentiable Snell's law to calculate the refraction angle of light on the lens surface. The formula is:

[0069]

[0070] Where n1 represents the refractive index of air, n2 represents the refractive index of the lens material, θ1 represents the incident angle, and θ2 represents the refraction angle of the light on the lens surface. The refraction angle calculation process is derived through automatic differentiation technology to generate the gradient information of the light propagation path;

[0071] The angle of incidence is determined by the dot product of the light direction vector and the lens surface normal vector, and the refractive index of air and the refractive index of the lens material are dynamically adjusted according to the ambient temperature;

[0072] The energy of the refracted light based on the refraction angle of the light on the lens surface is integrated along the propagation path to generate a simulated light field distribution diagram. The energy attenuation model takes into account the absorption rate of the lens material and the scattering loss of the surface roughness. The data comes from the lamp material property library.

[0073] Compare the simulated light field distribution map with the photon spatial distribution probability map generated by U-Net pixel by pixel, and calculate the mean square error L. The formula is:

[0074]

[0075] Among them, y represents the measured data of light intensity distribution, It is expressed as the simulation result of differentiable ray tracing, where N is the total number of pixels in the light field distribution. The mean square error weight is dynamically adjusted according to the illumination requirement of the target area. The correlation function between the lens surface temperature gradient and the operating temperature of the lamp is defined to limit the light focusing density in the high-temperature area. The mean square error of the light intensity distribution and the thermodynamic constraint loss are superimposed according to the dynamic weight. The weight coefficient is automatically learned through back propagation. Based on the joint gradient of the mean square error of the light intensity distribution and the thermodynamic constraint, the Adam optimizer is used to update the lens curvature parameters. After each round of optimization, the ray tracing simulation is re-executed to calculate the light intensity uniformity index. When the light intensity uniformity shows a significant improvement trend during the optimization process, the current optimal lens curvature parameters are locked. Otherwise, the Monte Carlo sampling density is increased and the light path is regenerated.

[0076] The optimized lens curvature parameters are input into the differentiable ray tracing simulator to generate dual-channel physical simulation results including light intensity distribution and temperature field distribution. The discriminator generates a physical feasibility score by comparing the simulation data with the historical optimal light field data. The score comprehensively considers the local light spot consistency, overall illumination uniformity and temperature gradient threshold compliance rate. When the score is too low, a re-optimization instruction is sent to the U-Net generator to trigger iterative optimization based on the updated photon spatial distribution probability map. At the same time, the loss function weight is adjusted according to the physical feasibility score. When the thermodynamic constraint score is lower than the light intensity distribution score, the thermodynamic weight is increased. When the thermodynamic constraint score is higher than the light intensity distribution score, the focus is on light intensity distribution optimization.

[0077] According to the differentiable ray tracing algorithm, the gradient of the loss function with respect to the lens curvature parameter is calculated. The gradient represents the effect of a small change in the lens curvature parameter on the loss function. The specific calculation method is as follows:

[0078]

[0079] Among them, H represents the loss function, μ represents the lens curvature parameter, Expressed as the gradient of the loss function H with respect to the lens curvature parameter μ;

[0080] Using the gradient descent optimization algorithm, the update direction and step size of the lens curvature parameters are determined based on the calculated gradient of the loss function with respect to the lens curvature parameters. The specific steps are: selecting a suitable learning rate to control the amplitude of the parameter update, updating the lens curvature parameters in the opposite direction of the gradient to reduce the value of the loss function, and repeating the light propagation path simulation and parameter adjustment steps until the value of the loss function converges to a smaller value. After each iteration, the light propagation path is re-simulated, the new loss function value is calculated, and the lens curvature parameters are updated. After multiple iterative optimizations, the optimal lens curvature parameters are obtained to minimize the difference between the light propagation path and the desired photon spatial distribution.

[0081] S4. Based on the optimal lens curvature parameters, a dynamic light field control algorithm is used to perform multi-objective light efficiency optimization and thermal management collaborative control to generate an LED lighting energy-saving optimization strategy;

[0082] Initialize the relevant parameters of the dynamic light field control algorithm based on the optimal lens curvature parameters. The initialization parameters include the lens geometry, optical properties, and lighting requirements of the target area.

