A light source energy-saving control method and system based on a space-time model

Abstract

The application discloses a kind of lamp source energy-saving control method and system based on space-time model, it is related to lamp source control technical field.A kind of lamp source energy-saving control system based on space-time model, including have: space-time model construction module and lamp source energy-saving control module.The application is by introducing traffic flow, environment and illumination and other multi-source space-time perception data, constructs lamp source control space-time model, makes lamp source control no longer rely on fixed time period or single parameter, but can dynamically reflect the real lighting demand under different regions, different time and different environmental conditions, fundamentally improves the intelligent and fine level of lamp source control;Through point mapping and establishing lamp source grid unit to lamp source in geographic information model, upgrade grid collaborative control to traditional single lamp control, not only improve the scalability of large-scale lamp source management, also provide unified space carrier for regional cascade dimming and hierarchical control, effectively avoid brightness mutation and illumination split phenomenon.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of lamp source control, in particular to a lamp source energy-saving control method and system based on a space-time model. BACKGROUND

[0002] Under the background of global sustainable development, energy consumption and environmental problems are increasingly prominent. Traditional lamp source control often uses timing switch or unified dimming mode, without fully considering geographical location, time variation and environmental differences. As a result, situations such as full-power lighting in low-demand sections at night, and not dimming in time when there is sufficient natural light often occur, leading to serious energy waste and difficulty in dealing with lamp aging and changes in surrounding lighting conditions.

[0003] In order to improve the energy-saving level of the lighting system, an intelligent control method that can integrate space-time geographical information and lamp characteristics is needed to finely and differentially adjust the lamp source under the premise of meeting the lighting safety and use requirements, thereby effectively reducing energy consumption and carbon emissions. SUMMARY

[0004] The present application aims to provide a lamp source energy-saving control method and system based on a space-time model, which effectively reduces lamp source energy consumption while ensuring lighting uniformity, safety and visual comfort, and achieves the overall energy-saving goal.

[0005] A lamp source energy-saving control method based on a space-time model, comprising the following steps:

[0006] There are a plurality of to-be-adjusted lamp sources in the lamp source control area; the space-time perception data corresponding to each to-be-adjusted lamp source in the lamp source control area is obtained; each to-be-adjusted lamp source in the lamp source control area is mapped in a preset GIS model to obtain a lamp source control node; a lamp source control space-time model is established according to the space-time perception data; wherein the space-time perception data includes traffic space-time data, environmental space-time data and illumination space-time data; for each to-be-adjusted lamp source in the lamp source control area, the lamp source control space-time model is used for grid division to establish a lamp source grid unit;

[0007] The lighting demand of each lamp source grid unit is modeled to obtain a lamp source lighting demand heat map; in the lamp source lighting demand heat map, there are lamp source grid units and corresponding lamp source grid demand characteristics;

[0008] The lamp source control smoothing index between the lamp source grid units is determined in the lamp source lighting demand heat map; based on the lamp source lighting demand heat map, the lamp source control area is controlled in combination with the lamp source grid units to obtain a lamp source energy-saving control strategy.

[0009] As a preferred technical solution of the present application, the specific steps of establishing a lamp source control space-time model according to space-time perception data include:

[0010] The spatiotemporal sensing data is matched with the geographic coordinates and timestamps of the light sources to be adjusted within the light source control area to obtain the spatiotemporal feature vector of the light source control. The local spatial gradient of the light source and the average spatiotemporal features of the light source area are calculated based on the spatiotemporal feature vector of the light source control. Periodic analysis is performed on the spatiotemporal feature vector of the light source control to obtain the periodic pattern features of the light source.

[0011] In the GNN-RNN framework, calculations are performed based on the local spatial gradient of the light source, the periodic pattern features of the light source, and the average spatiotemporal features of the light source region to obtain the comprehensive feature vector of light source control for each light source control node within the light source control region; a spatiotemporal model of light source control is constructed based on all comprehensive feature vectors of light source control.

[0012] Among them, a prediction model framework is introduced into the spatiotemporal model of light source control. The prediction model framework is used to predict the comprehensive feature vector of light source control for each light source control node in the future time window.

[0013] As a preferred technical solution of the present invention, the specific steps for modeling the illumination requirements of each lamp source grid cell include:

[0014] Obtain the comprehensive feature vector of light source control for each light source control node output in the spatiotemporal model of light source control within a future time window; perform feature fusion on the comprehensive feature vectors of light source control for all light source control nodes in the light source grid cell within the future time window to obtain the comprehensive feature vector of light source grid control corresponding to the light source grid cell.

[0015] Semantic parsing is performed on the comprehensive feature vector of the light source grid control within the future time window to obtain the predicted spatiotemporal demand features of lighting; at the same time, real-time light source response demand is obtained based on the sensors within the light source control area; the predicted spatiotemporal demand features of lighting and the real-time light source response demand are fused to obtain the light source grid demand features corresponding to the light source grid unit.

[0016] A graph is constructed by combining all light source grid cells and their corresponding light source grid demand characteristics to obtain a heat map of light source illumination demand.

[0017] As a preferred embodiment of the present invention, the specific steps for determining the lamp source control smoothing index between lamp source grid units in the lamp source illumination demand heat map include:

[0018] The light source grid units are clustered according to functional zoning to obtain several basic grid unit clusters; for each basic grid unit cluster, the comprehensive feature vectors of light source control corresponding to all light source grid units within the cluster are extracted and their attributes are fused to obtain the comprehensive feature vector of the grid cluster.

[0019] The basic grid cell clusters are divided using spectral clustering and hierarchical clustering algorithms. When the functional similarity between multiple adjacent basic grid cell clusters exceeds a preset first threshold, several large grid lighting regions are obtained. When the functional similarity between a basic grid cell cluster and its adjacent basic grid cell clusters is less than a preset second threshold, several small grid lighting regions are obtained. Otherwise, the original basic grid cell clusters are retained. At the same time, the small grid lighting regions are local light source control units within the large grid lighting regions.

[0020] Identify the lighting coordination control boundaries between adjacent large grid lighting areas and between large grid lighting areas and their internal small grid lighting areas to obtain the light source control smoothness index.

[0021] As a preferred embodiment of the present invention, the specific steps for identifying the lighting coordination control boundaries between adjacent large grid lighting areas and between a large grid lighting area and its internal small grid lighting areas include:

[0022] Identify the light source grid cells located at the illumination cooperative control boundary to obtain the boundary grid set; calculate the feature distance of the comprehensive feature vector of the grid cluster between adjacent large grid illumination areas to obtain the grid control feature distance; if the grid control feature distance is greater than the preset gradient threshold, the corresponding illumination cooperative control boundary is determined to be a strong cooperative boundary, otherwise it is a weak cooperative boundary;

[0023] For the boundary mesh set on the strong cooperative boundary, the boundary mesh set is assigned a higher boundary transition weight than the boundary inside the large mesh lighting area based on the mesh control feature distance, and the large mesh light source control smoothing index is calculated; for the boundary mesh set on the weak cooperative boundary, the large mesh basic illumination smoothing coefficient is set.

[0024] For small grid lighting areas within a large grid lighting area, the deviation of the comprehensive feature vector of the grid cluster within the small grid lighting area from the lighting feature of its parent large grid lighting area is calculated. A light source control smoothing index for the small grid is generated based on this deviation. Specifically, when the deviation of the lighting feature of the small grid lighting area is a positive high-demand deviation, the light source control smoothing index of the parent large grid is superimposed to obtain the small grid light source control smoothing index for the small grid lighting area. When the deviation of the lighting feature of the small grid lighting area is not a positive high-demand deviation, a basic illumination smoothing coefficient for the small grid is set.

[0025] The light source control smoothing index is obtained by combining all the large grid light source control smoothing indices, large grid basic illumination smoothing coefficients, small grid light source control smoothing indices, and small grid basic illumination smoothing coefficients.

