Multi-level collaborative balancing method and system for smart urban lighting system

Abstract

The application provides a multi-level collaborative balancing method and system for a smart urban lighting system, and relates to the technical field of smart lighting. The method comprises the following steps: collecting light intensity and energy consumption parameters, constructing an operating state vector, and performing graph convolution operation to obtain a node evaluation value and a region coupling degree; constructing a multi-level topological structure through hierarchical clustering and dependency analysis, determining a conduction path and a delay coefficient; determining a corrected control amplitude based on a time sequence ordering control priority and combining energy consumption constraints to generate instructions for step-by-step control.

Description

Technical Field

[0001] This invention relates to the field of smart lighting technology, and in particular to a multi-level collaborative balancing method and system for smart urban lighting systems. Background Technology

[0002] In the construction of smart cities, intelligent control of lighting systems is a key link in improving energy efficiency and ensuring the quality of public lighting. In existing technologies, lighting control methods mainly rely on responsive adjustment of independent parameters of individual lighting terminals or local areas, or the use of time-sharing and zoning strategies. Based on a preset schedule or simple area division, groups of streetlights are managed to operate in a unified manner, usually only considering the real-time status of the terminal itself or the static attributes of the area to which it belongs.

[0003] However, existing technologies still lack effective modeling of the dynamic correlations within large-scale lighting networks, making it difficult to address the mutual influences between lighting terminals due to physical proximity or electrical connections, and neglecting the impact of system hierarchy and control command transmission timing. This results in control actions failing to form a coordinated and orderly step-by-step response, which may not only cause instantaneous grid load shocks, but also presents a contradiction between overall energy efficiency optimization and lighting quality maintenance when facing dynamic changes. Summary of the Invention

[0004] This invention provides a multi-level collaborative balancing method and system for smart city lighting systems, which can at least solve some of the problems existing in the prior art.

[0005] A first aspect of this invention provides a multi-level collaborative balancing method for a smart city lighting system, comprising:

[0006] The illuminance parameters and energy consumption parameters of the lighting terminals within the lighting area are collected. The illuminance parameters are statistically analyzed using a time-domain sliding window to obtain brightness fluctuation characteristics. An operating state vector is constructed based on the brightness fluctuation characteristics and the energy consumption parameters. A regional adjacency matrix is ​​constructed based on the topological connection relationship between the lighting terminals. Graph convolution operation is performed on the operating state vector to obtain node evaluation values ​​and regional coupling degree.

[0007] The load distribution pattern is obtained by hierarchically clustering the node evaluation values. The cross-regional influence factor is calculated based on the regional coupling degree and the hierarchical dependency relationship is obtained by combining the node set corresponding to the load distribution pattern. A multi-level topology structure is constructed based on the hierarchical dependency relationship and the inter-layer transmission path is extracted. The transmission delay coefficient is determined based on the inter-layer transmission path.

[0008] Based on the conduction delay coefficient, the inter-layer conduction path is sorted in time to obtain a control priority sequence. The control trigger time of each level node is determined according to the control priority sequence, and the control amplitude of each level node is determined in combination with the node evaluation value. The control amplitude is compared and verified based on the preset energy consumption constraints to obtain a corrected control amplitude. Based on the corrected control amplitude, a balanced control command is generated and the lighting terminals corresponding to each level node are controlled step by step.

[0009] In one alternative implementation,

[0010] The illuminance parameters within the lighting area and the energy consumption parameters of the lighting terminals are collected. Time-domain sliding window statistics are performed on the illuminance parameters to obtain brightness fluctuation characteristics. Based on the brightness fluctuation characteristics and the energy consumption parameters, an operating state vector is constructed, including:

[0011] Illumination intensity parameters at multiple spatial locations within the illumination area are acquired using distributed sensing devices. These parameters are then arranged according to their corresponding timestamps and a sliding window is initialized. Within each sliding window, the mean and variance of the illumination intensity parameters are calculated to obtain window statistical features. The window statistical features of adjacent windows are then differentially analyzed to obtain a rate of change sequence. Based on this rate of change sequence, the rising and falling phases of the illumination intensity parameters are identified and inflection points are marked. Finally, the window statistical features and the illumination intensity parameters corresponding to the inflection points are combined and encoded to obtain brightness fluctuation features.

[0012] The system obtains real-time current and voltage parameters of each lighting terminal from the distributed lighting terminal and calculates instantaneous power. It then performs time integration on the instantaneous power to obtain a cumulative energy consumption value. Finally, it normalizes the cumulative energy consumption value and a preset benchmark energy consumption value to obtain energy consumption parameters.

[0013] The brightness fluctuation features are arranged into a brightness feature matrix according to the spatial location of the lighting area, and the energy consumption parameters are arranged into an energy consumption feature vector according to the lighting terminal. A mapping matrix is ​​established based on the coverage relationship between the lighting terminal and the spatial location. The brightness feature matrix is ​​spatially aligned and transformed based on the mapping matrix to obtain an aligned brightness feature matrix. The aligned brightness feature matrix and the energy consumption feature vector are spliced ​​and dimension-reduced by projection to obtain the operating state vector.

[0014] In one alternative implementation,

[0015] Based on the topological connection relationship between lighting terminals, a region adjacency matrix is ​​constructed, and graph convolution operation is performed on the running state vector to obtain node evaluation values ​​and region coupling degrees, including:

[0016] The signal transmission delay and data interaction frequency between lighting terminals are obtained and a spatiotemporal coupling tensor is constructed. The spatiotemporal coupling tensor is decomposed and reconstructed by tensor decomposition and outer product to obtain a connection strength matrix. A region adjacency matrix is ​​constructed based on the connection strength matrix.

[0017] The running state vectors are arranged into a state feature matrix and combined with the region adjacency matrix to perform tensor shrinkage operation to obtain primary aggregation features. The region adjacency matrix is ​​then subjected to spectral decomposition to obtain a sequence of eigenvalues. A frequency domain filter is constructed based on the eigenvalue sequence and the primary aggregation features are filtered to obtain frequency domain enhancement features. An attention enhancement feature is calculated based on the frequency domain enhancement feature and the state feature matrix. The set of neighboring nodes for each lighting terminal is determined based on the region adjacency matrix, and the attention enhancement features are subjected to neighborhood aggregation to obtain propagation features. Node evaluation values ​​are then determined using graph pooling.

[0018] Based on the region adjacency matrix, a multi-hop neighborhood topological subgraph is extracted and isomorphic encoding is performed to obtain a topological pattern vector. A structural similarity matrix is ​​constructed based on the topological pattern vector and bilinearly pooled with the state feature matrix to obtain structural awareness features. Graph comparison learning is performed on the structural awareness features to determine information flow. Entropy measurement is performed on the information flow to calculate the region coupling degree.

[0019] In one alternative implementation,

[0020] Hierarchical clustering of the node evaluation values ​​yields the load distribution pattern. Based on the regional coupling degree, cross-regional influence factors are calculated, and the hierarchical dependencies are obtained by combining the node set corresponding to the load distribution pattern.

[0021] The node evaluation values ​​are divided into multiple load levels. The variance of the node evaluation values ​​of the lighting terminals in each load level is calculated, and the load levels are adjusted to obtain optimized load levels. The set of lighting terminals corresponding to each optimized load level is extracted and marked as a load distribution pattern. Based on the region adjacency matrix, the number of connection edges in each load distribution pattern is counted and the internal connection density is calculated. The number of cross-pattern connection edges between different load distribution patterns is counted. Based on the number of cross-pattern connection edges and the internal connection density, a pattern coupling strength matrix is ​​constructed.

[0022] Based on the load distribution pattern, the regional coupling degree is grouped and the mean regional coupling degree of each load distribution pattern is calculated to obtain the mean mode coupling. The product of the mean mode coupling and the corresponding element in the mode coupling strength matrix is ​​calculated to obtain the initial cross-regional influence factor. The initial cross-regional influence factor is attenuated and corrected according to the topological distance, and the corrected cross-regional influence factor is obtained by combining the pre-acquired frequency domain enhancement features.

[0023] The lighting terminal with the highest node evaluation value is selected from the node set corresponding to the load distribution mode as the mode representative node. The topology hop count between different mode representative nodes is calculated and the inter-mode dependency strength is determined by combining the modified cross-regional influence factor. A directed dependency graph is constructed based on the inter-mode dependency strength and the topology is sorted to obtain the hierarchical dependency relationship.