[0083] The validity of the received optimal lens curvature parameters is verified to ensure that the parameter values ​​are within the physically achievable range. When it is found that the parameter values ​​exceed the safe range of the material refractive index, the error information is recorded and an alarm is triggered. The verified optimal lens curvature parameters are passed to the dynamic light field control algorithm. The dynamic light field control algorithm will use the verified optimal lens curvature parameters to adjust the light field distribution of the lamp to achieve energy-saving optimization. Ensure that the data format and interface during the transmission process meet the requirements of the dynamic light field control algorithm. In the dynamic light field control algorithm, the relevant parameters are initialized according to the received optimal lens curvature parameters. The initialized relevant parameters need to be consistent with the received optimal lens curvature parameters.

[0084] According to the optimal lens curvature parameters, the optical characteristics of each lamp are analyzed. The spot angle is measured by a goniophotometer and the absolute value of the luminous flux is tested by an integrating sphere. The spot classification and luminous flux threshold are generated as the basis for lighting task allocation. According to the spot angle classification, luminous flux threshold and lighting task, the grouping criteria are defined: spot angle difference ≤ 5°, luminous flux deviation ≤ 10%, consistent task type and spatial spacing ≤ 2m. Based on the grouping criteria, the K-means clustering algorithm is used to group the lamps: the centroid is initialized as the spot angle and luminous flux mean, the lamps are traversed and assigned to the nearest centroid group, and the intra-group variance is iteratively optimized until convergence to ensure that the intra-group spot angle difference is ≤ 5° and the luminous flux deviation is ≤ 10%. When verifying the grouping results, if the intra-group spot angle range is > 10° or the luminous flux range is > 20%, the centroid is readjusted and iteratively assigned until the grouping criteria are met. The lamp number, spot angle mean, luminous flux range and task type of each group are output as input parameters of the dynamic light field control algorithm.

[0085] Under the premise of meeting the uniformity of illumination in the target area, three optimization goals are defined. The first is to minimize the total energy consumption of the lighting equipment, which is to minimize the energy consumption of the entire lighting equipment by reasonably adjusting the power and usage time of the lamps. The second is to minimize the surface temperature gradient of the lamps to ensure that the surface temperature of the lamps is evenly distributed during operation, avoid local overheating, and thus extend the service life of the lamps. In order to achieve the above optimization goals, the multi-objective particle swarm optimization algorithm is selected as the optimization tool. During the optimization process, the particle swarm needs to be initialized. Each particle represents a set of possible lamp dimming parameters and illumination angles. The initial position of the particle can be randomly generated. The randomly generated initial position can provide a wider search space for the optimization process. After initialization, each particle has an initial speed and position. According to the dimming parameters and illumination angles represented by each particle, the comprehensive fitness is calculated. The formula is:

[0086] F=w1·P+w2·U+w3·Q;

[0087] Among them, F is the comprehensive fitness, P is the total energy consumption of the lighting equipment, U is the illumination uniformity error, Q is the surface temperature gradient of the lamp, w1 is the weight coefficient of the energy consumption target (the value range is: 0.4≤w1≤0.7), w2 is the weight coefficient of the illumination uniformity target (the value range is: 0.4≤w2≤0.7), and w3 is the weight coefficient of the temperature gradient target (the value range is: 0.4≤w3≤0.7), which is used to balance the importance of different targets;

[0088] The fitness of each particle is evaluated according to the defined optimization objectives. This is the degree to which the particle satisfies the two optimization objectives under the current parameters. Based on the numerical evaluation results of the comprehensive fitness, the position and velocity of the particle are updated to guide the particle to move to a more optimal solution space. This process is repeated until the fitness no longer improves significantly. A set of Pareto optimal solutions is generated through the multi-objective particle swarm optimization algorithm. A set of solutions is selected from the Pareto optimal solutions as the final optimization strategy. The dimming parameters and illumination angles of each group of lamps are output and applied to actual LED lighting equipment to achieve the energy-saving optimization goal. Through this multi-objective optimization method, not only can the energy consumption of lighting equipment be effectively reduced, but also the stable operation of the lamps can be ensured and their service life can be extended, thereby providing an efficient, energy-saving and stable operation solution for LED lighting equipment.

[0089] The optimal dimming parameters and illumination angles of each group of lamps are extracted from the final results of the multi-objective particle swarm optimization algorithm and organized into specific optimization strategies. The generated optimization strategies are verified to meet the optimization goals. The verification content includes whether the energy consumption is minimized, whether the illumination uniformity meets the requirements, and whether the surface temperature distribution of the lamps is uniform. The generated optimization strategies are output and used for the regulation of actual lighting equipment. The optimization strategies are converted into a format suitable for the input of LED lamp control devices. The optimization strategies are transmitted to the control devices of LED lamps through wired and wireless communication. The control devices will automatically adjust the operating status of the lamps according to the received optimization strategies to achieve energy-saving and optimized operation. During the actual operation, the operating status of the lighting equipment is continuously monitored to ensure the effectiveness of the optimization strategies. If any abnormalities or deviations from the optimization goals are found, adjustments and optimizations are made in a timely manner.