[0026] As a preferred embodiment of the present invention, the specific steps for controlling all adjustable light sources within the light source control area based on a heat map of light source illumination demand combined with light source grid units include:

[0027] The heat map of light source illumination demand is analyzed to identify the characteristic values ​​of light source grid demand for each light source grid unit and construct an illumination demand amplitude array; the illumination demand amplitude array is used as the initial illumination control signal source.

[0028] The light signal penetration coefficient is calculated based on the light source control smoothing index. Specifically, when the light source control smoothing index indicates a large grid light source control smoothing index, the light signal penetration coefficient is a high light penetration coefficient; when the light source control smoothing index indicates a small grid light source control smoothing index, the light signal penetration coefficient is a low light penetration coefficient; otherwise, the light signal penetration coefficient is the basic light penetration coefficient.

[0029] A light source control diffusion network is constructed based on the light signal penetration coefficient. The initial light control signal source is used as input, and the signal intensity is iteratively calculated in the light source control diffusion network to generate the final light response envelope within the light source control area.

[0030] The values ​​in the light response envelope are mapped to the adjustment range set by the light source to be adjusted, thus obtaining the energy-saving control strategy for each light source to be adjusted.

[0031] A light source energy-saving control system based on a spatiotemporal model includes:

[0032] The spatiotemporal model construction module includes a model construction unit and a grid division unit. The model construction unit is used when there are several light sources to be adjusted within the light source control area. It acquires the spatiotemporal perception data corresponding to each light source to be adjusted within the light source control area. It maps each light source to be adjusted within the light source control area to a point in a preset GIS model to obtain light source control nodes. It establishes a spatiotemporal model for light source control based on the spatiotemporal perception data. The spatiotemporal perception data package contains traffic flow spatiotemporal data, environmental spatiotemporal data, and illumination spatiotemporal data. The grid division unit is used to divide the light source into grids using the light source control spatiotemporal model for each light source to be adjusted within the light source control area, and establishes light source grid units.

[0033] The lamp source energy-saving control module includes a feature recognition unit and an energy-saving control unit. The feature recognition unit is used to model the illumination demand of each lamp source grid unit to obtain a lamp source illumination demand heat map. The lamp source illumination demand heat map contains lamp source grid units and corresponding lamp source grid demand features. The energy-saving control unit is used to determine the lamp source control smoothing index between lamp source grid units in the lamp source illumination demand heat map. Based on the lamp source illumination demand heat map and the lamp source grid units, all lamp sources to be adjusted within the lamp source control area are controlled to obtain the lamp source energy-saving control strategy.

[0034] The present invention has the following advantages:

[0035] 1. This invention introduces multi-source spatiotemporal sensing data such as traffic flow, environment, and illumination to construct a spatiotemporal model for lighting source control. This enables lighting source control to move beyond fixed time periods or single parameters and dynamically reflect the real lighting needs under different regions, times, and environmental conditions, fundamentally improving the intelligence and precision of lighting source control. By mapping lighting sources to their locations in a geographic information model and establishing lighting source grid units, traditional single-lamp control is upgraded to grid-based collaborative control. This not only improves the scalability of large-scale lighting source management but also provides a unified spatial carrier for regional-level coordinated dimming and hierarchical control, effectively avoiding sudden brightness changes and lighting fragmentation.

[0036] 2. This invention utilizes the comprehensive feature vector output by the spatiotemporal model of lamp source control to model the illumination demand of lamp source grid units and generate an illumination demand heat map, enabling lighting demand to be expressed in an intuitive and quantitative way, which is conducive to accurately matching lighting supply with actual demand. By constructing a lamp source control formula that includes maximum control wattage, lamp source control weight, lamp source demand weight, and energy-saving coefficient, this invention achieves real-time and continuous adjustment of each lamp source to be adjusted while ensuring the safe operating range of the lamps, so that the overall energy-saving target can be smoothly and stably implemented at the individual lamp control level. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of a lamp source energy-saving control system based on a spatiotemporal model used in an embodiment of the present invention. Detailed Implementation

[0038] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this invention.

[0039] Example 1: In this example, taking a city road lighting control area as an example, a lamp source energy-saving control method based on a spatiotemporal model includes the following steps:

[0040] There are several light sources to be adjusted within the light source control area. Spatiotemporal sensing data corresponding to each light source to be adjusted within the light source control area is acquired. Each light source to be adjusted within the light source control area is mapped to a point in a pre-defined GIS model to obtain a light source control node. A spatiotemporal model for light source control is established based on the spatiotemporal sensing data. The spatiotemporal sensing data package contains traffic flow spatiotemporal data, environmental spatiotemporal data, and illumination spatiotemporal data. For each light source to be adjusted within the light source control area, a grid is created using the light source control spatiotemporal model to establish a light source grid unit.

[0041] The specific implementation method for acquiring spatiotemporal perception data for each adjustable light source within the light source control area is as follows: data is collected through a multi-dimensional heterogeneous sensor network deployed within the light source control area. This multi-dimensional heterogeneous sensor network includes millimeter-wave radar installed on streetlight poles, high-definition video surveillance probes, illuminance sensors, and environmental meteorological monitoring stations. Specifically, the vehicle flow spatiotemporal data is extracted in real-time using millimeter-wave radar and video stream analysis technology. This data includes traffic flow figures (representing the total number of vehicles passing through per unit time), average driving speed (representing the degree of road congestion and the required visual reaction distance), and lane occupancy rate (representing the spatial load status of the road). All of these data are accompanied by precise time data. The data includes stamps and road segment location tags; environmental spatiotemporal data is acquired through meteorological monitoring stations and light sensors, specifically including ambient light intensity values ​​(representing the contribution of natural light to illumination), atmospheric visibility distance (representing the degree of light attenuation due to haze or rain / snow), and rainfall data (representing potential factors affecting changes in road surface reflectivity); illumination spatiotemporal data is transmitted back through the intelligent controller inside the luminaire, including the current output power, voltage and current status, and lighting duration of the luminaire, representing the basic operating status and energy consumption baseline of the light source; after collection, the above spatiotemporal perception data undergoes time synchronization calibration and outlier cleaning, and is integrated into a dataset with unified spatiotemporal labels, serving as the input basis for subsequent model building.

[0042] The specific implementation method for mapping each adjustable light source within the light source control area to a pre-set GIS model to obtain the light source control node is as follows: A point mapping algorithm based on coordinate projection transformation and road network topology is adopted. First, a pre-set Geographic Information System (GIS) base map is loaded, which contains the precise road outline, lane centerline, and functional zoning vector layer of the corresponding light source control area. The original latitude and longitude coordinates reported by the built-in GPS module of each adjustable light source are read, and the Gauss-Kruger projection method is used to convert the spherical latitude and longitude coordinates into plane rectangular coordinates to eliminate geographic projection distortion. Impact on distance calculation; Execute a map matching algorithm based on a hidden Markov model to match and correct the transformed planar coordinates with the nearest neighbor road grid in the GIS model, correcting the positioning error caused by GPS signal drift, and ensuring that the virtual point accurately falls on the logical position of the road edge; Instantiate the corrected point into a light source control node. The light source control node not only contains geometric location information, but also automatically establishes topological connection relationships with adjacent light source nodes through spatial indexing algorithms (such as R-tree index), thereby generating a virtual control node object in the digital space that has both geographical attributes and network topology attributes.

