Intelligent light regulation method, apparatus, device, and storage medium

Abstract

The application provides a kind of intelligent light regulation method, device and equipment and storage medium, method includes: linear projection operation is carried out to environment parameter sequence and user behavior sequence, obtains basic multi-head attention matrix;Scene feature sequence and basic multi-head attention matrix are interacted with attention, and scene feature attention branch matrix is obtained;User preference sequence and basic multi-head attention matrix are interacted with attention, and user preference attention branch matrix is obtained;Scene feature attention branch matrix and user preference attention branch matrix are dynamically fused based on adaptive fusion coefficient, and fusion attention matrix is obtained;Intelligent light regulation parameter is determined based on fusion attention matrix;Intelligent light terminal is intelligently light regulated based on intelligent light regulation parameter.The application realizes the accurate matching of scene demand and user preference by constructing the attention module of scene feature-user preference, improves the adaptability and personalization of intelligent light regulation.

Description

Technical Field

[0001] This application belongs to the field of intelligent control technology for intelligent lighting, and relates to an intelligent lighting control method, device, equipment, and storage medium. Background Technology

[0002] With the deep integration of IoT and AI technologies, smart lighting systems are evolving from basic manual and timed control to a paradigm of environmental self-sensing and user-personalized adaptive control. However, existing smart lighting control strategies are mostly limited to hard logic threshold triggering based on single-modal sensor data (such as illuminance and infrared sensing) or rely on shallow machine learning models (such as BP neural networks) for decision-making, revealing significant technical bottlenecks when dealing with complex dynamic scenarios.

[0003] Therefore, how to improve the accuracy of intelligent lighting control based on the personalized needs of different users has become an urgent technical problem to be solved. Summary of the Invention

[0004] This application provides an intelligent lighting control method, device, equipment, and storage medium, which is used to improve the accuracy of intelligent lighting control based on the personalized needs of different users, and ensure that the personalized needs of different scenarios and different users are met.

[0005] In a first aspect, this application provides an intelligent lighting control method, the method comprising: collecting environmental parameter sequences, user behavior sequences, scene feature sequences, and user preference sequences required for intelligent lighting control; performing a linear projection operation on the environmental parameter sequences and the user behavior sequences to obtain a basic multi-head attention matrix; performing attention interaction between the scene feature sequences and the basic multi-head attention matrix to obtain a scene feature attention branch matrix; performing attention interaction between the user preference sequences and the basic multi-head attention matrix to obtain a user preference attention branch matrix; obtaining adaptive fusion coefficients; dynamically fusing the scene feature attention branch matrix and the user preference attention branch matrix based on the adaptive fusion coefficients to obtain a fused attention matrix; determining intelligent lighting control parameters based on the fused attention matrix; and performing intelligent lighting control on an intelligent lighting terminal based on the intelligent lighting control parameters.

[0006] This application addresses the problems of weak multi-factor coordination, insufficient personalization, and disconnect between the weight allocation of traditional attention mechanisms and lighting control scenarios in traditional intelligent lighting control by constructing a dual-branch fusion attention module based on scene features and user preferences. It achieves precise matching between scene requirements and user preferences, improving the adaptability, personalization, and energy efficiency of intelligent lighting control. Through formulaic attention weight allocation and adaptive fusion, it enhances the accuracy of lighting control, ensuring that the personalized needs of different scenarios and users are met.

[0007] In one implementation of the first aspect, the scene feature sequence and the basic multi-head attention matrix are subjected to attention interaction to obtain a scene feature attention branch matrix, including: determining a scene feature attention weight vector based on the scene feature sequence and the basic multi-head attention matrix; and determining a scene feature attention branch matrix based on the scene feature attention weight vector and the basic multi-head attention matrix.

[0008] In one implementation of the first aspect, the user preference sequence and the basic multi-head attention matrix are subjected to attention interaction to obtain a user preference attention branch matrix, including: determining a user preference attention weight vector based on the user preference sequence and the basic multi-head attention matrix; and determining a user preference attention branch matrix based on the user preference attention weight vector and the basic multi-head attention matrix.

[0009] In one implementation of the first aspect, the expression for obtaining the adaptive fusion coefficients is:

[0010]

[0011]

[0012] Where sim(·) is the cosine similarity function, Represents a sequence of scene features. Represents a sequence of user actions. ω represents the user preference sequence, E represents the current ambient light intensity, 100 lux represents the preset light-dark boundary threshold, and ω represents the environmental weight factor.

[0013] In one implementation of the first aspect, determining intelligent lighting control parameters based on the fused attention matrix includes: inputting the fused attention matrix into a fully connected layer to obtain fully connected control parameters output by the fully connected layer; and processing the fully connected control parameters through an activation function to generate intelligent lighting control parameters.

[0014] In one implementation of the first aspect, when performing intelligent lighting control on the intelligent lighting terminal based on the intelligent lighting control parameters, the method further includes: collecting user feedback information on the control results; concatenating the feedback information into the user behavior sequence to iterate the user preference sequence; and correcting the attention weight of the user preference sequence according to the feedback information to achieve closed-loop control.