[0024] In one alternative implementation,

[0025] Based on the hierarchical dependencies, a multi-level topology is constructed and inter-layer propagation paths are extracted. The propagation delay coefficient is then determined based on these inter-layer propagation paths, including:

[0026] Based on the hierarchical dependency relationship, the hierarchical sorting result is determined and the lighting terminals are assigned to the corresponding hierarchical node sets. Based on the dependency edges of the pre-constructed directed dependency graph, inter-layer connection edges are established between the hierarchical node sets to obtain an inter-layer connection edge set. The number of lighting terminals in the hierarchical node sets and the number of inter-layer connection edges between different levels are counted to construct an inter-layer connection density matrix. A multi-level topology structure is constructed based on the inter-layer connection edge set.

[0027] Lighting terminals are selected from the highest-level node set of the multi-level topology and the inter-layer connection edge set is traversed downwards to obtain the inter-layer conduction path set. The node set pairs passed through by each inter-layer conduction path are extracted and the corresponding connection density is queried based on the inter-layer connection density matrix. The effective inter-layer conduction path set is determined based on the connection density. The path connection weight corresponding to each effective inter-layer conduction path is extracted from the region adjacency matrix. The path conduction efficiency is calculated based on the length of the effective inter-layer conduction path and the path connection weight.

[0028] Extract the node evaluation values ​​of the starting node and the ending node on the effective interlayer conduction path and calculate the difference between the node evaluation values. Calculate the conduction impedance based on the difference between the node evaluation values ​​and the path conduction efficiency. Extract the proportion of low-frequency components of the starting node based on the predetermined frequency domain enhancement features and perform frequency domain correction on the conduction impedance to obtain the conduction delay coefficient.

[0029] In one alternative implementation,

[0030] Based on the conduction delay coefficient, the inter-layer conduction paths are time-series ordered to obtain a control priority sequence. The control trigger time of each level node is determined according to the control priority sequence, and the control amplitude of each level node is determined in conjunction with the node evaluation value, including:

[0031] The conduction delay coefficients corresponding to each effective inter-layer conduction path in the effective inter-layer conduction path set are sorted in ascending order. The lighting terminal sequence contained in each effective inter-layer conduction path is extracted in the sorting order and combined sequentially to obtain the control priority sequence. The position index of each lighting terminal is extracted from the control priority sequence and the initial trigger time is calculated based on the position index. The connection density corresponding to the set of nodes of each layer to which each lighting terminal belongs is extracted from the inter-layer connection density matrix and the initial trigger time is adjusted based on the connection density to obtain the control trigger time.

[0032] Extract the node evaluation value corresponding to each lighting terminal in the control priority sequence and calculate the mean of the evaluation values ​​of the control priority sequence. Calculate the deviation between the node evaluation value corresponding to each lighting terminal and the mean of the evaluation values ​​to obtain the evaluation deviation. Extract the path conduction efficiency of the effective inter-layer conduction path where each lighting terminal is located and perform conduction efficiency weighting on the evaluation deviation based on the path conduction efficiency to obtain the weighted evaluation deviation. Determine the control amplitude of each level node based on the weighted evaluation deviation.

[0033] In one alternative implementation,

[0034] The modified control amplitude is obtained by comparing and verifying the control amplitude based on preset energy consumption constraints. Based on the modified control amplitude, a balanced control command is generated, and the lighting terminals corresponding to each level node are controlled step-by-step, including:

[0035] Extract the global energy consumption upper limit and the hierarchical energy consumption upper limit set from the preset energy consumption constraints. Calculate the estimated energy consumption of each lighting terminal based on the control amplitude and accumulate them hierarchically to obtain the hierarchical energy consumption distribution. Compare the hierarchical energy consumption distribution with the hierarchical energy consumption upper limit set to identify the energy consumption exceeding the limit level and extract the corresponding path conduction efficiency to calculate the reduction coefficient. Based on the reduction coefficient, reduce the control amplitude of the lighting terminals within the energy consumption exceeding the limit level to obtain the hierarchical corrected control amplitude. Calculate the global cumulative energy consumption based on the hierarchical corrected control amplitude and compensate it in conjunction with the global energy consumption upper limit to obtain the corrected control amplitude.

[0036] Based on the correction control amplitude and the control trigger time, the control parameters of each lighting terminal are determined and a hierarchical control parameter set is constructed according to the hierarchy. The path conduction efficiency is extracted from the pre-acquired inter-layer connection edge set and the inter-layer conduction coefficient is calculated. Based on the inter-layer conduction coefficient, the hierarchical control parameter set is subjected to conduction compensation and a balanced control command is generated.

[0037] Based on the equalization control command and the conduction delay coefficient, the lighting terminals corresponding to each level node are controlled step by step along the multi-level topology.

[0038] A second aspect of the present invention provides a multi-level collaborative equalization system for smart city lighting, comprising:

[0039] The data acquisition unit is used to collect illuminance parameters and energy consumption parameters of lighting terminals within the lighting area, perform time-domain sliding window statistics on the illuminance parameters to obtain brightness fluctuation characteristics, construct an operating state vector based on the brightness fluctuation characteristics and the energy consumption parameters, construct a region adjacency matrix based on the topological connection relationship between lighting terminals, and perform graph convolution operation on the operating state vector to obtain node evaluation values ​​and region coupling degree.

[0040] The topology analysis unit is used to perform hierarchical clustering on the node evaluation values ​​to obtain the load distribution pattern, calculate the cross-regional influence factor based on the regional coupling degree and solve the hierarchical dependency relationship by combining the node set corresponding to the load distribution pattern, construct a multi-level topology structure based on the hierarchical dependency relationship and extract the inter-layer transmission path, and determine the transmission delay coefficient based on the inter-layer transmission path.

[0041] The timing control unit is used to sort the inter-layer conduction paths according to the conduction delay coefficient to obtain a control priority sequence, determine the control trigger time of each level node according to the control priority sequence and determine the control amplitude of each level node in combination with the node evaluation value, compare and verify the control amplitude based on the preset energy consumption constraints to obtain a corrected control amplitude, generate a balanced control command based on the corrected control amplitude and perform step-by-step control on the lighting terminals corresponding to each level node.

[0042] A third aspect of the present invention provides an electronic device, comprising:

[0043] A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0044] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0045] In this invention, by performing time-domain sliding window statistics on illuminance parameters, brightness fluctuation characteristics are accurately extracted. Combined with energy consumption parameters of lighting terminals, an operating state vector is constructed, comprehensively depicting the real-time working state of lighting nodes. This effectively integrates topological connections, accurately calculates node evaluation values ​​reflecting the node's own state, and characterizes the regional coupling degree of mutual influence between regions, providing a reliable data foundation for subsequent collaborative analysis. Based on node evaluation values, hierarchical clustering automatically identifies the distribution patterns of different load levels within the system, forming logical node groups. According to the transmission delay coefficient determined by the transmission path, the time loss of control commands transmitted between different levels is quantified. Based on the transmission delay coefficient, the inter-layer transmission paths are sequentially ordered to generate a control priority sequence, ensuring that control commands are issued in a reasonable order, avoiding system oscillations caused by command conflicts or timing errors. By comparing and verifying the control amplitude through preset energy consumption constraints, a corrected control amplitude is obtained, ensuring that all control operations are performed within the overall system energy consumption limit framework, achieving the energy-saving goal. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating the multi-level collaborative balancing method for a smart city lighting system according to an embodiment of the present invention.

[0047] Figure 2 This is a flowchart illustrating the inter-level conduction control timing optimization of the multi-level collaborative balancing method for a smart city lighting system according to an embodiment of the present invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0050] Figure 1 This is a flowchart illustrating the multi-level collaborative balancing method for a smart city lighting system according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0051] The illuminance parameters and energy consumption parameters of the lighting terminals within the lighting area are collected. The illuminance parameters are statistically analyzed using a time-domain sliding window to obtain brightness fluctuation characteristics. An operating state vector is constructed based on the brightness fluctuation characteristics and the energy consumption parameters. A regional adjacency matrix is ​​constructed based on the topological connection relationship between the lighting terminals. Graph convolution operation is performed on the operating state vector to obtain node evaluation values ​​and regional coupling degree.