[0043] The specific steps for establishing a spatiotemporal model for light source control based on spatiotemporal perception data include:

[0044] Based on the geographical coordinates and timestamps of the light sources to be adjusted within the light source control area, the spatiotemporal sensing data is matched to obtain the spatiotemporal feature vector of the light source control. A spatial index library centered on the light source control nodes is constructed, and a time tolerance window (e.g., 5 seconds before and after) and a spatial neighborhood radius (e.g., the distance covering the width of a single lane) are set. For each inflow of traffic, environmental, or illumination data, the Euclidean distance between its generation location and each light source control node is calculated, and the data is associated with the nearest light source node whose timestamp is within the tolerance window. The successfully matched heterogeneous data are vectorized and concatenated. Specifically, the normalized traffic flow, vehicle speed, ambient illuminance, rainfall, and other physical quantities are arranged in a predetermined order to construct a multidimensional array, namely the light source control spatiotemporal feature vector. The light source control spatiotemporal feature vector mathematically represents the microscopic operating environment state of a specific light source at a specific moment, providing a standardized input format for subsequent model calculations.

[0045] Based on the spatiotemporal feature vector of light source control, the local spatial gradient and average spatiotemporal characteristics of the light source control area are calculated. To capture the drastic spatial changes in illumination demand, a discrete difference algorithm is used to calculate the local spatial gradient of the light source. Specifically, for any light source control node, its directly connected neighboring nodes in the topology are retrieved, and the numerical difference between the node and its neighboring nodes in the spatiotemporal feature vector of light source control (such as traffic flow or illuminance dimension) is calculated. This numerical difference is then divided by the physical distance between the nodes to obtain the gradient vector representing the rate of spatial change, i.e., the local spatial gradient of the light source. The local spatial gradient of the light source reflects the abrupt changes in traffic flow or illumination on the road segment. At the same time, to obtain a macroscopic background reference, a global mean statistical algorithm is used to calculate the average spatiotemporal characteristics of the light source area. This involves arithmetically averaging the spatiotemporal feature vectors of light source control of all nodes within the light source control area at the same time to obtain a benchmark vector representing the overall average level of the area, i.e., the average spatiotemporal characteristics of the light source area, which is used to subsequently determine the degree of deviation of the state of a single point relative to the overall environment.

[0046] Periodic analysis is performed on the spatiotemporal feature vector of light source control to obtain the light source periodic pattern features. The historical spatiotemporal feature vector sequence of the light source control node over the past several days (e.g., the past 30 days) is retrieved, and the time series is separated into long-term trend components, seasonal periodic components, and random residual components using the Seasonal Trend Decomposition (STL decomposition) process. The seasonal periodic components are extracted to identify the daily variation pattern with a 24-hour period (e.g., the distribution of morning and evening peak hours) and the weekly variation pattern with a 7-day period. These waveform data reflecting time patterns are parameterized to generate the light source periodic pattern features. The light source periodic pattern features can quantify whether the current moment is in a peak, trough, or transitional phase of traffic patterns, providing prior knowledge for the prediction model.

[0047] In the GNN-RNN framework, calculations are performed based on the local spatial gradient of the light source, the periodic pattern features of the light source, and the average spatiotemporal features of the light source region to obtain the comprehensive feature vector of light source control for each light source control node within the light source control area. A deep spatiotemporal fusion neural network framework is constructed, which combines the aggregation ability of graph neural networks (GNN) for topological spatial information with the memory ability of recurrent neural networks (RNN) for time series. In the specific calculation process, firstly, the local spatial gradient of the light source is processed using graph neural network layers, and the gradient information of the current node and its neighboring nodes is convolved and aggregated to capture the spatial mutual influence; simultaneously, ... The periodic pattern features of the light source and the average spatiotemporal features of the light source region are used as contextual auxiliary inputs. They are processed temporally through gated recurrent units (GRU) or long short-term memory networks (LSTM) to update the hidden state of the nodes. After nonlinear transformation and feature fusion of multiple layers of networks, the GNN-RNN framework finally outputs a high-dimensional dense vector, namely the comprehensive feature vector of light source control. The comprehensive feature vector of light source control is no longer a simple stack of physical quantities, but a deep semantic representation containing spatiotemporal dependencies encoded by deep learning. These vector sets of all nodes together constitute the digital kernel of the spatiotemporal model of light source control.

[0048] A spatiotemporal model for light source control is constructed based on the comprehensive feature vectors of all light source control systems. This model, in terms of data structure, is a graph structure object containing topological relationships and dynamic attributes. It abstracts the physical road network connections as edges in graph theory and each light source control node as a vertex. The specific process of constructing the spatiotemporal model involves using the comprehensive feature vectors of light source control, which contain deep semantic information and are calculated in the previous steps, as attribute values. These vectors are mapped and filled into the corresponding vertices of the graph structure, while the adjacency matrix of the road network is used to define the association weights between vertices. At this point, the spatiotemporal model is not merely a static map, but a mathematical entity that represents the current instantaneous state of the entire regional lighting network. It exists in computer memory as a high-dimensional matrix or tensor, where each row corresponds to a light source node and each column corresponds to a dimension of the feature vector. This matrix can be directly read and processed by deep learning algorithms, thus achieving a complete mapping from physical space state to digital vector space.

[0049] In this study, a predictive model framework is introduced into the spatiotemporal model of light source control. This framework is used to predict the comprehensive feature vector of light source control for each light source control node within a future time window. The predictive model framework is a deep learning submodule embedded in the aforementioned spatiotemporal model of light source control, used for time series extrapolation. It typically adopts a sequence-to-sequence (Seq2Seq) architecture or a Transformer decoder architecture. The mechanism of this framework is not a simple linear extrapolation, but is based on the principle of memory-generation. It utilizes the historical hidden states maintained by recurrent neural network units (i.e., the model's memory encoding of traffic and environmental changes over a period of time), combined with the current comprehensive feature vector of light source control, and calculates the state transition probability through a nonlinear activation function. The core task of the predictive model framework is to learn the evolution law of the comprehensive feature vector of light source control on the time axis, that is, to understand how complex features such as traffic density and ambient light change over time, thereby giving the model the ability to predict the future.

[0050] The prediction process employs an autoregressive iterative approach. First, the integrated feature vector of the light source control at the current time t and the current hidden state of the model are input into the decoder of the prediction model framework. The decoder outputs the predicted feature vector for the next time t+1. Next, the system uses this predicted vector at time t+1 as a pseudo-true value and inputs it back into the decoder to calculate the vector at time t+2. This process is repeated until the set future time window (e.g., the next 30 minutes) is covered. During this process, the prediction model framework also incorporates the spatial aggregation function of graph neural networks to ensure that when predicting the future state of a node, not only its own time trend is considered, but also the predicted future states of its neighboring nodes are integrated, thereby generating a future feature sequence with spatiotemporal consistency.

[0051] The specific implementation method for establishing light source grid units by meshing using the light source control spatiotemporal model for each light source to be adjusted within the light source control area is as follows: the continuous physical road space is discretized into non-overlapping logical control units that cover the entire area. Specifically, a Thiessen polygon generation algorithm based on road network constraints is adopted. The geometric coordinates of all light source control nodes are extracted from the light source control spatiotemporal model as generators, and Thiessen polygons are constructed on a two-dimensional plane, such that the distance from any point within each polygon to the corresponding light source node is less than the distance to any other light source node, thereby establishing the initial value of each light source. The system governs the surrounding area; it introduces road geographic information data from the GIS model and performs Boolean operations to find the intersection; it then uses the actual road edge contours to trim the generated Thiessen polygons, removing non-road areas (such as buildings or green belts), resulting in the final irregular geometric shape that is the light source grid unit; finally, it instantiates the grid unit as a data object and establishes an index association between it and the corresponding light source control comprehensive feature vector, so that each grid unit not only has a clear spatial geographic boundary, but also inherits the spatiotemporal perception data attributes of the area, serving as a spatial carrier for subsequent refined lighting demand calculations.