[0090] This embodiment also provides an intelligent LED lighting energy-saving optimization system based on an AI algorithm, including:

[0091] Acquisition module, detection module, simulation module and optimization module,

[0092] An acquisition module is used to collect raw data and perform preprocessing to generate a multimodal data set; the multimodal data set includes a thermal map of personnel positions, light intensity values, operating temperature, vibration acceleration signals, and effective current values;

[0093] The detection module is used to input the multimodal data set into the multi-task neural network model, perform abnormal energy consumption detection, identify energy consumption anomalies, and generate photon spatial distribution optimization instructions through fault root cause reasoning;

[0094] The simulation module is used to optimize the photon spatial distribution instructions, use the U-Net generator in the generative adversarial network to obtain the photon spatial distribution probability map, use the discriminator to verify the photon spatial distribution probability map, and perform physical simulation through differentiable ray tracing to obtain the optimal lens curvature parameters;

[0095] The optimization module is used to perform multi-objective light efficiency optimization and thermal management collaborative control based on the optimal lens curvature parameters through a dynamic light field control algorithm to generate an LED lighting energy-saving optimization strategy.

[0096] This embodiment also provides a computer device, which is suitable for the case of an intelligent LED lighting energy-saving optimization method based on an AI algorithm, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the intelligent LED lighting energy-saving optimization method based on the AI ​​algorithm proposed in the above embodiment.

[0097] The computer device may be a terminal, comprising a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse.

[0098] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the intelligent LED lighting energy-saving optimization method based on the AI ​​algorithm proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk.

[0099] In summary, the present invention adopts a multi-task neural network combined with a cross-modal attention mechanism and a reinforcement learning strategy to dynamically detect energy consumption anomalies and infer the root cause of the fault, generate targeted photon distribution instructions, improve the abnormal response speed and processing accuracy, use the U-Net generative adversarial network to generate a photon spatial distribution probability map, and optimize the lens curvature parameters through differentiable ray tracing physical simulation to ensure that the light efficiency distribution conforms to the physical laws while reducing thermodynamic risks.

[0100] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An intelligent LED lighting energy-saving optimization method based on AI algorithm, characterized by: include: Collecting raw data and preprocessing it to generate a multimodal data set; the multimodal data set includes a thermal map of personnel positions, light intensity values, operating temperatures, vibration acceleration signals, and effective current values; Input multimodal data sets into a multi-task neural network model to perform abnormal energy consumption detection, identify energy consumption anomalies, and generate photon spatial distribution optimization instructions through fault root cause reasoning; Based on the photon spatial distribution optimization instruction, the U-Net generator in the generative adversarial network is used to obtain the photon spatial distribution probability map, which is then verified using the discriminator. Physical simulation is then performed through differentiable ray tracing to obtain the optimal lens curvature parameters. According to the optimal lens curvature parameters, multi-objective light efficiency optimization and thermal management coordinated control are performed through the dynamic light field control algorithm to generate an LED lighting energy-saving optimization strategy.

2. The AI ​​algorithm-based intelligent LED lighting energy-saving optimization method according to claim 1, characterized in that: Construction of multi-task neural network models, including model architecture and training strategies; The model architecture is divided into an anomaly detection branch and an instruction generation branch; The anomaly detection branch performs abnormal energy consumption detection on multimodal data sets through a multi-scale temporal convolutional network and a cross-modal attention mechanism, and outputs the anomaly level; The instruction generation branch is based on the abnormality level. It uses the causal graph neural network and the reinforcement learning strategy network to analyze the root cause of the fault and output the description of the root cause of the fault. In addition, it dynamically generates instructions for optimizing the photon spatial distribution through collaboration with the reinforcement learning strategy network. Based on multimodal time series data, cross-modal feature extraction and anomaly detection training are performed on the anomaly detection branch, and the instruction generation branch is jointly trained through dynamic weight adjustment and differentiable physical simulation methods.

3. The AI ​​algorithm-based intelligent LED lighting energy-saving optimization method according to claim 1, characterized in that: The U-Net generator in the generative adversarial network is used to obtain the photon spatial distribution probability map. The specific steps are: In the generative adversarial network, the encoder uses a cross-modal attention mechanism to dynamically fuse the thermal characteristics of the person, the natural lighting characteristics, and the vibration energy characteristics to form a high-dimensional semantic representation; The decoder gradually upsamples through the deconvolution layer, combines the detailed features of each level of the encoder with the decoding features in combination with jump connections, and uses the residual gating mechanism to suppress noise interference and generate a photon spatial distribution probability map.