[0052] The load distribution pattern is obtained by hierarchically clustering the node evaluation values. The cross-regional influence factor is calculated based on the regional coupling degree and the hierarchical dependency relationship is obtained by combining the node set corresponding to the load distribution pattern. A multi-level topology structure is constructed based on the hierarchical dependency relationship and the inter-layer transmission path is extracted. The transmission delay coefficient is determined based on the inter-layer transmission path.

[0053] Based on the conduction delay coefficient, the inter-layer conduction path is sorted in time to obtain a control priority sequence. The control trigger time of each level node is determined according to the control priority sequence, and the control amplitude of each level node is determined in combination with the node evaluation value. The control amplitude is compared and verified based on the preset energy consumption constraints to obtain a corrected control amplitude. Based on the corrected control amplitude, a balanced control command is generated and the lighting terminals corresponding to each level node are controlled step by step.

[0054] In one alternative implementation,

[0055] The illuminance parameters within the lighting area and the energy consumption parameters of the lighting terminals are collected. Time-domain sliding window statistics are performed on the illuminance parameters to obtain brightness fluctuation characteristics. Based on the brightness fluctuation characteristics and the energy consumption parameters, an operating state vector is constructed, including:

[0056] Illumination intensity parameters at multiple spatial locations within the illumination area are acquired using distributed sensing devices. These parameters are then arranged according to their corresponding timestamps and a sliding window is initialized. Within each sliding window, the mean and variance of the illumination intensity parameters are calculated to obtain window statistical features. The window statistical features of adjacent windows are then differentially analyzed to obtain a rate of change sequence. Based on this rate of change sequence, the rising and falling phases of the illumination intensity parameters are identified and inflection points are marked. Finally, the window statistical features and the illumination intensity parameters corresponding to the inflection points are combined and encoded to obtain brightness fluctuation features.

[0057] The system obtains real-time current and voltage parameters of each lighting terminal from the distributed lighting terminal and calculates instantaneous power. It then performs time integration on the instantaneous power to obtain a cumulative energy consumption value. Finally, it normalizes the cumulative energy consumption value and a preset benchmark energy consumption value to obtain energy consumption parameters.

[0058] The brightness fluctuation features are arranged into a brightness feature matrix according to the spatial location of the lighting area, and the energy consumption parameters are arranged into an energy consumption feature vector according to the lighting terminal. A mapping matrix is ​​established based on the coverage relationship between the lighting terminal and the spatial location. The brightness feature matrix is ​​spatially aligned and transformed based on the mapping matrix to obtain an aligned brightness feature matrix. The aligned brightness feature matrix and the energy consumption feature vector are spliced ​​and dimension-reduced by projection to obtain the operating state vector.

[0059] Illumination intensity parameters at multiple spatial locations within a lighting area are acquired using distributed sensing devices. Each sensing device includes a photosensitive sensor and a data acquisition module. The photosensitive sensor measures the ambient light intensity and outputs an analog voltage signal. The data acquisition module converts the analog voltage signal into digital illumination intensity parameters, ranging from 0 to 10000 lux. The illumination intensity parameters acquired at each spatial location are arranged according to their corresponding timestamps to form a time series. A sliding window is initialized with a window length of 10 minutes and a step size of 5 minutes, ensuring 50% overlap between adjacent windows. Within each sliding window, the sum of all illumination intensity parameters within the window is divided by the number of data points to obtain the mean. The sum of squared differences between each data point and the mean is calculated and divided by the number of data points to obtain the variance. The mean and variance constitute the window's statistical characteristics. Difference operations are performed on the window statistical characteristics of adjacent windows. The mean of the subsequent window is subtracted from the mean of the previous window, and the difference is divided by the time interval to obtain the rate of change of the mean. The variance is calculated using the same method to obtain the rate of change of the variance. These two values ​​form a rate of change sequence. The sequence of rate of change is iterated. When the mean rate of change is greater than a preset threshold of 0.5 lux per minute, it is marked as an upward phase; when it is less than -0.5 lux per minute, it is marked as a downward phase. The moment of phase transition is marked as the turning point. The mean and variance in the window statistical features are concatenated with the light intensity parameter value corresponding to the turning point to form a feature vector. The labels of the upward and downward phases are converted into binary codes through one-hot encoding and appended to the end of the feature vector to obtain the brightness fluctuation features.

[0060] Real-time current and voltage parameters are obtained from distributed lighting terminals. Each lighting terminal has a built-in current transformer and voltage divider circuit. The current transformer outputs a current signal proportional to the load current, which is converted into a voltage signal by a sampling resistor and then converted to a digital signal via analog-to-digital conversion (ADC) to obtain the current parameter. The voltage divider circuit proportionally reduces the supply voltage before inputting it to the ADC to obtain the voltage parameter. The current parameter and voltage parameter are multiplied to obtain the instantaneous power, in watts. The instantaneous power is integrated over time using the trapezoidal integral method. The sum of the instantaneous power at two adjacent sampling times is multiplied by the time interval and then divided by 2. The cumulative energy consumption value, in kilowatt-hours, is obtained by summing the integrals over all time periods. A preset baseline energy consumption value is the product of the rated power of the lighting terminals in the area and the statistical period. The cumulative energy consumption value is divided by the baseline energy consumption value and limited to between 0 and 1 to obtain the energy consumption parameter.

[0061] The brightness fluctuation characteristics of each spatial location are arranged into a matrix according to two-dimensional coordinates, with rows corresponding to the vertical coordinates and columns corresponding to the horizontal coordinates, forming a brightness feature matrix. The energy consumption parameters of each lighting terminal are arranged into a one-dimensional vector according to the terminal number, forming an energy consumption feature vector. A mapping matrix is ​​established based on the correspondence between the physical installation location of the lighting terminal and the spatial location of the sensing device. A matrix element of 1 indicates that the terminal covers the spatial location, and an element of 0 indicates that it does not cover it. The brightness feature matrix is ​​left-multiplied by the mapping matrix to achieve the alignment transformation between the lighting terminal dimension and the spatial location dimension, resulting in an aligned brightness feature matrix. The aligned brightness feature matrix is ​​flattened into a one-dimensional vector and concatenated end-to-end with the energy consumption feature vector to form a high-dimensional feature vector. Principal component analysis is used to perform dimensionality reduction projection on the high-dimensional feature vector. The covariance matrix is ​​calculated, and the eigenvectors corresponding to the three largest eigenvalues ​​are extracted as the projection basis. The high-dimensional feature vector is then projected onto three-dimensional space to obtain the operating state vector.

[0062] In this embodiment, the illuminance of multiple spatial locations within the lighting area is continuously collected by distributed sensing devices. A sliding window based on timestamps is used to perform mean and variance statistical analysis of the illuminance, enabling a phased characterization of the illuminance change process. This accurately identifies the brightness increase and decrease phases, thereby improving the perception accuracy of dynamic fluctuations in the lighting environment and more comprehensively reflecting the changing trends of the lighting state. This enhances the stability and reliability of the lighting state characteristics. Instantaneous power is calculated from the real-time current and voltage parameters of the lighting terminals, and cumulative energy consumption is obtained through time integration. This allows for comparative analysis of energy consumption data from different lighting terminals under different operating conditions on a unified scale, improving the comparability and accuracy of energy consumption assessment. By establishing a coverage relationship mapping between lighting terminals and spatial locations and performing spatial alignment transformation on the brightness feature matrix, spatial illuminance information and the corresponding lighting terminal's energy consumption information are associated and expressed under the same structure, significantly improving the accuracy of data association analysis.

[0063] In one alternative implementation,

[0064] Based on the topological connection relationship between lighting terminals, a region adjacency matrix is ​​constructed, and graph convolution operation is performed on the running state vector to obtain node evaluation values ​​and region coupling degrees, including:

[0065] The signal transmission delay and data interaction frequency between lighting terminals are obtained and a spatiotemporal coupling tensor is constructed. The spatiotemporal coupling tensor is decomposed and reconstructed by tensor decomposition and outer product to obtain a connection strength matrix. A region adjacency matrix is ​​constructed based on the connection strength matrix.