[0052] Lighting demand modeling is performed for each light source grid cell to obtain a light source lighting demand heatmap; the light source lighting demand heatmap contains light source grid cells and corresponding light source grid demand features;

[0053] The specific steps for modeling the illumination requirements of each light source grid cell include:

[0054] Obtain the comprehensive feature vector of light source control for each light source control node output in the spatiotemporal model of light source control within a future time window; extract multiple predicted time points of the future time window with a uniform time granularity (e.g., every 1 minute or every 5 minutes), extract the comprehensive feature vector of the corresponding time for each node and complete time alignment to obtain the comprehensive feature vector of light source control.

[0055] The comprehensive feature vectors of light source control for all light source control nodes within a future time window are fused to obtain the comprehensive feature vector of light source grid control for that light source grid unit. Based on spatial affiliation, all light source control nodes within the same light source grid unit are aggregated into a node set. Then, the vector sequence of this set within the future time window is fused: First, node-dimensional fusion is performed for each future time point, which can be done using weighted summation or attention-weighted fusion (for example, the weights are obtained by evaluating the importance of the nodes, such as normalizing the weight vector formed by road grade, node historical failure rate, node coverage area, or lamp rated power), to obtain the grid-level vector at that time. Then, time-dimensional fusion is performed on the grid-level vectors of multiple times within the future time window, for example, using time decay weighted averaging (the closer to the current time, the higher the weight) or trend extraction (calculating statistics such as slope and volatility and concatenating them with the mean vector), finally forming a unique comprehensive feature vector of light source grid control for that light source grid unit, which is used as input for grid-level semantic parsing.

[0056] Based on the comprehensive feature vector of the light source grid control within the future time window, semantic analysis is performed to obtain the predicted spatiotemporal demand features of lighting. Semantic analysis refers to converting high-dimensional vectors that are difficult to use directly for control into interpretable and quantifiable descriptions of lighting demand. Specifically, a feature mapping model, such as a regression / classifier composed of multilayer perceptrons, is used to map the comprehensive feature vector of grid control into several demand semantic quantities, including predicted demand intensity (continuous values ​​corresponding to target illuminance or dimming level), predicted demand level (such as discrete categories of safety lighting level / regular lighting level / energy-saving lighting level), and demand change trend (rising / stable / falling, which can be obtained by differential statistics of adjacent future segments). Subsequently, the above semantic quantities are corrected by rules. For example, when the prediction shows that the demand is very low at night but the grid belongs to the main road or key area, the demand intensity is raised to a minimum value according to a preset safety lower limit threshold, and this minimum constraint is marked in the predicted spatiotemporal demand features of lighting.

[0057] Simultaneously, real-time lighting response requirements are acquired based on sensors within the lighting control area. Real-time lighting response requirements refer to the immediate lighting needs triggered by sensors and business events within the control area, used to correct lags or biases caused by pure prediction. Specifically, real-time observations such as illuminance, visibility, rainfall / snowfall, road occupancy, and pedestrian / vehicle density extracted from video / radar are periodically collected, and time synchronization and outlier processing are performed (such as median filtering to remove instantaneous spikes and thresholding to remove unreasonable readings). Then, spatial aggregation is performed by grid cells (for example, weighted averaging or taking quantiles of multi-sensor readings within the grid to enhance robustness). At the same time, event signals such as temporary activities, security alarms, and road closures are accessed and transcribed into structured fields such as event intensity and duration, ultimately forming the real-time response requirement vector of the grid cell at the current moment, providing an immediate correction term for subsequent fusion with predicted requirements.

[0058] The predicted spatiotemporal lighting demand features and real-time light source response demands are fused to obtain the light source grid demand features corresponding to the light source grid units. First, the two types of features are aligned to the same dimension (e.g., both are mapped to the demand intensity range of 0 to 1, and one-hot encoding is performed for categorical semantic quantities). Then, a dual-channel fusion is used to obtain the final light source grid demand features. One channel retains the predicted demand to provide future lead time, while the other channel retains the real-time demand to be sensitive to sudden changes. The fusion method can be gated weighting, that is, the weights are adaptively allocated according to the real-time reliability coefficient (e.g., the real-time weight is increased when the sensor integrity rate is high and the fluctuation is significant; the prediction weight is increased when the sensor is missing or the noise is high). After fusion, control constraint fields such as safety lower limit and maximum allowable dimming rate are added to form grid demand features that can be directly used for heat map assignment and subsequent control calculations.

[0059] A graph is constructed by combining all light source grid units and their corresponding light source grid demand characteristics to obtain a heat map of light source illumination demand. Graph construction refers to organizing grid units into an attributed topology: First, each light source grid unit is used as a node of the graph, and the light source grid demand characteristics of that unit are used as node attributes; then, edges are established based on geographical adjacency or road connectivity (e.g., grid units sharing a boundary are connected, or adjacent units on the road network topology are connected), and edge attributes such as distance, road grade difference, and functional area difference are assigned to the edges; finally, the core fields such as demand intensity / grade in the node attributes are projected onto the grid patches of the geographic information system base map, and the colors or grayscale are rendered according to the numerical values ​​to generate a visual heat map layer, while preserving the underlying graph structure for subsequent calculation of inter-grid collaboration and smooth control, thus forming a light source illumination demand heat map that is both visually expressive and computationally comprehensible.

[0060] The lamp source control smoothing index between lamp source grid units is determined in the lamp source illumination demand heat map; based on the lamp source illumination demand heat map and the lamp source grid units, all lamp sources to be adjusted within the lamp source control area are controlled to obtain the lamp source energy-saving control strategy.

[0061] The specific steps for determining the lamp source control smoothing index between lamp source grid cells in the lamp source illumination demand heatmap include:

[0062] The light source grid units are clustered according to functional zoning to obtain several basic grid unit clusters. The light source grid units are also clustered according to functional zoning to merge grids with similar usage scenarios into basic grid unit clusters. The functional zoning layer in the geographic information model and the semantic labels of each grid unit (e.g., teaching area, living area, main road, etc.) are read, and the functional labels, road level, historical pedestrian and vehicle activity statistics, nighttime safety level, etc. of the grid units are combined into functional attribute vectors. Then, the functional similarity between grid units is calculated (e.g., using cosine similarity or normalized Euclidean distance), and grids with high similarity and spatial adjacency are merged into the same basic grid unit cluster.

[0063] For a basic grid unit cluster, the comprehensive feature vectors of light source control corresponding to all light source grid units within it are extracted and their attributes are fused to obtain the comprehensive feature vector of the grid cluster. For each basic grid unit cluster, the comprehensive feature vectors of light source control corresponding to all light source grid units within it (from the grid-level comprehensive representation output by the spatiotemporal model) are extracted and their attributes are fused to obtain the comprehensive feature vector of the grid cluster. The fusion process can adopt weighted averaging or robust statistics (for example, first removing extreme values ​​from each dimension, and then calculating the average value by weighting the number of nodes or road weights), so that the comprehensive vector of the cluster can simultaneously represent the consistency of functional attributes and the consistency of control semantics.

[0064] The basic grid cell clusters are divided using spectral clustering and hierarchical clustering algorithms. When the functional similarity between multiple adjacent basic grid cell clusters exceeds a preset first threshold, several large grid lighting regions are obtained. When the functional similarity between a basic grid cell cluster and its adjacent basic grid cell clusters is less than a preset second threshold, several small grid lighting regions are obtained. Otherwise, the original basic grid cell clusters are retained. At the same time, the small grid lighting regions are local light source control units within the large grid lighting regions.

[0065] After obtaining the basic grid cell clusters, spectral clustering and hierarchical clustering are used to further divide them to form a hierarchical control structure of large grid lighting regions / small grid lighting regions. The processing steps of spectral clustering are as follows: first, construct a cluster-cluster similarity matrix (calculate the similarity based on the comprehensive feature vector of the grid clusters and assign higher connection weights to spatial adjacency relationships), then calculate the Laplacian matrix corresponding to the matrix and obtain the first few feature vectors. After mapping the basic clusters to a low-dimensional embedding space, clustering is performed to obtain candidate regions. The processing steps of hierarchical clustering are as follows: take adjacent basic clusters as the initial clusters, and merge or divide them layer by layer according to functional similarity from high to low to generate candidate regions at different scales.