4. The AI ​​algorithm-based intelligent LED lighting energy-saving optimization method according to claim 1, characterized in that: The specific steps of using the discriminator to verify the photon spatial distribution probability map are as follows: Perform spatial registration between the photon spatial distribution probability map and the real scene light effect distribution map; The discriminator adopts a joint mechanism of global discrimination and local discrimination. The global discrimination compares the overall distribution of the light field, and the local discrimination randomly samples the area to analyze the statistical characteristics of the light spot. Overfitting is suppressed through dynamic label smoothing technology, while gradient penalty constrains the discriminator's sensitivity to the edge of the spot; During the physical verification phase, the photon spatial distribution probability map is converted into a power allocation scheme, and the temperature field distribution is verified in combination with the thermal resistance model. The confidence score of the exceeding standard area is lowered, and the light intensity error, thermodynamic score and light spot consistency are integrated to generate the final judgment result.

5. The AI ​​algorithm-based intelligent LED lighting energy-saving optimization method according to claim 1, characterized in that: The physical simulation is performed by differentiable ray tracing to obtain the optimal lens curvature parameters. The specific steps are: Based on the photon spatial distribution probability map, the lens spherical curvature parameters are initialized, and a light path covering the light output half-angle is generated through Monte Carlo sampling. The refraction angle of the light is calculated using the differentiable Snell's law. The light energy attenuation is integrated by combining the lens material absorptivity and the surface scattering model to generate a simulated light field distribution. The simulated light field is compared pixel by pixel with the probability map to calculate the mean square error of the light intensity. At the same time, the thermal resistance model is used to verify whether the temperature gradient exceeds the standard, and the lens curvature parameters that meet the light intensity distribution and thermodynamic constraints are output.

6. The AI ​​algorithm-based intelligent LED lighting energy-saving optimization method according to claim 1, characterized in that: The multi-objective light efficiency optimization and thermal management coordinated control are performed by the dynamic light field control algorithm, and the specific steps are as follows: Initialize the lamp grouping based on the optimal lens curvature parameters, define total energy consumption, temperature gradient and illumination uniformity as optimization objectives, and use a multi-objective particle swarm algorithm to dynamically adjust the weights to generate a Pareto optimal solution; After verifying that the total energy consumption, temperature gradient and illumination uniformity meet the standards through physical simulation, the optimization strategy is converted into lighting control instructions to achieve coordinated optimization of light field distribution and thermodynamic constraints.

7. The AI ​​algorithm-based intelligent LED lighting energy-saving optimization method according to claim 1, characterized in that: The specific steps of generating the LED lighting energy-saving optimization strategy are as follows: Based on the optimal lens curvature parameters, the dynamic light field control algorithm is used to group the lamps. Through the multi-objective particle swarm optimization algorithm, the dimming parameters and illumination angles of each group of lamps are optimized with the lowest energy consumption and the minimization of the lamp surface temperature gradient as the optimization goals, while meeting the illumination uniformity of the target area, to generate an LED lighting energy-saving optimization strategy.

8. An AI-based intelligent LED lighting energy-saving optimization system, based on the AI-based intelligent LED lighting energy-saving optimization method according to any one of claims 1 to 7, characterized in that: include: Acquisition module, detection module, simulation module and optimization module, An acquisition module is used to collect raw data and perform preprocessing to generate a multimodal data set; the multimodal data set includes a thermal map of personnel positions, light intensity values, operating temperature, vibration acceleration signals, and effective current values; The detection module is used to input the multimodal data set into the multi-task neural network model, perform abnormal energy consumption detection, identify energy consumption anomalies, and generate photon spatial distribution optimization instructions through fault root cause reasoning; The simulation module is used to optimize the photon spatial distribution instructions, use the U-Net generator in the generative adversarial network to obtain the photon spatial distribution probability map, use the discriminator to verify the photon spatial distribution probability map, and perform physical simulation through differentiable ray tracing to obtain the optimal lens curvature parameters; The optimization module is used to perform multi-objective light efficiency optimization and thermal management collaborative control based on the optimal lens curvature parameters through a dynamic light field control algorithm to generate an LED lighting energy-saving optimization strategy.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the intelligent LED lighting energy-saving optimization method based on the AI ​​algorithm are implemented in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the intelligent LED lighting energy-saving optimization method based on the AI ​​algorithm according to any one of claims 1 to 7 are implemented.

Images