[0066] The running state vectors are arranged into a state feature matrix and combined with the region adjacency matrix to perform tensor shrinkage operation to obtain primary aggregation features. The region adjacency matrix is ​​then subjected to spectral decomposition to obtain a sequence of eigenvalues. A frequency domain filter is constructed based on the eigenvalue sequence and the primary aggregation features are filtered to obtain frequency domain enhancement features. An attention enhancement feature is calculated based on the frequency domain enhancement feature and the state feature matrix. The set of neighboring nodes for each lighting terminal is determined based on the region adjacency matrix, and the attention enhancement features are subjected to neighborhood aggregation to obtain propagation features. Node evaluation values ​​are then determined using graph pooling.

[0067] Based on the region adjacency matrix, a multi-hop neighborhood topological subgraph is extracted and isomorphic encoding is performed to obtain a topological pattern vector. A structural similarity matrix is ​​constructed based on the topological pattern vector and bilinearly pooled with the state feature matrix to obtain structural awareness features. Graph comparison learning is performed on the structural awareness features to determine information flow. Entropy measurement is performed on the information flow to calculate the region coupling degree.

[0068] The system acquires signal round-trip transmission time data and data packet interaction frequency statistics between lighting terminals within a time window. The time window is divided into multiple time slices of fixed duration. For each time slice, the average signal transmission delay and cumulative data interaction frequency of each terminal pair are extracted and used as the time and spatial dimensions of the spatiotemporal coupling tensor, respectively, to construct a third-order spatiotemporal coupling tensor. CP decomposition is performed on the aforementioned third-order spatiotemporal coupling tensor to obtain a factor matrix set with a preset rank. The resulting spatial pattern factor matrix, temporal pattern factor matrix, and interaction pattern factor matrix are then reconstructed through an outer product operation. The connection strength matrix is ​​recovered through this outer product reconstruction, with matrix elements representing the comprehensive connection strength values ​​between corresponding terminal pairs. After normalizing the connection strength matrix, the diagonal elements are set to zero, resulting in a region adjacency matrix describing the spatial adjacency relationships of the lighting terminals.

[0069] The operating state vectors of each lighting terminal are arranged in order of terminal number to form a state feature matrix, with the matrix rows corresponding to node indices and the columns corresponding to feature dimensions. Tensor contraction is performed on the state feature matrix and the region adjacency matrix, shrinking along the node dimensions using Einstein's summation convention to obtain a primary aggregated feature matrix that fuses neighborhood information. Eigenvalue decomposition is performed on the region adjacency matrix to obtain a descending sequence of eigenvalues. The eigenvectors corresponding to the first few dominant eigenvalues ​​are selected to construct a frequency domain filter matrix, achieving selective amplification of spectral components through eigenvalue weighting. The primary aggregated feature matrix is ​​projected onto the frequency domain space spanned by the eigenvectors, and a left multiplication operation of the filter matrix is ​​performed. Then, an inverse transformation is performed back to the spatial domain to obtain the frequency domain enhanced feature matrix.

[0070] The dot product similarity between the frequency domain enhancement feature matrix and the state feature matrix is ​​calculated and then normalized using softmax to obtain the attention weight matrix. This weight matrix is ​​multiplied by the state feature matrix to obtain the attention enhancement feature. The neighbor node index set for each lighting terminal is determined based on the positions of non-zero elements in the region adjacency matrix. For each central node, its neighbor node feature vectors are extracted from the attention enhancement feature, and mean aggregation is performed. The aggregation result is concatenated with the central node's own features and then passed through a non-linear activation function to obtain the propagation feature. Global pooling is performed on the propagation feature matrix to calculate the norm of each node's feature vector as the node evaluation value.

[0071] Starting from the region adjacency matrix, a breadth-first traversal is performed to extract the two-hop and three-hop neighbor nodes of each node. The extracted multi-hop neighbor nodes and their connecting edges form a topological subgraph. For each topological subgraph, a Weisfeiler-Lehman isomorphism test is performed iteratively to update node labels. The multi-layer node label sequence generated during the iteration is converted into a fixed-length topological pattern vector through hash encoding. The cosine similarity between any two topological pattern vectors is calculated to construct a structural similarity matrix. Bilinear pooling is performed on the structural similarity matrix and the state feature matrix. An interaction between structural similarity and node features is established through a bilinear transformation matrix, resulting in a structure-aware feature matrix containing topological structure-aware information.

[0072] A comparative learning process is performed by constructing positive and negative sample pairs. The original graph is used as the positive sample, while negative sample graphs are generated by randomly deleting edges or perturbing node features. The similarity of the representations of the positive and negative sample pairs in the structure-aware feature space is calculated, and a contrastive loss function is used to optimize the feature extraction process. After optimization, the feature activation intensity on each edge is counted as the information flow value. Shannon entropy is calculated for the information flow distribution within each subgraph region. A higher entropy value indicates a more balanced information propagation within the region. The normalized entropy value is used as the regional coupling index for the region, quantitatively describing the degree of collaborative balance among nodes within the region.

[0073] In this embodiment, by acquiring the signal transmission delay and data interaction frequency between lighting terminals and constructing a spatiotemporal coupling tensor, and then obtaining the connection strength matrix and constructing a region adjacency matrix through tensor decomposition and outer product reconstruction, the communication relationship and interaction strength between lighting terminals can be expressed in a structured way, improving the accuracy of network topology modeling and more realistically reflecting the cooperative operation relationship between lighting terminals. By constructing the operating state vector into a state feature matrix and combining it with the region adjacency matrix for feature aggregation, and then using spectral decomposition to construct a frequency domain filter to perform frequency domain enhancement processing on the aggregated features, key change patterns can be extracted from the frequency domain space of the graph structure. By effectively suppressing noise information and strengthening key state features, the stability and discriminative ability of state information expression are improved. By introducing frequency domain filtering based on feature value sequences and attention enhancement mechanisms, and combining the propagation aggregation of neighboring nodes with graph pooling evaluation, the importance of different lighting terminals in the network can be dynamically identified, improving the sensitivity and recognition ability to key equipment state changes. By extracting multi-hop neighborhood topology subgraphs and performing subgraph isomorphic encoding, a structural similarity matrix is ​​further constructed and bilinear pooling is performed in combination with state features, which can simultaneously perceive operating state information and topological structure information, improving the ability to recognize complex network structure patterns.

[0074] In one alternative implementation,

[0075] Hierarchical clustering of the node evaluation values ​​yields the load distribution pattern. Based on the regional coupling degree, cross-regional influence factors are calculated, and the hierarchical dependencies are obtained by combining the node set corresponding to the load distribution pattern.

[0076] The node evaluation values ​​are divided into multiple load levels. The variance of the node evaluation values ​​of the lighting terminals in each load level is calculated, and the load levels are adjusted to obtain optimized load levels. The set of lighting terminals corresponding to each optimized load level is extracted and marked as a load distribution pattern. Based on the region adjacency matrix, the number of connection edges in each load distribution pattern is counted and the internal connection density is calculated. The number of cross-pattern connection edges between different load distribution patterns is counted. Based on the number of cross-pattern connection edges and the internal connection density, a pattern coupling strength matrix is ​​constructed.

[0077] Based on the load distribution pattern, the regional coupling degree is grouped and the mean regional coupling degree of each load distribution pattern is calculated to obtain the mean mode coupling. The product of the mean mode coupling and the corresponding element in the mode coupling strength matrix is ​​calculated to obtain the initial cross-regional influence factor. The initial cross-regional influence factor is attenuated and corrected according to the topological distance, and the corrected cross-regional influence factor is obtained by combining the pre-acquired frequency domain enhancement features.

[0078] The lighting terminal with the highest node evaluation value is selected from the node set corresponding to the load distribution mode as the mode representative node. The topology hop count between different mode representative nodes is calculated and the inter-mode dependency strength is determined by combining the modified cross-regional influence factor. A directed dependency graph is constructed based on the inter-mode dependency strength and the topology is sorted to obtain the hierarchical dependency relationship.

[0079] After obtaining the node evaluation values ​​corresponding to all lighting terminals, the evaluation values ​​are divided into several initial load levels according to their numerical values. For each initial load level, the variance of all node evaluation values ​​within the level is calculated. When the variance within a certain load level exceeds a preset threshold, the level is divided into two sub-levels, with the sub-level boundaries positioned at the median of the evaluation values ​​within the level. This level division operation is repeated until the variance within all levels is below the preset threshold, at which point the optimized load level set is obtained. The set of lighting terminal numbers within each optimized load level is extracted, and each set is labeled as a load distribution pattern.