[0066] Specifically, the comprehensive feature vector of each basic grid unit cluster is used as the main representation, supplemented by auxiliary attributes such as the functional zoning label, road grade ratio, and nighttime pedestrian and vehicle activity statistics of the cluster. First, the features of each dimension are unified to a comparable scale (e.g., continuous features are standardized, and categorical features are numerically encoded and normalized). Then, the functional similarity between any two basic clusters is calculated. To reflect the control principle of prioritizing adjacency, the system only calculates the effective similarity for basic clusters that are spatially adjacent or connected by road networks. For non-adjacent clusters, the similarity is set to extremely low or directly to zero, thereby limiting the secondary partitioning problem to a locally collaborative range and avoiding non-physical merging across regions.

[0067] In spectral clustering, the similarity between basic clusters is transformed into structural information that can be used for graph segmentation, thereby obtaining candidate large-scale lighting regions. First, a cluster-cluster similarity matrix is ​​constructed, where each element consists of two parts: one part is the functional similarity calculated from the comprehensive feature vectors of the grid clusters, and the other part is the connection weight of spatial adjacency (e.g., multiply by a higher weight if adjacent, and set to zero if not adjacent), thus forming strong connections that are similar and adjacent. Then, the Laplacian matrix is ​​calculated from the similarity matrix, and its first few feature vectors are obtained to map each basic cluster to a low-dimensional embedding space. In this space, clusters that are highly similar and closely connected will naturally cluster together. Next, clustering is performed in the low-dimensional space to obtain a set of candidate regions, and the spatial connectivity of the candidate regions is checked (if there are non-connected sub-blocks, they are split according to connected components). Finally, a set of large-scale candidate lighting regions that meet the requirements of high similarity within regions and large differences between regions is obtained, providing a basis for subsequent threshold decision.

[0068] In hierarchical clustering, different scales of region partitioning results are generated in a fine-to-coarse manner to identify small local regions that need to be stripped or large continuous regions that need to be merged. Each basic grid cell cluster is used as the initial cluster, and the merging cost is calculated first between spatially adjacent clusters. The merging cost can be given by "1 - functional similarity", so that the higher the similarity, the easier it is to merge. Then, the clusters are merged layer by layer according to the merging cost from small to large, forming a hierarchical tree structure, where each layer corresponds to a region scale. In order to support both merging and segmentation, local anomalies are checked in reverse on the hierarchical tree: if the functional similarity between a basic cluster and its surrounding clusters is generally low, the cluster will become an edge node at a coarser scale. Based on this, it is marked as a potential segmentable unit and retained as an independent candidate at a finer scale level so as to form small grid lighting regions later.

[0069] A dual-threshold decision step is introduced. When the functional similarity between multiple adjacent basic clusters exceeds a preset first threshold, they are merged into a large grid lighting area. When the functional similarity between a basic cluster and its adjacent basic clusters is less than a preset second threshold, it is separated from the periphery and forms a small grid lighting area. If it is between the two thresholds, the original basic cluster is retained unchanged. At the same time, it is stipulated that the small grid lighting area belongs to the local control unit within its large grid lighting area, which is used to characterize fine dimming for sudden changes in local demand or special scenes.

[0070] The candidate region results output by spectral clustering and hierarchical clustering are uniformly judged to strictly implement three result forms: large grid lighting region, small grid lighting region, and retention of basic clusters. First, the average similarity within the region and the boundary similarity with the neighborhood are calculated for each candidate region. The average similarity within the region is used to measure whether it can be considered a large grid lighting region (i.e., multiple adjacent basic clusters are sufficiently similar overall), and the boundary similarity is used to measure whether there are small grid lighting regions that need to be separated (i.e., a basic cluster has significantly low similarity with its surroundings). When the average similarity within a group of adjacent basic clusters exceeds a preset first threshold, it is identified as a large grid lighting region, and all basic clusters in the region are assigned a unified upper-level control identifier. When the functional similarity between a basic cluster and its adjacent basic clusters is lower than a preset second threshold, it is separated from the current candidate large region or neighborhood structure and identified as a small grid lighting region. When the similarity is between the two thresholds, it means that neither the strong merging condition nor the strong separation condition is met. In this case, the basic cluster is retained as an independent control unit to avoid excessive merging causing a one-size-fits-all lighting strategy or excessive segmentation causing control fragmentation.

[0071] Specifically, a small grid lighting area must be completely within the boundary of a large grid lighting area. If it crosses multiple large grid boundaries, it will be assigned according to the rule of which large grid boundary it has the lowest similarity to (highest difference) or which large grid interior it has the highest similarity to (highest fit). When generating the control strategy, the large grid lighting area provides the overall baseline brightness and smoothness constraints, while the small grid lighting area only makes local corrections to the baseline within its coverage area (e.g., increasing it to cope with sudden high demand, or decreasing it to perform more aggressive energy saving). And the local corrections must meet the safety lower limit and change rate limit given by the large grid.

[0072] Identify the lighting coordination control boundaries between adjacent large grid lighting areas, and between a large grid lighting area and its internal small grid lighting areas, to obtain the light source control smoothness index;

[0073] The specific steps for identifying the lighting coordination control boundaries between adjacent large grid lighting areas, and between a large grid lighting area and its internal small grid lighting areas, include:

[0074] Identify the light source grid cells located at the illumination co-control boundary to obtain the boundary grid set. When identifying the light source grid cells located at the illumination co-control boundary, first convert the light source illumination demand heat map into a region labeling raster. That is, write the illumination region identifier of the large grid to which each light source grid cell belongs and (if it exists) the illumination region identifier of the small grid to which it belongs. Then, scan the adjacency relationship of each grid cell one by one to see if the region identifier of each grid cell and its adjacent grids (four-neighbor or eight-neighbor) changes. If a grid cell is different from the large grid region identifier of at least one adjacent grid cell, or if it is located inside a large grid cell but is different from the small grid identifier of an adjacent grid cell, it is determined that the grid cell is on the illumination co-control boundary. The grids that meet the determination conditions are archived according to the boundary type (large grid-large grid boundary, large grid-small grid boundary), and the boundary grids are merged for connectivity and denoised (e.g., removing isolated single-point boundaries and filling 1-grid-wide breaks), finally forming the boundary grid set for subsequent smoothing calculations.

[0075] The characteristic distance of the grid cluster comprehensive feature vector between adjacent large grid lighting areas is calculated to obtain the grid control feature distance. When calculating the grid control feature distance between adjacent large grid lighting areas, the grid cluster comprehensive feature vector corresponding to each large grid lighting area is used as the region-level representation, and the comprehensive feature vectors of the two large grid lighting areas on both sides of a boundary are taken for distance calculation. In order to make the distance reflect the semantic difference of lighting control rather than the difference in dimensions, the dimensions of the comprehensive feature vector are first uniformly scaled (e.g., normalized according to historical statistical range or standardized according to mean and variance), and then the grid control feature distance is obtained by using weighted Euclidean distance or cosine distance. The weights can be set according to the principle that "safety-related dimensions are more important, energy-saving dimensions are second, and comfort-related dimensions are third". When the boundary is long and the heterogeneity within the two sides of the region is strong, the system can also take the local mean of several grid cluster comprehensive feature vectors close to the boundary on both sides of the boundary as the boundary-side representative vector that is closer to the actual state of the boundary, and then calculate the distance based on this, so as to avoid using the average representation of the entire region to cover up the real differences near the boundary.