[0015] In one implementation of the first aspect, the method further includes: performing preprocessing operations on the collected environmental parameter sequence, user behavior sequence, scene feature sequence, and user preference sequence to eliminate the influence of units.

[0016] Secondly, this application provides an intelligent lighting control device, the device comprising: a multi-dimensional sequence data acquisition module, used to acquire environmental parameter sequences, user behavior sequences, scene feature sequences, and user preference sequences required for intelligent lighting control; a basic multi-head attention matrix determination module, used to perform linear projection operations on the environmental parameter sequences and the user behavior sequences to obtain a basic multi-head attention matrix; a scene feature attention branch matrix determination module, used to perform attention interaction between the scene feature sequences and the basic multi-head attention matrix to obtain a scene feature attention branch matrix; a user preference attention branch matrix determination module, used to perform attention interaction between the user preference sequences and the basic multi-head attention matrix to obtain a user preference attention branch matrix; an adaptive fusion coefficient acquisition module, used to acquire adaptive fusion coefficients; a fusion attention determination module, used to dynamically fuse the scene feature attention branch matrix and the user preference attention branch matrix based on the adaptive fusion coefficients to obtain a fusion attention matrix; an intelligent lighting control parameter determination module, used to determine intelligent lighting control parameters based on the fusion attention matrix; and an intelligent lighting control module, used to perform intelligent lighting control on an intelligent lighting terminal based on the intelligent lighting control parameters.

[0017] Thirdly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the intelligent lighting control method described in any one of the first aspects of embodiments of this application.

[0018] Fourthly, embodiments of this application provide an electronic device, the electronic device comprising: a memory storing a computer program; and a processor communicatively connected to the memory, which executes the intelligent lighting control method described in any one of the first aspects of embodiments of this application when the computer program is invoked.

[0019] As described above, the intelligent lighting control method, apparatus, device, and storage medium of this application have the following beneficial effects:

[0020] 1) The innovative improved Transformer attention mechanism in this application constructs a dual-branch fusion attention module of "scene features-user preferences", which solves the problems of weak multi-factor coordination ability, insufficient personalization and disconnect between the weight allocation of traditional attention mechanism and the lighting control scene in traditional intelligent lighting control. It realizes the accurate matching of scene requirements and user preferences, and improves the adaptability, personalization and energy efficiency of intelligent lighting control. Through formulaic attention weight allocation and adaptive fusion, the accuracy of lighting control is improved, ensuring that the personalized needs of different scenes and different users are met.

[0021] 2) This application obtains the scene feature attention branch matrix through scene feature attention weight vector and basic multi-head attention matrix, realizing fine-grained decoupling and differentiated focus of features within the scene; the system can autonomously identify key influencing factors in the current scene (e.g., in the "conference scene", it accurately identifies that the lighting weight of the "front stage presentation area" is much higher than that of the "back row rest area"), avoiding indiscriminate global feature extraction, and providing a weight benchmark with clear direction for subsequent precise control; by first extracting the adaptive weight vector within the scene, and then using the weight vector to conditionally reconstruct the basic attention output, redundant noise in the scene is effectively filtered out, generating a high-purity and highly directional dynamic feature representation of the scene, laying a data foundation for subsequent precise adaptive fusion of multiple branches.

[0022] 3) The method for determining the user preference attention branch matrix in this application breaks the traditional network's fixed encoding mode of user features and realizes dynamic feature reconstruction based on individual differences. The multi-head attention mechanism no longer processes data in a one-size-fits-all manner, but performs adaptive feature amplification and suppression based on each user's unique "weight vector". This enables the final generated user preference branch matrix to accurately reflect the micro-intention of a specific user in a specific state; and provides a high-purity user-side data base for efficient and accurate fusion with scene features.

[0023] 4) The method for generating intelligent lighting control parameters in this application achieves precise decoupling and dimensionality reduction mapping from a high-dimensional multimodal semantic space to a low-dimensional physical control space. The fully connected layer transforms the abstract fusion features containing complex scenes and user intentions into concrete, continuous numerical quantities (fully connected control parameters) that can be recognized by the underlying actuators, completely opening up the data link between the perception decision layer and the physical control layer; it constructs a safety constraint boundary for control commands, avoiding the impact of illegal parameters on hardware devices. Through the nonlinear transformation of the activation function, the unbounded continuous values ​​are forcibly truncated or mapped to the legal physical range supported by the intelligent lighting hardware, fundamentally ensuring the physical security and operational stability of the control system; by realizing the dimensionality reduction mapping from high-dimensional fusion features to physical control quantities through the fully connected layer, and using the activation function to construct a physical safety boundary, it not only opens up the link from decision to execution, but also strictly constrains the control parameters within the legal range of the hardware, effectively avoiding device abnormalities caused by illegal commands, while ensuring the smoothness of lighting state switching and visual comfort.