[0080] For each marked load distribution pattern, the number of connection edges between lighting terminals within the pattern is counted based on the region adjacency matrix, and denoted as the internal edge count. The internal edge count is divided by the square of the number of lighting terminals within the pattern to obtain the internal connection density of the current pattern. For any two different load distribution patterns, the number of connection edges between lighting terminals within one pattern and those within the other pattern in the region adjacency matrix is ​​counted, and denoted as the cross-pattern connection edge count. A pattern coupling strength matrix is ​​constructed, where the matrix row and column indices correspond to each load distribution pattern, and the matrix element values ​​are the ratios of the cross-pattern connection edge count to the sum of the internal connection densities of the two patterns.

[0081] The regional coupling degree is grouped according to the load distribution pattern to which the lighting terminals belong. For each load distribution pattern, the average regional coupling degree of all lighting terminals within the pattern is calculated to obtain the pattern coupling mean vector. Elements of the pattern coupling strength matrix are extracted and multiplied by the pattern coupling mean of the load distribution pattern to obtain the initial cross-regional influence factor between patterns.

[0082] For the initial cross-regional impact factor, the shortest topological distance between nodes is calculated, denoted as the topological hop count. An attenuation coefficient α is set, and the initial cross-regional impact factor is multiplied by α raised to the power of the topological hop count to achieve distance attenuation correction. Simultaneously, a Fast Fourier Transform (FFT) is performed on the collected illumination intensity parameters, and the top three frequency components with the highest energy percentage in the transformed spectrum are extracted. The amplitudes of these frequency components are normalized and then weighted and summed to obtain the frequency domain enhancement feature. The attenuation-corrected cross-regional impact factor is multiplied by the frequency domain enhancement feature to obtain the corrected cross-regional impact factor.

[0083] Iterate through the node set corresponding to each load distribution pattern, selecting the lighting terminal with the highest node evaluation value as the representative node of that pattern. Calculate the shortest path hop count between any two pattern representative nodes in the topology. Multiply the reciprocal of the hop count by the corresponding modified cross-regional influence factor to obtain the inter-pattern dependency strength between the two patterns. Construct a directed dependency graph, where nodes represent load distribution patterns and directed edge weights represent inter-pattern dependency strengths. Perform a topological sort on the directed dependency graph, prioritizing nodes with zero in-degree and the smallest sum of dependency strengths. The sorting result represents the hierarchical dependency relationship, with patterns with lower dependency strengths located in shallower levels and patterns with higher dependency strengths located in deeper levels.

[0084] In this embodiment, by dividing the node evaluation values ​​into multiple load levels and adaptively adjusting the levels according to the variance of the node evaluation values ​​within each load level, the load characteristics of lighting terminals within the same level become more consistent, forming a more representative load distribution pattern. This improves the rationality and stability of the load pattern division and more accurately reflects the distribution pattern of lighting terminals under different operating loads. By combining the regional adjacency matrix to statistically analyze the internal connection density of each load distribution pattern and the number of cross-mode connection edges between different patterns, and constructing a mode coupling strength matrix accordingly, the structural correlation between different load distribution patterns can be quantitatively expressed, improving the ability to characterize the overall structural relationship. By grouping and statistically analyzing the regional coupling degree according to the load distribution pattern and calculating the cross-regional influence factor in conjunction with the mode coupling strength matrix, and then correcting it through topological distance attenuation and frequency domain enhancement features, the influence relationship between modes can simultaneously consider spatial structural distance, information propagation characteristics, and state change characteristics, improving the accuracy and reliability of cross-regional correlation analysis.

[0085] In one alternative implementation,

[0086] Based on the hierarchical dependencies, a multi-level topology is constructed and inter-layer propagation paths are extracted. The propagation delay coefficient is then determined based on these inter-layer propagation paths, including:

[0087] Based on the hierarchical dependency relationship, the hierarchical sorting result is determined and the lighting terminals are assigned to the corresponding hierarchical node sets. Based on the dependency edges of the pre-constructed directed dependency graph, inter-layer connection edges are established between the hierarchical node sets to obtain an inter-layer connection edge set. The number of lighting terminals in the hierarchical node sets and the number of inter-layer connection edges between different levels are counted to construct an inter-layer connection density matrix. A multi-level topology structure is constructed based on the inter-layer connection edge set.

[0088] Lighting terminals are selected from the highest-level node set of the multi-level topology and the inter-layer connection edge set is traversed downwards to obtain the inter-layer conduction path set. The node set pairs passed through by each inter-layer conduction path are extracted and the corresponding connection density is queried based on the inter-layer connection density matrix. The effective inter-layer conduction path set is determined based on the connection density. The path connection weight corresponding to each effective inter-layer conduction path is extracted from the region adjacency matrix. The path conduction efficiency is calculated based on the length of the effective inter-layer conduction path and the path connection weight.

[0089] Extract the node evaluation values ​​of the starting node and the ending node on the effective interlayer conduction path and calculate the difference between the node evaluation values. Calculate the conduction impedance based on the difference between the node evaluation values ​​and the path conduction efficiency. Extract the proportion of low-frequency components of the starting node based on the predetermined frequency domain enhancement features and perform frequency domain correction on the conduction impedance to obtain the conduction delay coefficient.

[0090] When constructing a multi-level topology, the identified lighting terminals are first assigned based on hierarchical dependencies. All lighting terminals marked as belonging to the same level in the hierarchical dependency relationship are assigned to their corresponding hierarchical node sets, forming a hierarchical ranking result. This ranking result reflects the priority position of each lighting terminal in the control logic. After completing the hierarchical node set division, inter-level connections are established using a pre-constructed directed dependency graph. The dependency edges in the directed dependency graph represent the control transmission direction between lighting terminals. When the starting and ending points of a dependency edge are located in different hierarchical node sets, an inter-level connection edge is established between these two hierarchical node sets. All inter-level connection edges constitute the inter-level connection edge set.

[0091] To quantify the tightness of inter-layer connections, the number of lighting terminals contained in each layer's node set is counted, along with the number of inter-layer connection edges between any two layer node sets. For example, the number of connection edges between the i-th and j-th layers is divided by the product of the number of lighting terminals in the two layers to obtain a normalized connection density value. The connection density values ​​between all layer pairs form an inter-layer connection density matrix. This matrix reflects the activity level of control signal transmission between different layers. Based on the inter-layer connection edge sets and the inclusion relationships of each layer's node sets, a complete multi-level topology is constructed.

[0092] Starting from the highest-level node set in the multi-level topology, a lighting terminal is selected as the traversal starting point, and the process proceeds layer by layer downwards along the set of inter-layer connection edges. During the traversal, all traversed lighting terminals and their respective levels are recorded until the lowest level is reached or further downward propagation is impossible, forming an inter-layer transmission path. This traversal operation is performed on all possible starting points to obtain a set of inter-layer transmission paths. For each inter-layer transmission path, the pairs of adjacent level node sets it traverses are extracted, and the connection density value at the corresponding position is queried in the inter-layer connection density matrix. When the connection density value of all level pairs contained in a path is greater than a preset threshold, the path is included in the set of valid inter-layer transmission paths, ensuring that the selected path has sufficient control transmission capability.

[0093] The adjacency weights of adjacent lighting terminals on effective inter-layer transmission paths are extracted from the region adjacency matrix, and the path connection weights are obtained by summing all the adjacency weights on the path. The path transmission efficiency is calculated as the ratio of the path connection weight to the path length, where the path length is defined as the number of lighting terminals traversed by the path minus one. This ratio reflects the control signal transmission intensity per unit path span.

[0094] The node evaluation values ​​of the starting and ending nodes of the effective inter-layer conduction path are extracted, and the absolute difference between them is calculated as the node evaluation value difference. The conduction impedance is calculated by dividing the node evaluation value difference by the path conduction efficiency, representing the equivalent obstruction encountered by the control signal when propagating along the path. To further consider the impact of frequency domain characteristics on conduction delay, the proportion of low-frequency components corresponding to the starting node is extracted from the predetermined frequency domain enhancement features. The conduction impedance is multiplied by the sum of the low-frequency component proportion and an empirical coefficient to complete the frequency domain correction, obtaining the conduction delay coefficient. The conduction delay coefficient comprehensively reflects the path conduction characteristics and frequency domain response characteristics, providing a quantitative basis for subsequent timing sequencing.