[0076] If the distance of the grid control features is greater than the preset gradient threshold, the corresponding lighting coordination control boundary is determined to be a strong coordination boundary; otherwise, it is a weak coordination boundary. When determining strong and weak coordination boundaries, the above-mentioned distance of the grid control features is compared with the preset gradient threshold. The gradient threshold is used to characterize the maximum semantic gradient that allows cross-regional smoothing. When the distance of the grid control features is greater than the gradient threshold, it indicates that the lighting demand or control semantics on both sides of the boundary are significantly different. If large-scale diffusion and averaging are still forcibly performed at the boundary, it is easy to cause the high-demand side to be pulled down or the low-demand side to be pulled up, which may lead to safety risks or energy-saving failures. Therefore, this boundary is marked as a strong coordination boundary and a transition zone is required at the boundary to achieve controlled smoothing. When the distance of the grid control features is not greater than the preset gradient threshold, it indicates that the control semantics on both sides are similar, and the benefits of cross-boundary smoothing coordination outweigh the risks. Therefore, this boundary is marked as a weak coordination boundary and a simpler regional smoothing coefficient is allowed to meet the continuity requirements.

[0077] For the boundary mesh set on the strong cooperative boundary, the boundary mesh set is assigned a higher boundary transition weight than the boundary inside the large mesh lighting area based on the mesh control feature distance, and the large mesh light source control smoothing index is calculated; for the boundary mesh set on the weak cooperative boundary, the large mesh basic illumination smoothing coefficient is set.

[0078] When calculating the smoothing index for large-grid light source control for strong cooperative boundaries, the boundary grid set corresponding to the strong cooperative boundary is first taken as the object. Each boundary grid is assigned a boundary transition weight, which must be higher than the default weight of the grid inside the large-grid lighting area to emphasize the continuity and controllability of brightness changes at the boundary. The weight assignment is related to the distance of the grid control feature, which can be achieved by monotonic mapping: the greater the distance, the higher the boundary transition weight (equivalent to requiring a wider and smoother transition band). When the distance is close to the threshold, the weight can be slightly higher than that inside. Then, within each large-grid lighting area, the smoothing index for light source control of the large grid is calculated by combining its internal default smoothing weight and the boundary transition weight of the boundary grid (which can be understood as the overall smoothing intensity parameter of the area when controlling diffusion / cooperative smoothing). This index is written into the area control parameter table and used to constrain the change amplitude near the boundary when the heat map demand signal is diffused later. For weak cooperative boundaries, since the difference between the two sides is small and there is no need to construct a separate transition band, the basic illumination smoothing coefficient of the large grid is directly set as a smoothing constraint for the corresponding area to reduce the computational complexity and maintain control stability.

[0079] For small grid lighting areas within a large grid lighting area, the deviation of the comprehensive feature vector of the grid cluster within the small grid lighting area from the lighting feature of its parent large grid lighting area is calculated. A light source control smoothing index for the small grid is generated based on this deviation. Specifically, when the deviation of the lighting feature of the small grid lighting area is a positive high-demand deviation, the light source control smoothing index of the parent large grid is superimposed to obtain the small grid light source control smoothing index for the small grid lighting area. When the deviation of the lighting feature of the small grid lighting area is not a positive high-demand deviation, a basic illumination smoothing coefficient for the small grid is set.

[0080] When generating a smoothing index for small grid lighting areas within a large grid lighting area, the lighting feature deviation is first calculated. This means the direction and magnitude of the difference between the comprehensive feature vector of the small grid cluster and the comprehensive feature vector of the large grid lighting area to which it belongs. After processing both at the same scale, the difference vector is calculated, and the projection of the difference vector on the demand intensity-related dimension is used as the basis for determining the deviation direction, while the weighted distance on all dimensions is used as the basis for quantifying the deviation magnitude. When the deviation direction shows a positive high demand deviation (i.e., the small grid is significantly higher than the large grid baseline in key demand dimensions, such as higher pedestrian and vehicle density, more active key entrances and exits, or higher security level), in order to ensure continuous lighting for local high demand and avoid sudden brightening or dimming in some areas, the smoothing index is adjusted within the small grid itself. The index is superimposed with the light source control smoothing index of the large grid, so that the small grid can be locally adjusted while maintaining a smooth connection with the large grid. When the deviation is not a positive high demand deviation (such as low demand deviation or insignificant difference), in order to avoid excessively transmitting the collaborative constraints of the large grid to the already energy-saving local area, thereby limiting the energy-saving space, the small grid's basic illumination smoothing coefficient will be directly set as a smoothing constraint. Finally, all the light source control smoothing indices of the large grid and the basic illumination smoothing coefficient of the large grid, as well as all the light source control smoothing indices of the small grid and the basic illumination smoothing coefficient of the small grid, are uniformly summarized and combined according to the key value method of area identifier - smoothing parameter to generate a light source control smoothing index set, which is used for subsequent diffusion smoothing and power distribution to the light sources to be adjusted in the whole area.

[0081] The light source control smoothing index is obtained by combining all the large grid light source control smoothing indices, large grid basic illumination smoothing coefficients, small grid light source control smoothing indices, and small grid basic illumination smoothing coefficients.

[0082] The specific steps for controlling all adjustable light sources within the light source control area based on the heat map of light source illumination demand and the light source grid unit include:

[0083] The process involves analyzing a heatmap of light source illumination demand to identify the characteristic values ​​of each light source grid unit and constructing a light source demand amplitude array. This array serves as the initial light control signal source. The characteristic values ​​of each light source grid unit are retrieved from the heatmap, containing information such as the future illumination demand intensity, demand level, and demand trend for that grid unit. Based on these characteristic values, a light source demand amplitude array is generated. This array is a set of vectors representing the illumination demand of each light source grid unit, with each element representing the illumination demand intensity of that unit. This array acts as the initial light control signal source to initiate subsequent control calculations. The light source demand amplitude array provides the basic input for subsequent control signals, ensuring that the control process aligns with actual lighting demand.

[0084] The light signal penetration coefficient is calculated based on the light source control smoothing index. Specifically, when the light source control smoothing index indicates a large grid light source control smoothing index, the light signal penetration coefficient is a high light penetration coefficient; when the light source control smoothing index indicates a small grid light source control smoothing index, the light signal penetration coefficient is a low light penetration coefficient; otherwise, the light signal penetration coefficient is the basic light penetration coefficient.

[0085] The light source control smoothness index is derived from the control requirements of the light source grid cells, with different smoothness requirements in different areas. When the light source control smoothness index indicates a large grid light source control smoothness index, it means that the light control demand in this area is relatively extensive and uniform. Therefore, a high light penetration coefficient is assigned to the light signal penetration coefficient, meaning that the signal can diffuse relatively freely in this area. Conversely, when the light source control smoothness index indicates a small grid light source control smoothness index, it means that the control demand in this area is relatively local and specific. A lower light signal penetration coefficient is assigned to avoid excessive diffusion leading to local control failure. If the light signal penetration coefficient is not classified as a large grid or small grid control smoothness index, a basic penetration coefficient is assigned to this area, indicating that the signal diffusion in this area is neither too strong nor too weak, maintaining a balance.

[0086] A light source control diffusion network is constructed based on the light signal penetration coefficient. The initial light control signal source is used as input, and the signal intensity is iteratively calculated in the light source control diffusion network to generate the final light response envelope within the light source control area.

[0087] After calculating the light signal penetration coefficient, it is applied to the construction of a light source control diffusion network. This network is a graph structure where each light source grid cell is a node, connections between nodes represent adjacency relationships, and the weight of each connection is determined by the light signal penetration coefficient. The purpose of the light source control diffusion network is to simulate the propagation of light signals within the light source control area through a diffusion mechanism. Using an initial light control signal source as input, the initial value of the signal source is a control signal generated based on the light demand amplitude array. These signals diffuse along the connections in the network, adjusting their intensity according to the penetration coefficient of each adjacent node. The process is iteratively calculated throughout the network until the signal intensity converges. This process iterates multiple times, resulting in a final light response value for each node (light source grid cell), which represents the final control signal for that grid cell.