[0024] 5) The closed-loop control method corresponding to the intelligent lighting control method in this application realizes the transformation of the user preference model from static solidification to dynamic growth, effectively eliminating the lag effect of historical offline data. By directly injecting real-time feedback as the latest time step into the behavior sequence, the system can capture the user's instantaneous intention changes at all times, ensuring that the user profile is always close to the user's true current state. It endows the model with the ability to correct deviations online and self-evolve parameters, avoiding repetitive control errors. It directly intervenes and corrects the allocation logic of attention weights based on feedback information, forcing the network to reduce the attention of ineffective or negative features and improve the response sensitivity of effective features. It constructs a rigorous negative feedback closed-loop control link, ensuring the long-term stability and strategy convergence of the control system. By using the real control results (feedback) of the physical world as the output of the calibration anchor and reverse constraint decision model, it effectively suppresses the accumulation of prediction errors caused by environmental disturbances or user preference drift, ensuring that the system's control accuracy continuously converges and optimizes during dynamic operation. Attached Figure Description

[0025] Figure 1A The diagram shown illustrates the application scenario of the intelligent lighting control method provided in this application embodiment.

[0026] Figure 1B The flowchart shown is a process for an intelligent lighting control method provided in an embodiment of this application.

[0027] Figure 2 The flowchart shown is a process for determining the attention branch matrix of scene features provided in an embodiment of this application.

[0028] Figure 3 The flowchart shown is a process for determining the user preference attention branch matrix provided in an embodiment of this application.

[0029] Figure 4 The flowchart shown is a process for generating intelligent lighting control parameters provided in an embodiment of this application.

[0030] Figure 5 The flowchart shown is a closed-loop control flowchart corresponding to the intelligent lighting control method provided in the embodiments of this application.

[0031] Figure 6 The diagram shown is a structural diagram of the intelligent lighting control device provided in an embodiment of this application.

[0032] Figure 7 The diagram shown is a structural diagram of an electronic device provided in an embodiment of this application.

[0033] Component designation explanation

[0034] S11~S18 step 65 Adaptive fusion coefficient acquisition module S21~S22 step 66 Fusion Attention Determination Module S31~S32 step 67 Intelligent lighting control parameter determination module S41~S42 step 68 Intelligent lighting control module S51~S52 step 70 electronic devices 60 Intelligent lighting control device 71 processor 61 Multi-dimensional sequence data acquisition module 72 Non-volatile storage media 62 Basic Multi-Head Attention Matrix Determination Module 73 System bus 63 Scene Feature Attention Branch Matrix Determination Module 74 Internal memory 64 User preference attention branch matrix determination module 75 Network interface Detailed Implementation

[0035] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0036] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. Therefore, the drawings only show the components related to this application and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0037] The following embodiments of this application provide intelligent lighting control methods, devices, equipment, and storage media, including but not limited to the hardware application scenarios listed in these embodiments, such as... Figure 1A As shown in the diagram, this embodiment provides an application scenario illustration of an intelligent lighting control method, specifically including: a data acquisition sensor, an electronic device, and an intelligent lighting terminal. The data acquisition sensor, electronic device, and intelligent lighting terminal are connected in pairs. The data acquisition sensor comprises a light sensor, a human infrared sensor, a temperature and humidity sensor, and a user interaction terminal, used to collect environmental parameter sequences, user behavior sequences, scene feature sequences, and user preference sequences. The collected environmental parameter sequences, user behavior sequences, scene feature sequences, and user preference sequences are sent to the electronic device for data processing to obtain intelligent lighting control parameters; these parameters are then sent to the intelligent lighting terminal to achieve intelligent lighting control. Furthermore, when intelligently controlling the intelligent lighting terminal based on the intelligent lighting control parameters, the data acquisition sensor collects user feedback information on the control results; this feedback information is concatenated to the user behavior sequence to iterate the user preference sequence, and the attention weight of the user preference sequence is corrected based on the feedback information to achieve closed-loop control.

[0038] The technical solutions in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0039] like Figure 1B As shown in the flowchart, this application provides a method for intelligent lighting control. Figure 1B As shown, the intelligent lighting control method provided in this application includes the following steps S11 to S18.

[0040] S11 collects the environmental parameter sequence, user behavior sequence, scene feature sequence, and user preference sequence required for intelligent lighting control.

[0041] The environmental parameter sequence X includes ambient light intensity x1, ambient temperature x2, and ambient humidity x3. The sampling frequency is 1 time / minute, forming the sequence X = [x1, x2, x3]ᵀ∈R³ˣᵀ (T is the sampling time length).

[0042] The user behavior sequence Y includes whether the user is present (y1), the type of user activity (y2, such as reading / resting / working), and the user's manual adjustment record (brightness and color temperature adjustment values), forming the sequence Y = [y1, y2, y3]ᵀ ∈ R³ˣᵀ.

[0043] For example, the user behavior sequence Y is: [1 (presence status, 1 indicates presence, 0 indicates absence), reading (activity type, digital conversion: predefined reading = 0.8, rest = 0.3, entertainment = 0.7, here we take 0.8), historical adjustment record (brightness 80%, color temperature 5000K, digital conversion: first normalize then weighted fusion, brightness normalization = 80 / 100 = 0.8, color temperature normalization = (5000-2700) / (6500-2700)≈0.61, fused with brightness weight 0.6 and color temperature weight 0.4, fusion value = 0.8×0.6+0.61×0.4≈0.72, simplified to 0.8 for calculation)].