[0095] In this embodiment, the hierarchical order of lighting terminals is determined based on hierarchical dependencies, and the terminals are assigned to corresponding hierarchical node sets. Inter-layer connections are established between different levels based on the directed dependency graph, and an inter-layer connection density matrix is ​​constructed. This allows the dependencies between terminals in the lighting system to be organized and expressed in a multi-level topological structure, more clearly depicting the association strength and structural distribution between nodes at different levels. This improves the ability to express the hierarchical characteristics of complex lighting network structures and more accurately reflects the hierarchical dependencies between terminals. By traversing downwards from higher-level nodes in the multi-level topology to form inter-layer transmission paths, and combining inter-layer connection density to filter effective inter-layer transmission paths, critical paths with practical transmission significance in the network structure can be identified. This avoids interference from invalid or weak connection paths, improving the effectiveness of path identification and the reliability of analysis results. By combining path connection weights and path lengths to calculate path transmission efficiency, the impact of connection strength and transmission distance on information or state propagation capabilities can be comprehensively considered, improving the accuracy of characterizing network transmission characteristics.

[0096] Figure 2 This is a flowchart illustrating the inter-level conduction control timing optimization of the multi-level collaborative balancing method for a smart city lighting system according to an embodiment of the present invention.

[0097] exist Figure 2 middle,

[0098] Based on the conduction delay coefficient, the inter-layer conduction paths are time-series ordered to obtain a control priority sequence. The control trigger time of each level node is determined according to the control priority sequence, and the control amplitude of each level node is determined in conjunction with the node evaluation value, including:

[0099] The conduction delay coefficients corresponding to each effective inter-layer conduction path in the effective inter-layer conduction path set are sorted in ascending order. The lighting terminal sequence contained in each effective inter-layer conduction path is extracted in the sorting order and combined sequentially to obtain the control priority sequence. The position index of each lighting terminal is extracted from the control priority sequence and the initial trigger time is calculated based on the position index. The connection density corresponding to the set of nodes of each layer to which each lighting terminal belongs is extracted from the inter-layer connection density matrix and the initial trigger time is adjusted based on the connection density to obtain the control trigger time.

[0100] Extract the node evaluation value corresponding to each lighting terminal in the control priority sequence and calculate the mean of the evaluation values ​​of the control priority sequence. Calculate the deviation between the node evaluation value corresponding to each lighting terminal and the mean of the evaluation values ​​to obtain the evaluation deviation. Extract the path conduction efficiency of the effective inter-layer conduction path where each lighting terminal is located and perform conduction efficiency weighting on the evaluation deviation based on the path conduction efficiency to obtain the weighted evaluation deviation. Determine the control amplitude of each level node based on the weighted evaluation deviation.

[0101] After obtaining the conduction delay coefficients, the inter-layer conduction paths need to be time-sequentially ordered to form a control priority sequence. The conduction delay coefficient of each path is extracted from the set of effective inter-layer conduction paths, and then sorted in ascending order. After sorting, paths with smaller conduction delay coefficients are listed first, indicating faster signal transmission speeds and thus requiring priority control. Following the sorting order, the lighting terminal sequences contained in each effective inter-layer conduction path are extracted sequentially. For example, if the first path contains terminals A, B, and C, and the second path contains terminals D and E, these sequences are combined sequentially to form the control priority sequence ABCDE.

[0102] Determine the control trigger time by extracting the position index of each lighting terminal from the control priority sequence. The position index represents the terminal's position in the sequence; for example, terminal A's position index is 1, and terminal E's position index is 5. Calculate the initial trigger time based on the position index by multiplying it by a preset time interval. The preset time interval is typically set to 0.5 to 2 seconds to ensure that control commands are transmitted sequentially without conflict. Since the connection density varies between nodes at different levels, the initial trigger time needs adjustment. Extract the connection density value corresponding to the node set of the level to which each lighting terminal belongs from the inter-level connection density matrix. A higher connection density value indicates a stronger connection between the node at that level and other nodes, requiring more time for control response. Multiply the initial trigger time by a connection density adjustment factor, calculated as 1 plus the ratio of the connection density value to a threshold, typically set to 0.3. After adjustment, the control trigger time for each lighting terminal is obtained.

[0103] The control amplitude is determined, and the node evaluation value corresponding to each lighting terminal in the control priority sequence is extracted. The node evaluation value is obtained through graph convolution operation, reflecting the importance of the node in the entire network. The arithmetic mean of all node evaluation values ​​in the control priority sequence is calculated to obtain the mean evaluation value. For each lighting terminal, the difference between its node evaluation value and the mean evaluation value is calculated, and the absolute value is divided by the mean evaluation value to obtain the evaluation deviation. The evaluation deviation reflects the degree of difference between the terminal and the overall level. Further, the path conduction efficiency of the effective inter-layer conduction path of each lighting terminal is extracted. The path conduction efficiency is calculated by combining the path length and the reciprocal of the conduction delay coefficient; a larger value indicates a better conduction effect. The evaluation deviation is multiplied by the path conduction efficiency, and a conduction efficiency weighting operation is performed to obtain the weighted evaluation deviation. The weighted evaluation deviation takes into account both the node's own state and path transmission characteristics. The control range is determined based on the weighted evaluation deviation. The control range is proportional to the weighted evaluation deviation. The control range is usually set as the weighted evaluation deviation multiplied by the benchmark control coefficient. The benchmark control coefficient ranges from 0.2 to 0.8 and is used to map the deviation to the actual executable control range.

[0104] In this embodiment, by sorting the conduction delay coefficients corresponding to the effective inter-layer conduction paths and extracting the lighting terminal sequence in the path based on the sorting results to construct the control priority sequence, it is more in line with the internal state propagation law of the system, which can reduce control conflicts and response lags and improve the coordination and stability of the overall control process. By combining the position index of the lighting terminal in the control priority sequence to calculate the initial trigger time and using the inter-layer connection density to dynamically adjust the trigger time, the matching degree between the control trigger mechanism and the network structure characteristics is improved. By calculating the deviation of the evaluation value of each lighting terminal node in the control priority sequence from the overall mean and weighting it in combination with the path conduction efficiency, the importance of the node state and its conduction ability in the network can be considered at the same time, and the key nodes that need to be controlled more accurately can be identified more accurately. By determining the control amplitude of each level node based on the weighted evaluation deviation, nodes at different levels can be controlled differently according to the deviation of the operating state and the influence ability in the network, which can effectively improve the flexibility and accuracy of the lighting system operation control and help improve the overall operating efficiency and energy consumption optimization effect.

[0105] In one alternative implementation,

[0106] The modified control amplitude is obtained by comparing and verifying the control amplitude based on preset energy consumption constraints. Based on the modified control amplitude, a balanced control command is generated, and the lighting terminals corresponding to each level node are controlled step-by-step, including:

[0107] Extract the global energy consumption upper limit and the hierarchical energy consumption upper limit set from the preset energy consumption constraints. Calculate the estimated energy consumption of each lighting terminal based on the control amplitude and accumulate them hierarchically to obtain the hierarchical energy consumption distribution. Compare the hierarchical energy consumption distribution with the hierarchical energy consumption upper limit set to identify the energy consumption exceeding the limit level and extract the corresponding path conduction efficiency to calculate the reduction coefficient. Based on the reduction coefficient, reduce the control amplitude of the lighting terminals within the energy consumption exceeding the limit level to obtain the hierarchical corrected control amplitude. Calculate the global cumulative energy consumption based on the hierarchical corrected control amplitude and compensate it in conjunction with the global energy consumption upper limit to obtain the corrected control amplitude.

[0108] Based on the correction control amplitude and the control trigger time, the control parameters of each lighting terminal are determined and a hierarchical control parameter set is constructed according to the hierarchy. The path conduction efficiency is extracted from the pre-acquired inter-layer connection edge set and the inter-layer conduction coefficient is calculated. Based on the inter-layer conduction coefficient, the hierarchical control parameter set is subjected to conduction compensation and a balanced control command is generated.