[0088] The values ​​in the light response envelope are mapped to the adjustment range set by the light source to be adjusted, thus obtaining the energy-saving control strategy for each light source to be adjusted.

[0089] The values ​​in the obtained illumination response envelope are mapped to the adjustment range of the light source to be adjusted. Each light source has a fixed adjustment range, which is set according to parameters such as the rated power, minimum dimming limit, and dimming accuracy of the light source. By mapping the calculation results in the illumination response envelope to these adjustment ranges, a specific energy-saving control strategy is generated for each light source. The energy-saving control strategy assigns a precise adjustment value to each light source to adjust its brightness, power, and other control parameters to achieve a balance between lighting needs and energy-saving goals. Through the above method, the light source control system can automatically adjust the brightness of each light source according to the needs of the area and the control strategy, thereby ensuring that the entire area meets lighting needs while maximizing energy savings.

[0090] In this embodiment, common LED streetlights may have rated power of 50W, 100W, 150W, etc., while the adjustable power range is generally between 20% and 100%. That is, a 50W LED streetlight has an adjustable power range of 10W-50W. During energy-saving control, the power of the lamps should not be adjusted beyond their adjustable range, otherwise the lamps may malfunction, affecting the lighting effect or even damaging the lamps. Specifically, in this embodiment, it is assumed that on a suburban road late at night, after calculation, the lighting power needs to be reduced to 30% of the normal power. It is assumed that the rated power of the streetlights on this road is 100W. If the adjustable power range of a street light is 20%-100%, then the required control wattage is 100W × 30%. Since 30W is within the adjustable power range of this street light, the power of the street light can be adjusted to 30W. However, if the calculated control wattage is lower than the lower limit of the adjustable power of the lamp, such as a calculated control wattage of 5W, while the lower limit of the adjustable power of the street light is 20W, then the power of the lamp will be adjusted to the lower limit of the adjustable power of 20W to ensure the normal operation of the lamp. In this way, under the premise of meeting energy-saving requirements, the wattage range of the lamp is fully considered to achieve reasonable power control for each lamp.

[0091] In this embodiment, for example, energy-saving control of lighting sources is implemented in a university campus scenario. A geographic base map and functional zoning layers are imported, covering areas such as main roads, teaching areas, living areas, sports areas, and green belts. Each area's lighting grid is divided according to road direction and lighting density. Each grid contains several lighting control nodes (such as LED streetlights), and the adjacency relationships between grids are associated. Specifically, main roads and surrounding areas such as teaching and living areas form continuous or semi-continuous grid strips, with each grid containing multiple control nodes. A topological structure is maintained between adjacent grids for subsequent collaborative control. Real-time monitoring is performed using spatiotemporal sensing data (such as traffic flow and environmental data) and sensor data (such as illuminance sensor data) to generate illumination demand characteristics for each grid. Based on these characteristics, a lighting demand heatmap is generated, reflecting the intensity of lighting demand in different areas. The demand characteristic values ​​of each grid are extracted and aggregated into an illumination demand amplitude array, representing the intensity of lighting demand in each grid. This array becomes the initial input to the control signal for subsequent power adjustment calculations.

[0092] The system determines the light signal penetration coefficient based on the light source control smoothing index. Specifically, when the light source control smoothing index indicates a large-grid light source control smoothing index, the light signal penetration coefficient is a high light penetration coefficient; when it indicates a small-grid light source control smoothing index, the light signal penetration coefficient is a low light penetration coefficient; otherwise, it is set to the basic light penetration coefficient. Using the signal penetration coefficient, the system constructs a light source control diffusion network to simulate the propagation of the light control signal between grid cells.

[0093] The construction of the light source control diffusion network involves calculating the connection relationship between each grid cell and its adjacent cells, and weighting it by combining the light signal penetration coefficient. Taking the initial light control signal as input, the diffusion network performs multiple iterative calculations to finally generate a light response envelope. This response envelope reflects the final lighting control signal of each grid cell, combining the needs and control signals of the surrounding grid cells.

[0094] After obtaining the final illumination response envelope, the system maps the values ​​in the response envelope to the power adjustment range of the light source to be adjusted. For Class A (rated power 100W, minimum controllable power 20W) and Class B (rated power 60W, minimum controllable power 12W) luminaires, adjustments are made based on the calculated theoretical power. For example, assuming that a Class A luminaire on a main road needs to be adjusted to 30W, it is legal within its adjustable power range (10W to 50W), so it can be directly adjusted to 30W. If the calculated control wattage is lower than the minimum adjustable power (e.g., 5W), it is adjusted to 20W (the minimum controllable power of a Class A luminaire). In this way, the power adjustment of each luminaire is ensured to comply with its equipment limitations, avoiding malfunctions or damage to the luminaires.

[0095] During winter (December), taking 200 lights in the park as an example, the system is divided into three control periods: evening peak, off-peak, and late night. At late night, Class A lights in main roads and teaching areas maintain a higher power (approximately 78W), while lights in smaller grid areas such as green belts may be reduced to 20W. However, this is still constrained by the smooth transition within the larger grid lighting area to ensure uniform lighting. Ultimately, the system ensures that the total energy consumption of the entire park is reduced by approximately 15% relative to the full power baseline, from 235.2kWh to 200.6kWh, achieving an energy saving target close to 15%.

[0096] The above embodiments are only a reference for an executable solution and do not represent data under all operating conditions.

[0097] Example 2: A lamp source energy-saving control system based on a spatiotemporal model, see [link / reference]. Figure 1 As shown, it includes:

[0098] The spatiotemporal model construction module includes a model construction unit and a grid division unit. The model construction unit is used when there are several light sources to be adjusted within the light source control area. It acquires the spatiotemporal perception data corresponding to each light source to be adjusted within the light source control area. It maps each light source to be adjusted within the light source control area to a point in a preset GIS model to obtain light source control nodes. It establishes a spatiotemporal model for light source control based on the spatiotemporal perception data. The spatiotemporal perception data package contains traffic flow spatiotemporal data, environmental spatiotemporal data, and illumination spatiotemporal data. The grid division unit is used to divide the light source into grids using the light source control spatiotemporal model for each light source to be adjusted within the light source control area, and establishes light source grid units.

[0099] The lamp source energy-saving control module includes a feature recognition unit and an energy-saving control unit. The feature recognition unit is used to model the illumination demand of each lamp source grid unit to obtain a lamp source illumination demand heat map. The lamp source illumination demand heat map contains lamp source grid units and corresponding lamp source grid demand features. The energy-saving control unit is used to determine the lamp source control smoothing index between lamp source grid units in the lamp source illumination demand heat map. Based on the lamp source illumination demand heat map and the lamp source grid units, all lamp sources to be adjusted within the lamp source control area are controlled to obtain the lamp source energy-saving control strategy.

[0100] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims. Parts not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A method for energy-saving control of light sources based on a spatiotemporal model, characterized in that, Includes the following steps: There are several light sources to be adjusted within the light source control area; acquire the spatiotemporal sensing data corresponding to each light source to be adjusted within the light source control area; Each light source to be adjusted within the light source control area is mapped to a point in a preset GIS model to obtain a light source control node; A spatiotemporal model for light source control is established based on spatiotemporal perception data. The spatiotemporal perception data package contains traffic flow spatiotemporal data, environmental spatiotemporal data, and illumination spatiotemporal data. For each light source to be adjusted within the light source control area, the spatiotemporal model for light source control is used to divide the grid and establish light source grid units. Lighting demand modeling is performed for each light source grid cell to obtain a light source lighting demand heatmap; the light source lighting demand heatmap contains light source grid cells and corresponding light source grid demand features; The lamp source control smoothing index between lamp source grid units is determined in the lamp source illumination demand heat map; based on the lamp source illumination demand heat map and the lamp source grid units, all lamp sources to be adjusted within the lamp source control area are controlled to obtain the lamp source energy-saving control strategy.