[0044] Wherein, the scene feature sequence S: predefined feature vectors for different scenes, such as reading scene S1 = [0.8, 0.2, 0.6] (corresponding to brightness weight, color temperature weight, and energy consumption weight respectively), rest scene S2 = [0.3, 0.8, 0.2], forming a sequence S ∈ R³ˣᴷ (K is the number of scene types).

[0045] Wherein, the user preference sequence P is: learned through historical adjustment data, including user preferences for brightness p1, color temperature p2, and energy consumption p3, forming a sequence P ∈ R³ˣᵁ (U is the number of users).

[0046] In some embodiments, the method further includes performing preprocessing operations on the collected environmental parameter sequence, user behavior sequence, scene feature sequence, and user preference sequence to eliminate the influence of units.

[0047] For example, the collected environmental parameter sequences, user behavior sequences, scene feature sequences, and user preference sequences are standardized and preprocessed to eliminate the influence of units. The corresponding preprocessing formula is as follows:

[0048]

[0049] in, For the first The sampling time, the first The raw data of each dimension For the first The mean of data in each dimension For the first The standard deviation of the data in each dimension This is the standardized data after preprocessing.

[0050] S12, perform a linear projection operation on the environmental parameter sequence and the user behavior sequence to obtain the basic multi-head attention matrix.

[0051] For example, a linear projection is performed on the preprocessed environmental parameter sequence X and user behavior sequence Y to obtain the query vector Q, key vector K, and value vector V. The corresponding projection formula is as follows:

[0052]

[0053] in, These are the linear projection weight matrices for the query, key, and value, respectively. These are the bias terms, and the dimensions are set according to the actual data dimensions (e.g., R³ˣ). 64 ).

[0054] Specifically, the formula for calculating the basic multi-head attention matrix is ​​as follows:

[0055]

[0056]

[0057] Where h is the number of attention heads (h=8 in this application). Let i be the projection weight of the i-th attention head. For the dimensions of Q and K (as defined in this application) ), This is a linear projection weight matrix of the multi-head attention output. The softmax function is used to normalize the attention scores to weights between 0 and 1. This represents the basic multi-head attention matrix.

[0058] S13, perform attention interaction between the scene feature sequence and the basic multi-head attention matrix to obtain the scene feature attention branch matrix.

[0059] S14, perform attention interaction between the user preference sequence and the basic multi-head attention matrix to obtain the user preference attention branch matrix.

[0060] S15, obtain the adaptive fusion coefficients.

[0061] In some embodiments, the expression for obtaining the adaptive fusion coefficient is:

[0062]

[0063]

[0064] Where sim(·) is the cosine similarity function, Represents a sequence of scene features. Represents a sequence of user actions. ω represents the user preference sequence, E represents the current ambient light intensity, 100 lux represents the preset light-dark boundary threshold, and ω represents the environmental weight factor.

[0065] S16. Based on the adaptive fusion coefficient, the scene feature attention branch matrix and the user preference attention branch matrix are dynamically fused to obtain the fused attention matrix.

[0066] For example, based on the adaptive fusion coefficients, the scene feature attention branch matrix and the user preference attention branch matrix are dynamically fused to obtain the expression corresponding to the fused attention matrix:

[0067]

[0068] in, This represents the attention branch matrix for scene features. This represents the user preference attention branch matrix. This represents the fused attention matrix.

[0069] S17, determine the intelligent lighting control parameters based on the fused attention matrix.

[0070] S18, intelligent lighting control is performed on the intelligent lighting terminal based on the intelligent lighting control parameters.

[0071] This application provides an intelligent lighting control method. This method innovatively improves the Transformer attention mechanism and constructs a dual-branch fusion attention module of "scene features-user preferences". It solves the problems of weak multi-factor coordination ability, insufficient personalization, and disconnect between the weight allocation of traditional attention mechanisms and lighting control scenarios in traditional intelligent lighting control. It achieves accurate matching between scene requirements and user preferences, and improves the adaptability, personalization and energy efficiency of intelligent lighting control. Through formulaic attention weight allocation and adaptive fusion, it improves the accuracy of lighting control and ensures that the personalized needs of different scenarios and different users are met.

[0072] like Figure 2As shown in the figure, this application embodiment provides a flowchart for determining the scene feature attention branch matrix, as follows: Figure 2 As shown, the method for determining the scene feature attention branch matrix provided in this application embodiment includes the following steps S21 to S22.

[0073] S21, determine the scene feature attention weight vector based on the scene feature sequence and the basic multi-head attention matrix.

[0074] For example, the scene feature sequence S is interacted with the output of the basic multi-head attention matrix to capture the correlation between scene features and the environment and user behavior, and the scene feature attention weights are calculated. The formula is as follows:

[0075]

[0076] in, The dimension of the scene feature sequence S (as defined in this application) ), This represents the scene feature attention weight vector (with dimensions K×T).