[0109] Based on the equalization control command and the conduction delay coefficient, the lighting terminals corresponding to each level node are controlled step by step along the multi-level topology.

[0110] Preset energy consumption constraints are obtained, which include two types of limiting parameters: a global energy consumption upper limit, representing the maximum power allowed to be consumed by the entire lighting area within a unit cycle, and a hierarchical energy consumption upper limit set corresponding to the independent energy consumption thresholds of each level of node group. After extracting the control amplitude, the instantaneous power increment of each lighting terminal after the control action is calculated, and the estimated energy consumption is obtained by combining it with the terminal's baseline power. The estimated energy consumption of all terminals is categorized and accumulated according to their respective levels to form a hierarchical energy consumption distribution array, where each element corresponds to the total estimated energy consumption of a level.

[0111] The hierarchical energy consumption distribution array is compared item by item with the hierarchical energy consumption upper limit set. When the estimated energy consumption of a certain level exceeds the corresponding upper limit, the level is marked as an energy consumption exceeding limit level. All internal transmission paths of the energy consumption exceeding limit level are extracted from the multi-level topology. The average adjacency degree and edge weight of the nodes on the path are calculated to determine the path transmission efficiency. The path transmission efficiency reflects the propagation speed and energy conversion ratio of the control signal within the level; a higher value indicates a more sensitive control response. A reduction coefficient is constructed based on the reciprocal of the path transmission efficiency. This coefficient is used to quantify the proportion of control amplitude that needs to be reduced.

[0112] For each lighting terminal within the energy consumption exceedance level, its original control amplitude is multiplied by a reduction factor to obtain the tiered corrected control amplitude. Priority is given to reducing the control amount of terminals on paths with lower conduction efficiency to ensure the control effect on critical paths. The corrected control amplitudes of all levels after reduction are summarized, and the total energy consumption of each level is recalculated and accumulated based on the estimated energy consumption of the terminals to obtain the global cumulative energy consumption. The global cumulative energy consumption is compared with the global energy consumption upper limit. If there is still a difference exceeding the limit, compensation is allocated according to the energy consumption proportion of each level. The compensation amount is achieved through secondary fine-tuning based on the tiered corrected control amplitude to obtain a corrected control amplitude that satisfies both global and tiered constraints.

[0113] Based on the correction adjustment range and the adjustment trigger time, a control parameter structure is generated for each lighting terminal. This structure includes the terminal identifier, the corrected brightness adjustment percentage, and the timestamp of the action trigger. The control parameter structures are grouped according to the terminal's level to form a hierarchical control parameter set. Each subset in the set corresponds to the control information of all terminals at one level.

[0114] The system reads the connection edge information between adjacent layers from a pre-established set of inter-layer connections. Each connection edge carries a path transmission efficiency attribute. The inter-layer transmission coefficient is calculated; this coefficient is equal to the weighted average of the path transmission efficiency of the connection edge and the number of edges, reflecting the intensity of the influence of upper-layer control actions on lower-layer nodes. The system iterates through the set of layer control parameters. For each non-top-level layer, the system extracts the executed control amplitude from the control parameters of the layer above it and multiplies it by the inter-layer transmission coefficient to obtain the transmission compensation amount. This transmission compensation amount is then added to the corrected control amplitude of each terminal in the current layer to ensure that the control actions of lower-layer terminals can offset the illumination influence transmitted from the upper layer, achieving inter-layer coordination. After completing the transmission compensation, the control parameters of all layers are integrated into a unified data packet and encapsulated into a balanced control command.

[0115] Based on the hierarchical information and conduction delay coefficient in the equalization control command, the control execution order is determined along the multi-level topology. The conduction delay coefficient represents the time delay required for signal transmission between layers, and this delay is superimposed on the control triggering time of each level node. Starting from the top-level node, control commands are sent to the lower levels sequentially according to the control priority sequence. After receiving the command, each level node waits for the conduction delay time before executing the brightness adjustment action of the corresponding lighting terminal, thus achieving sequential coordination of hierarchical control.

[0116] In this embodiment, by extracting the global energy consumption upper limit and hierarchical energy consumption upper limit from the preset energy consumption constraints, and performing hierarchical cumulative analysis on the estimated energy consumption of each lighting terminal in conjunction with the control amplitude, it is possible to identify the levels where energy consumption may exceed the limit in advance, thereby improving the foresight and control accuracy of energy consumption constraint management and helping to avoid the problem of energy consumption exceeding the standard. By calculating the reduction coefficient based on the path transmission efficiency for the identified energy consumption exceeding levels, and reducing the control amplitude of the lighting terminals in the corresponding levels, it is possible to maintain the balance of the overall control effect while satisfying the hierarchical energy consumption constraints, thereby improving the flexibility and overall coordination of the energy consumption control strategy. By constructing a hierarchical control parameter set by combining the modified control amplitude and the control trigger time, and using the inter-level transmission coefficient to conduct transmission compensation for the control parameters, the control strategies between different levels can be dynamically coordinated according to the transmission relationship in the network structure, thereby improving the overall consistency of the collaborative control of the multi-level lighting system.

[0117] A second aspect of the present invention provides a multi-level collaborative equalization system for smart city lighting, comprising:

[0118] The data acquisition unit is used to collect illuminance parameters and energy consumption parameters of lighting terminals within the lighting area, perform time-domain sliding window statistics on the illuminance parameters to obtain brightness fluctuation characteristics, construct an operating state vector based on the brightness fluctuation characteristics and the energy consumption parameters, construct a region adjacency matrix based on the topological connection relationship between lighting terminals, and perform graph convolution operation on the operating state vector to obtain node evaluation values ​​and region coupling degree.

[0119] The topology analysis unit is used to perform hierarchical clustering on the node evaluation values ​​to obtain the load distribution pattern, calculate the cross-regional influence factor based on the regional coupling degree and solve the hierarchical dependency relationship by combining the node set corresponding to the load distribution pattern, construct a multi-level topology structure based on the hierarchical dependency relationship and extract the inter-layer transmission path, and determine the transmission delay coefficient based on the inter-layer transmission path.

[0120] The timing control unit is used to sort the inter-layer conduction paths according to the conduction delay coefficient to obtain a control priority sequence, determine the control trigger time of each level node according to the control priority sequence and determine the control amplitude of each level node in combination with the node evaluation value, compare and verify the control amplitude based on the preset energy consumption constraints to obtain a corrected control amplitude, generate a balanced control command based on the corrected control amplitude and perform step-by-step control on the lighting terminals corresponding to each level node.

[0121] A third aspect of the present invention provides an electronic device, comprising:

[0122] A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0123] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0124] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-level collaborative balancing method for smart city lighting systems, characterized in that, include: The illuminance parameters and energy consumption parameters of the lighting terminals within the lighting area are collected. The illuminance parameters are statistically analyzed using a time-domain sliding window to obtain brightness fluctuation characteristics. An operating state vector is constructed based on the brightness fluctuation characteristics and the energy consumption parameters. A regional adjacency matrix is ​​constructed based on the topological connection relationship between the lighting terminals. Graph convolution operation is performed on the operating state vector to obtain node evaluation values ​​and regional coupling degree. The load distribution pattern is obtained by hierarchically clustering the node evaluation values. The cross-regional influence factor is calculated based on the regional coupling degree and the hierarchical dependency relationship is obtained by combining the node set corresponding to the load distribution pattern. A multi-level topology structure is constructed based on the hierarchical dependency relationship and the inter-layer transmission path is extracted. The transmission delay coefficient is determined based on the inter-layer transmission path. Based on the conduction delay coefficient, the inter-layer conduction path is sorted in time to obtain the control priority sequence. The control trigger time of each level node is determined according to the control priority sequence, and the control amplitude of each level node is determined in combination with the node evaluation value. The control amplitude is compared and verified based on the preset energy consumption constraints to obtain the corrected control amplitude. Based on the corrected control amplitude, a balanced control command is generated and the lighting terminals corresponding to each level node are controlled step by step. Based on the topological connection relationship between lighting terminals, a region adjacency matrix is ​​constructed, and graph convolution operation is performed on the running state vector to obtain node evaluation values ​​and region coupling degrees, including: The signal transmission delay and data interaction frequency between lighting terminals are obtained and a spatiotemporal coupling tensor is constructed. The spatiotemporal coupling tensor is decomposed and reconstructed by tensor decomposition and outer product to obtain a connection strength matrix. A region adjacency matrix is ​​constructed based on the connection strength matrix. The running state vectors are arranged into a state feature matrix and combined with the region adjacency matrix to perform tensor shrinkage operation to obtain primary aggregation features. The region adjacency matrix is ​​then subjected to spectral decomposition to obtain a sequence of eigenvalues. A frequency domain filter is constructed based on the eigenvalue sequence and the primary aggregation features are filtered to obtain frequency domain enhancement features. An attention enhancement feature is calculated based on the frequency domain enhancement feature and the state feature matrix. The set of neighboring nodes for each lighting terminal is determined based on the region adjacency matrix, and the attention enhancement features are subjected to neighborhood aggregation to obtain propagation features. Node evaluation values ​​are then determined using graph pooling. Based on the region adjacency matrix, a multi-hop neighborhood topological subgraph is extracted and isomorphic encoding is performed to obtain a topological pattern vector. A structural similarity matrix is ​​constructed based on the topological pattern vector and bilinearly pooled with the state feature matrix to obtain structural awareness features. Graph comparison learning is performed on the structural awareness features to determine information flow. Entropy measurement is performed on the information flow to calculate the region coupling degree.