2. The energy-saving control method for light sources based on a spatiotemporal model according to claim 1, characterized in that, The specific steps for establishing a spatiotemporal model for light source control based on spatiotemporal perception data include: The spatiotemporal sensing data is matched with the geographic coordinates and timestamps of the light sources to be adjusted within the light source control area to obtain the spatiotemporal feature vector of the light source control. The local spatial gradient of the light source and the average spatiotemporal features of the light source area are calculated based on the spatiotemporal feature vector of the light source control. Periodic analysis is performed on the spatiotemporal feature vector of the light source control to obtain the periodic pattern features of the light source. In the GNN-RNN framework, calculations are performed based on the local spatial gradient of the light source, the periodic pattern features of the light source, and the average spatiotemporal features of the light source region to obtain the comprehensive feature vector of light source control for each light source control node within the light source control region; a spatiotemporal model of light source control is constructed based on all comprehensive feature vectors of light source control. Among them, a prediction model framework is introduced into the spatiotemporal model of light source control. The prediction model framework is used to predict the comprehensive feature vector of light source control for each light source control node in the future time window.

3. The energy-saving control method for light sources based on a spatiotemporal model according to claim 2, characterized in that, The specific steps for modeling the illumination requirements of each light source grid cell include: Obtain the comprehensive feature vector of light source control for each light source control node output in the spatiotemporal model of light source control within a future time window; perform feature fusion on the comprehensive feature vectors of light source control for all light source control nodes in the light source grid cell within the future time window to obtain the comprehensive feature vector of light source grid control corresponding to the light source grid cell. Semantic parsing is performed on the comprehensive feature vector of the light source grid control within the future time window to obtain the predicted spatiotemporal demand features of lighting; at the same time, real-time light source response demand is obtained based on the sensors within the light source control area; the predicted spatiotemporal demand features of lighting and the real-time light source response demand are fused to obtain the light source grid demand features corresponding to the light source grid unit. A graph is constructed by combining all light source grid cells and their corresponding light source grid demand characteristics to obtain a heat map of light source illumination demand.

4. The energy-saving control method for light sources based on a spatiotemporal model according to claim 3, characterized in that, The specific steps for determining the lamp source control smoothing index between lamp source grid cells in the lamp source illumination demand heatmap include: The light source grid units are clustered according to functional zoning to obtain several basic grid unit clusters; for each basic grid unit cluster, the comprehensive feature vectors of light source control corresponding to all light source grid units within the cluster are extracted and their attributes are fused to obtain the comprehensive feature vector of the grid cluster. The basic grid cell clusters are divided using spectral clustering and hierarchical clustering algorithms. When the functional similarity between multiple adjacent basic grid cell clusters exceeds a preset first threshold, several large grid lighting regions are obtained. When the functional similarity between a basic grid cell cluster and its adjacent basic grid cell clusters is less than a preset second threshold, several small grid lighting regions are obtained. Otherwise, the original basic grid cell clusters are retained. At the same time, the small grid lighting regions are local light source control units within the large grid lighting regions. Identify the lighting coordination control boundaries between adjacent large grid lighting areas and between large grid lighting areas and their internal small grid lighting areas to obtain the light source control smoothness index.

5. The energy-saving control method for light sources based on a spatiotemporal model according to claim 4, characterized in that, The specific steps for identifying the lighting coordination control boundaries between adjacent large grid lighting areas, and between a large grid lighting area and its internal small grid lighting areas, include: Identify the light source grid cells located at the illumination cooperative control boundary to obtain the boundary grid set; calculate the feature distance of the comprehensive feature vector of the grid cluster between adjacent large grid illumination areas to obtain the grid control feature distance; if the grid control feature distance is greater than the preset gradient threshold, the corresponding illumination cooperative control boundary is determined to be a strong cooperative boundary, otherwise it is a weak cooperative boundary; For the boundary mesh set on the strong cooperative boundary, the boundary mesh set is assigned a higher boundary transition weight than the boundary inside the large mesh lighting area based on the mesh control feature distance, and the large mesh light source control smoothing index is calculated; for the boundary mesh set on the weak cooperative boundary, the large mesh basic illumination smoothing coefficient is set. For small grid lighting areas within a large grid lighting area, the deviation of the comprehensive feature vector of the grid cluster within the small grid lighting area from the lighting feature of its parent large grid lighting area is calculated. A light source control smoothing index for the small grid is generated based on this deviation. Specifically, when the deviation of the lighting feature of the small grid lighting area is a positive high-demand deviation, the light source control smoothing index of the parent large grid is superimposed to obtain the small grid light source control smoothing index for the small grid lighting area. When the deviation of the lighting feature of the small grid lighting area is not a positive high-demand deviation, a basic illumination smoothing coefficient for the small grid is set. The light source control smoothing index is obtained by combining all the large grid light source control smoothing indices, large grid basic illumination smoothing coefficients, small grid light source control smoothing indices, and small grid basic illumination smoothing coefficients.

6. The energy-saving control method for light sources based on a spatiotemporal model according to claim 5, characterized in that, The specific steps for controlling all adjustable light sources within the light source control area based on the heat map of light source illumination demand and the light source grid unit include: The heat map of light source illumination demand is analyzed to identify the characteristic values ​​of light source grid demand for each light source grid unit and construct an illumination demand amplitude array; the illumination demand amplitude array is used as the initial illumination control signal source. The light signal penetration coefficient is calculated based on the light source control smoothing index. Specifically, when the light source control smoothing index indicates a large grid light source control smoothing index, the light signal penetration coefficient is a high light penetration coefficient; when the light source control smoothing index indicates a small grid light source control smoothing index, the light signal penetration coefficient is a low light penetration coefficient; otherwise, the light signal penetration coefficient is the basic light penetration coefficient. A light source control diffusion network is constructed based on the light signal penetration coefficient. The initial light control signal source is used as input, and the signal intensity is iteratively calculated in the light source control diffusion network to generate the final light response envelope within the light source control area. The values ​​in the light response envelope are mapped to the adjustment range set by the light source to be adjusted, thus obtaining the energy-saving control strategy for each light source to be adjusted.

7. A lamp source energy-saving control system based on a spatiotemporal model, characterized in that, The system employs a spatiotemporal model-based energy-saving control method for light sources as described in any one of claims 1-6, comprising: The spatiotemporal model construction module includes a model construction unit and a grid division unit. The model construction unit is used when there are several light sources to be adjusted within the light source control area. It acquires the spatiotemporal perception data corresponding to each light source to be adjusted within the light source control area. It maps each light source to be adjusted within the light source control area to a point in a preset GIS model to obtain light source control nodes. It establishes a spatiotemporal model for light source control based on the spatiotemporal perception data. The spatiotemporal perception data package contains traffic flow spatiotemporal data, environmental spatiotemporal data, and illumination spatiotemporal data. The grid division unit is used to divide the light source into grids using the light source control spatiotemporal model for each light source to be adjusted within the light source control area, and establishes light source grid units. The lamp source energy-saving control module includes a feature recognition unit and an energy-saving control unit. The feature recognition unit is used to model the illumination demand of each lamp source grid unit to obtain a lamp source illumination demand heat map. The lamp source illumination demand heat map contains lamp source grid units and corresponding lamp source grid demand features. The energy-saving control unit is used to determine the lamp source control smoothing index between lamp source grid units in the lamp source illumination demand heat map. Based on the lamp source illumination demand heat map and the lamp source grid units, all lamp sources to be adjusted within the lamp source control area are controlled to obtain the lamp source energy-saving control strategy.

Images