[0077] It's important to note that conventional gating primarily targets single-modal sequence dependencies and cannot effectively model cross-modal feature associations (such as the interaction between visual scene features and user behavior features). Scene feature branches, however, need to capture cross-domain associations between "scene-environment-user behavior," and conventional gating lacks mechanisms for cross-modal interaction. Weighted summation is a linear combination and cannot achieve dynamic, non-linear weight allocation. Scene feature branches, through multi-head attention mechanisms (such as MultiHead(QKV) in the formula), can dynamically adjust the contribution of scene features based on semantic relevance between features (such as the attention weight of user behavior to scene elements). This non-linear weight allocation is unattainable with conventional weighted summation. The interaction logic of conventional gating / weighted summation is simple (such as linear weighting or gating switches), while scene feature branches, through the ternary interaction of Q, K, and V, can capture fine-grained feature dependencies (such as user click behavior's attention to specific objects in the scene). This deep interaction is irreplaceable by conventional methods.

[0078] S22, determine the scene feature attention branch matrix based on the scene feature attention weight vector and the basic multi-head attention matrix.

[0079] For example, the expression for determining the scene feature attention branch matrix based on the scene feature attention weight vector and the basic multi-head attention matrix is ​​as follows:

[0080]

[0081] in, This represents the attention branch matrix for scene features.

[0082] This application provides a method for determining the scene feature attention branch matrix. In this method, the scene feature attention branch matrix is ​​obtained through the scene feature attention weight vector and the basic multi-head attention matrix, realizing fine-grained decoupling and differentiated focusing of features within the scene. The system can autonomously identify key influencing factors in the current scene (e.g., in a "conference scene," accurately identifying that the lighting weight of the "front-stage presentation area" is much higher than that of the "back-row rest area"), avoiding indiscriminate global feature extraction and providing a clearly directional weight benchmark for subsequent precise control. By first extracting the adaptive weight vector within the scene and then using this weight vector to conditionally reconstruct the basic attention output, redundant noise in the scene is effectively filtered out, generating a high-purity, highly directional dynamic feature representation of the scene, laying a data foundation for subsequent precise adaptive fusion of multiple branches.

[0083] like Figure 3 As shown in the figure, this application embodiment provides a flowchart for determining the user preference attention branch matrix, as follows: Figure 3 As shown in the embodiments of this application, the method for determining the user preference attention branch matrix includes the following steps S31 to S32.

[0084] S31, determine the user preference attention weight vector based on the user preference sequence and the basic multi-head attention matrix.

[0085] For example, the user preference sequence P is interacted with the output of the basic multi-head attention matrix to learn the matching relationship between user preferences and the real-time scene, and the user preference attention weights are calculated. The formula is as follows:

[0086]

[0087] in, The dimension representing the user preference sequence P (as defined in this application) ), This represents the user preference attention weight vector (with dimensions U×T).

[0088] It should be noted that the training of the user preference sequence P includes: supervised learning using historical user adjustment data (such as adjustment values ​​for brightness, color temperature, and energy consumption), with the adjustment values ​​as labels, training the model to obtain user preference parameters p1, p2, and p3 for brightness, color temperature, and energy consumption, which constitute the sequence P.

[0089] Cold start processing includes: new users are initialized using the average preferences of the group (e.g., the mean of p1, p2, and p3 of all users); new scenarios are initialized based on the similarity of scenario features (e.g., the cosine similarity between the new scenario and the scenarios in S), taking the preferences of similar scenarios as the initial values.

[0090] User / scenario modeling includes: P is modeled independently for each user (each user has exclusive p1, p2, p3), and S is predefined according to the scenario. When calculating the user preference attention weight, the user preference sequence P is interacted with the scenario feature sequence S (e.g., dot product or attention weighting) to achieve joint user-scenario adaptation.

[0091] S32, determine the user preference attention branch matrix based on the user preference attention weight vector and the basic multi-head attention matrix.

[0092] For example, the expression for determining the user preference attention branch matrix based on the user preference attention weight vector and the basic multi-head attention matrix is ​​as follows:

[0093]

[0094] in, This represents the user preference attention branch matrix.

[0095] This application provides a method for determining a user preference attention branch matrix. This method breaks the traditional network's fixed encoding mode of user features and realizes dynamic feature reconstruction based on individual differences. The multi-head attention mechanism no longer processes data in a one-size-fits-all manner, but performs adaptive feature amplification and suppression based on each user's unique "weight vector". This enables the final generated user preference branch matrix to accurately reflect the micro-intentions of a specific user in a specific state. It provides a high-purity user-side data base for efficient and accurate fusion with scene features.

[0096] like Figure 4 As shown in the figure, this application provides a flowchart for generating intelligent lighting control parameters, as follows: Figure 4 As shown, the method for determining intelligent lighting control parameters provided in this application embodiment includes the following steps S41 to S42.

[0097] S41, input the fused attention matrix into the fully connected layer to obtain the fully connected layer output fully connected control parameters.

[0098] Specifically, the fused attention matrix is ​​input into the fully connected layer, and the expression corresponding to the fully connected layer's output fully connected control parameters is obtained as follows:

[0099] All-Connect=

[0100] in, This represents the weight matrix of the fully connected layer. This indicates the bias term of the fully connected layer, and All-Connect indicates the control parameters of the fully connected layer.

[0101] S42, after the fully connected control parameters are processed by the activation function, intelligent lighting control parameters are generated.

[0102] Specifically, after processing the fully connected control parameters through an activation function, the expression for generating the intelligent lighting control parameters is as follows:

[0103]

[0104] The value of L ranges from 0 to 100 (corresponding to the percentage of light brightness), and the value of C ranges from 2700K to 6500K (corresponding to warm light to cool light).

[0105] This application provides a method for generating intelligent lighting control parameters. This method achieves precise decoupling and dimensionality reduction mapping from a high-dimensional multimodal semantic space to a low-dimensional physical control space. The fully connected layer transforms the abstract fusion features containing complex scenes and user intentions into concrete, continuous numerical quantities (fully connected control parameters) that can be recognized by the underlying actuators, completely opening up the data link between the perception decision layer and the physical control layer. It constructs a safety constraint boundary for control commands, avoiding the impact of illegal parameters on hardware devices. Through the nonlinear transformation of the activation function, unbounded continuous values ​​are forcibly truncated or mapped to the legal physical range supported by the intelligent lighting hardware, fundamentally ensuring the physical security and operational stability of the control system. By realizing the dimensionality reduction mapping from high-dimensional fusion features to physical control quantities through the fully connected layer and constructing a physical safety boundary using the activation function, it not only opens up the link from decision to execution but also strictly constrains the control parameters within the legal range of the hardware, effectively avoiding device anomalies caused by illegal commands, while ensuring the smoothness of lighting state switching and visual comfort.

[0106] like Figure 5 As shown in the figure, this application provides a flowchart of the closed-loop control corresponding to an intelligent lighting control method, as follows: Figure 5 As shown, the closed-loop control method corresponding to the intelligent lighting control method provided in this application embodiment includes the following steps S51 to S52.

[0107] S51 collects user feedback on the control results.

[0108] For example, feedback information includes manual secondary adjustments, satisfaction ratings, etc.

[0109] S52, the feedback information is concatenated to the user behavior sequence to iterate the user preference sequence, and the attention weight of the user preference sequence is corrected according to the feedback information to achieve closed-loop control.

[0110] For example, the collected real-time feedback information (including but not limited to user manual adjustment commands, environmental sensor compensation data, and energy consumption index deviations) is subjected to feature extraction and embedding encoding, and converted into a feedback feature vector with fixed dimensions.

[0111] The feedback feature vector is used as the latest time step node and concatenated along the time dimension to the end of the current historical user behavior sequence to generate an expanded behavior sequence; the expanded behavior sequence is then subjected to a time-series update operation (such as sliding window truncation or state vector weighting) to iteratively generate an updated user preference sequence P.

[0112] The updated user preference sequence P is input into the user preference attention branch matrix. The correlation scores between each historical behavior node and the current feedback in the current state are recalculated. The initial updated attention weights are obtained through Softmax normalization.

[0113] Calculate the deviation loss value between the feedback feature vector and the expected control target (or the user's actual intervention intention); determine the correction gradient for the initial updated attention weights based on the deviation loss value; fine-tune the parameters of the initial updated attention weights using the correction gradient (e.g., through attention bias term injection or residual connection correction) to obtain the final corrected attention weights.

[0114] Based on the final corrected attention weights, the updated user preference sequence P is subjected to feature weighting and aggregation. The aggregated features are then input into the control decision network corresponding to the smart lighting terminal to generate the precise control command for the next moment and issue it for execution, thus completing a single closed-loop control link.

[0115] This application provides a closed-loop control method corresponding to an intelligent lighting control method. This method realizes the transformation of the user preference model from static solidification to dynamic growth, effectively eliminating the lag effect of historical offline data. By directly injecting real-time feedback as the latest time step into the behavior sequence, the system can capture the user's instantaneous intention changes at all times, ensuring that the user profile always closely reflects the user's true current state. It endows the model with the ability to correct deviations online and self-evolve parameters, avoiding repetitive control errors. Based on feedback information, it directly intervenes and corrects the attention weight allocation logic, forcing the network to reduce the attention of ineffective or negative features and improve the response sensitivity of effective features. A rigorous negative feedback closed-loop control link is constructed to ensure the long-term stability and strategy convergence of the control system. By using the real control results (feedback) of the physical world as the output of the calibration anchor and reverse constraint decision model, it effectively suppresses the accumulation of prediction errors caused by environmental disturbances or user preference drift, ensuring that the system's control accuracy continuously converges and optimizes during dynamic operation.

[0116] The scope of protection of the intelligent lighting control method described in this application is not limited to the execution order of the steps listed in this embodiment. Any solution implemented by adding, subtracting, or replacing steps in the prior art based on the principles of this application is included within the scope of protection of this application.

[0117] This application also provides an intelligent lighting control device, which can implement the intelligent lighting control method described in this application. However, the implementation device of the intelligent lighting control method described in this application includes, but is not limited to, the structure of the intelligent lighting control device listed in this embodiment. All structural modifications and substitutions of the prior art made based on the principles of this application are included within the protection scope of this application.

[0118] like Figure 6 As shown, in one embodiment, the intelligent lighting control device 60 of this application includes a multi-dimensional sequence data acquisition module 61, a basic multi-head attention matrix determination module 62, a scene feature attention branch matrix determination module 63, a user preference attention branch matrix determination module 64, an adaptive fusion coefficient acquisition module 65, a fusion attention determination module 66, an intelligent lighting control parameter determination module 67, and an intelligent lighting control module 68.

[0119] The multi-dimensional sequence data acquisition module 61 is used to acquire the environmental parameter sequence, user behavior sequence, scene feature sequence and user preference sequence required for intelligent lighting control.

[0120] The basic multi-head attention matrix determination module 62 is used to perform a linear projection operation on the environmental parameter sequence and the user behavior sequence to obtain the basic multi-head attention matrix.

[0121] The scene feature attention branch matrix determination module 63 is used to perform attention interaction between the scene feature sequence and the basic multi-head attention matrix to obtain the scene feature attention branch matrix.

[0122] The user preference attention branch matrix determination module 64 is used to perform attention interaction between the user preference sequence and the basic multi-head attention matrix to obtain the user preference attention branch matrix.

[0123] The adaptive fusion coefficient acquisition module 65 is used to acquire the adaptive fusion coefficient.

[0124] The fusion attention determination module 66 is used to dynamically fuse the scene feature attention branch matrix and the user preference attention branch matrix based on the adaptive fusion coefficient to obtain the fusion attention matrix.

[0125] The intelligent lighting control parameter determination module 67 is used to determine the intelligent lighting control parameters based on the fused attention matrix.

[0126] The intelligent lighting control module 68 is used to perform intelligent lighting control on the intelligent lighting terminal based on the intelligent lighting control parameters.

[0127] The structure and principle of the multi-dimensional sequence data acquisition module 61, the basic multi-head attention matrix determination module 62, the scene feature attention branch matrix determination module 63, the user preference attention branch matrix determination module 64, the adaptive fusion coefficient acquisition module 65, the fusion attention determination module 66, the intelligent lighting control parameter determination module 67, and the intelligent lighting control module 68 correspond one-to-one with the steps in the above-mentioned intelligent lighting control method, so they will not be described in detail here.

[0128] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, or methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules / units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or units may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of apparatuses or modules or units may be electrical, mechanical, or other forms.

[0129] The modules / units described as separate components may or may not be physically separate. The components shown as modules / units may or may not be physical modules; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules / units can be selected to achieve the objectives of the embodiments of this application, depending on actual needs. For example, the functional modules / units in the various embodiments of this application may be integrated into one processing module, or each module / unit may exist physically separately, or two or more modules / units may be integrated into one module / unit.

[0130] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0131] This application also provides an electronic device. Figure 7 The diagram shown is a structural schematic of an electronic device 70 in one embodiment of this application. The intelligent lighting control method provided in this embodiment can be applied to… Figure 7 The electronic device 70 shown is an example, but not limited to it. For example... Figure 7 As shown, the electronic device 70 includes a processor 71, a memory, a system bus 73, and a network interface 75. The memory may include a non-volatile storage medium 72 and internal memory 74.

[0132] The non-volatile storage medium 72 can store an operating system and a computer program. The computer program includes program instructions that, when executed, cause the processor to perform any of the intelligent lighting control methods provided in the embodiments of this application.

[0133] The processor provides computing and control capabilities, supporting the operation of the entire computer device.

[0134] The internal memory 74 provides an environment for the execution of a computer program in a non-volatile storage medium. When the computer program is executed by the processor, it enables the processor to execute any of the intelligent lighting control methods provided in the embodiments of this application.

[0135] This network interface 75 is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 7The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0136] It should be understood that processor 71 can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, the general-purpose processor can be a microprocessor or any conventional processor.

[0137] The electronic device 70 in this application embodiment may include terminal devices such as tablet computers, laptop computers, mobile phones, supercomputers, and smart wearable devices. It can also be applied to databases, servers, and service response systems based on terminal artificial intelligence. This application embodiment does not impose any restrictions on the specific type of electronic device.

[0138] For example, electronic devices can be stations (STAs) in WLANs, cellular phones, cordless phones, Session Initiation Protocol (SIP) phones, Wireless Local Loop (WLL) stations, handheld devices with wireless communication capabilities, computing devices or other processing devices connected to a wireless modem, computers, laptops, handheld communication devices, handheld computing devices, and / or other devices for communicating over wireless systems, as well as next-generation communication systems, such as mobile terminals in 5G networks, mobile terminals in future evolved Public Land Mobile Networks (PLMNs), or mobile terminals in future evolved Non-terrestrial Networks (NTNs).

[0139] This application also provides a computer-readable storage medium. Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing a processor. The program can be stored in a computer-readable storage medium, which is a non-transitory medium, such as random access memory, read-only memory, flash memory, hard disk, solid-state drive, magnetic tape, floppy disk, optical disk, and any combination thereof. The storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., digital video disc (DVD)), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0140] This application embodiment may also provide a computer program product comprising one or more computer instructions. When the computer instructions are loaded and executed on a computing device, all or part of the processes or functions described in this application embodiment are generated. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0141] When the computer program product is executed by a computer, the computer performs the method described in the foregoing method embodiments. The computer program product can be a software installation package; when the foregoing method is required, the computer program product can be downloaded and executed on the computer.

[0142] The descriptions of the processes or structures corresponding to the above figures each have their own emphasis. For parts of a process or structure that are not described in detail, please refer to the relevant descriptions of other processes or structures.

[0143] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A method for intelligent lighting control, characterized in that, The method includes: Collect the environmental parameter sequence, user behavior sequence, scene feature sequence, and user preference sequence required for intelligent lighting control; A linear projection operation is performed on the environmental parameter sequence and the user behavior sequence to obtain the basic multi-head attention matrix; The scene feature sequence and the basic multi-head attention matrix are interacted to obtain the scene feature attention branch matrix; The user preference sequence and the basic multi-head attention matrix are interacted to obtain the user preference attention branch matrix; Obtain adaptive fusion coefficients; The scene feature attention branch matrix and the user preference attention branch matrix are dynamically fused based on the adaptive fusion coefficients to obtain the fused attention matrix; The intelligent lighting control parameters are determined based on the fused attention matrix; The intelligent lighting terminal performs intelligent lighting control based on the aforementioned intelligent lighting control parameters.

2. The method according to claim 1, characterized in that, The scene feature sequence and the basic multi-head attention matrix are interacted to obtain a scene feature attention branch matrix, including: Determine the scene feature attention weight vector based on the scene feature sequence and the basic multi-head attention matrix; The scene feature attention branch matrix is ​​determined based on the scene feature attention weight vector and the basic multi-head attention matrix.

3. The method according to claim 1, characterized in that, The user preference sequence and the basic multi-head attention matrix are interacted to obtain a user preference attention branch matrix, including: Determine the user preference attention weight vector based on the user preference sequence and the basic multi-head attention matrix; The user preference attention branch matrix is ​​determined based on the user preference attention weight vector and the basic multi-head attention matrix.

4. The method according to claim 1, characterized in that, The expression for obtaining the adaptive fusion coefficients is: Where sim(·) is the cosine similarity function, Represents a sequence of scene features. Represents a sequence of user actions. This represents the user preference sequence, where E represents the current ambient light intensity. ω represents the preset threshold for light and dark separation, and ω represents the environmental weight factor.

5. The method according to claim 1, characterized in that, The intelligent lighting control parameters are determined based on the fused attention matrix, including: The fused attention matrix is ​​input into the fully connected layer to obtain the fully connected layer's output fully connected control parameters; The fully connected control parameters are processed by an activation function to generate intelligent lighting control parameters.

6. The method according to claim 1, characterized in that, When performing intelligent lighting control on the intelligent lighting terminal based on the intelligent lighting control parameters, the method further includes: Collect user feedback on the control results; The feedback information is concatenated to the user behavior sequence to iterate the user preference sequence, and the attention weight of the user preference sequence is adjusted according to the feedback information to achieve closed-loop control.

7. The method according to claim 1, characterized in that, The method further includes: Preprocessing operations are performed on the collected environmental parameter sequence, user behavior sequence, scene feature sequence, and user preference sequence to eliminate the influence of units.

8. An intelligent lighting control device, characterized in that, The device includes: The multi-dimensional sequence data acquisition module is used to collect environmental parameter sequences, user behavior sequences, scene feature sequences, and user preference sequences required for intelligent lighting control. The basic multi-head attention matrix determination module is used to perform a linear projection operation on the environmental parameter sequence and the user behavior sequence to obtain the basic multi-head attention matrix. The scene feature attention branch matrix determination module is used to perform attention interaction between the scene feature sequence and the basic multi-head attention matrix to obtain the scene feature attention branch matrix. The user preference attention branch matrix determination module is used to perform attention interaction between the user preference sequence and the basic multi-head attention matrix to obtain the user preference attention branch matrix. The adaptive fusion coefficient acquisition module is used to acquire the adaptive fusion coefficient. The fusion attention determination module is used to dynamically fuse the scene feature attention branch matrix and the user preference attention branch matrix based on the adaptive fusion coefficient to obtain a fusion attention matrix; The intelligent lighting control parameter determination module is used to determine the intelligent lighting control parameters based on the fused attention matrix. The intelligent lighting control module is used to perform intelligent lighting control on the intelligent lighting terminal based on the intelligent lighting control parameters.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

10. An electronic device, characterized in that, The electronic device includes: A memory that stores a computer program; The processor, which is communicatively connected to the memory, executes the method of any one of claims 1 to 7 when the computer program is invoked.

Images