2. The method according to claim 1, characterized in that, The illuminance parameters within the lighting area and the energy consumption parameters of the lighting terminals are collected. Time-domain sliding window statistics are performed on the illuminance parameters to obtain brightness fluctuation characteristics. Based on the brightness fluctuation characteristics and the energy consumption parameters, an operating state vector is constructed, including: Illumination intensity parameters at multiple spatial locations within the illumination area are acquired using distributed sensing devices. These parameters are then arranged according to their corresponding timestamps and a sliding window is initialized. Within each sliding window, the mean and variance of the illumination intensity parameters are calculated to obtain window statistical features. The window statistical features of adjacent windows are then differentially analyzed to obtain a rate of change sequence. Based on this rate of change sequence, the rising and falling phases of the illumination intensity parameters are identified and inflection points are marked. Finally, the window statistical features and the illumination intensity parameters corresponding to the inflection points are combined and encoded to obtain brightness fluctuation features. The system obtains real-time current and voltage parameters of each lighting terminal from the distributed lighting terminal and calculates instantaneous power. It then performs time integration on the instantaneous power to obtain a cumulative energy consumption value. Finally, it normalizes the cumulative energy consumption value and a preset benchmark energy consumption value to obtain energy consumption parameters. The brightness fluctuation features are arranged into a brightness feature matrix according to the spatial location of the lighting area, and the energy consumption parameters are arranged into an energy consumption feature vector according to the lighting terminal. A mapping matrix is ​​established based on the coverage relationship between the lighting terminal and the spatial location. The brightness feature matrix is ​​spatially aligned and transformed based on the mapping matrix to obtain an aligned brightness feature matrix. The aligned brightness feature matrix and the energy consumption feature vector are spliced ​​and dimension-reduced by projection to obtain the operating state vector.

3. The method according to claim 1, characterized in that, Hierarchical clustering of the node evaluation values ​​yields the load distribution pattern. Based on the regional coupling degree, cross-regional influence factors are calculated, and the hierarchical dependencies are obtained by combining the node set corresponding to the load distribution pattern. The node evaluation values ​​are divided into multiple load levels. The variance of the node evaluation values ​​of the lighting terminals in each load level is calculated, and the load levels are adjusted to obtain optimized load levels. The set of lighting terminals corresponding to each optimized load level is extracted and marked as a load distribution pattern. Based on the region adjacency matrix, the number of connection edges in each load distribution pattern is counted and the internal connection density is calculated. The number of cross-pattern connection edges between different load distribution patterns is counted. Based on the number of cross-pattern connection edges and the internal connection density, a pattern coupling strength matrix is ​​constructed. Based on the load distribution pattern, the regional coupling degree is grouped and the mean regional coupling degree of each load distribution pattern is calculated to obtain the mean mode coupling. The product of the mean mode coupling and the corresponding element in the mode coupling strength matrix is ​​calculated to obtain the initial cross-regional influence factor. The initial cross-regional influence factor is attenuated and corrected according to the topological distance, and the corrected cross-regional influence factor is obtained by combining the pre-acquired frequency domain enhancement features. The lighting terminal with the highest node evaluation value is selected from the node set corresponding to the load distribution mode as the mode representative node. The topology hop count between different mode representative nodes is calculated and the inter-mode dependency strength is determined by combining the modified cross-regional influence factor. A directed dependency graph is constructed based on the inter-mode dependency strength and the topology is sorted to obtain the hierarchical dependency relationship.

4. The method according to claim 1, characterized in that, Based on the hierarchical dependencies, a multi-level topology is constructed and inter-layer propagation paths are extracted. The propagation delay coefficient is then determined based on these inter-layer propagation paths, including: Based on the hierarchical dependency relationship, the hierarchical sorting result is determined and the lighting terminals are assigned to the corresponding hierarchical node sets. Based on the dependency edges of the pre-constructed directed dependency graph, inter-layer connection edges are established between the hierarchical node sets to obtain an inter-layer connection edge set. The number of lighting terminals in the hierarchical node sets and the number of inter-layer connection edges between different levels are counted to construct an inter-layer connection density matrix. A multi-level topology structure is constructed based on the inter-layer connection edge set. Lighting terminals are selected from the highest-level node set of the multi-level topology and the inter-layer connection edge set is traversed downwards to obtain the inter-layer conduction path set. The node set pairs passed through by each inter-layer conduction path are extracted and the corresponding connection density is queried based on the inter-layer connection density matrix. The effective inter-layer conduction path set is determined based on the connection density. The path connection weight corresponding to each effective inter-layer conduction path is extracted from the region adjacency matrix. The path conduction efficiency is calculated based on the length of the effective inter-layer conduction path and the path connection weight. Extract the node evaluation values ​​of the starting node and the ending node on the effective interlayer conduction path and calculate the difference between the node evaluation values. Calculate the conduction impedance based on the difference between the node evaluation values ​​and the path conduction efficiency. Extract the proportion of low-frequency components of the starting node based on the predetermined frequency domain enhancement features and perform frequency domain correction on the conduction impedance to obtain the conduction delay coefficient.

5. The method according to claim 1, characterized in that, Based on the conduction delay coefficient, the inter-layer conduction paths are time-series ordered to obtain a control priority sequence. The control trigger time of each level node is determined according to the control priority sequence, and the control amplitude of each level node is determined in conjunction with the node evaluation value, including: The conduction delay coefficients corresponding to each effective inter-layer conduction path in the effective inter-layer conduction path set are sorted in ascending order. The lighting terminal sequence contained in each effective inter-layer conduction path is extracted in the sorting order and combined sequentially to obtain the control priority sequence. The position index of each lighting terminal is extracted from the control priority sequence and the initial trigger time is calculated based on the position index. The connection density corresponding to the set of nodes of each layer to which each lighting terminal belongs is extracted from the inter-layer connection density matrix and the initial trigger time is adjusted based on the connection density to obtain the control trigger time. Extract the node evaluation value corresponding to each lighting terminal in the control priority sequence and calculate the mean of the evaluation values ​​of the control priority sequence. Calculate the deviation between the node evaluation value corresponding to each lighting terminal and the mean of the evaluation values ​​to obtain the evaluation deviation. Extract the path conduction efficiency of the effective inter-layer conduction path where each lighting terminal is located and perform conduction efficiency weighting on the evaluation deviation based on the path conduction efficiency to obtain the weighted evaluation deviation. Determine the control amplitude of each level node based on the weighted evaluation deviation.

6. The method according to claim 1, characterized in that, The modified control amplitude is obtained by comparing and verifying the control amplitude based on preset energy consumption constraints. Based on the modified control amplitude, a balanced control command is generated, and the lighting terminals corresponding to each level node are controlled step-by-step, including: Extract the global energy consumption upper limit and the hierarchical energy consumption upper limit set from the preset energy consumption constraints. Calculate the estimated energy consumption of each lighting terminal based on the control amplitude and accumulate them hierarchically to obtain the hierarchical energy consumption distribution. Compare the hierarchical energy consumption distribution with the hierarchical energy consumption upper limit set to identify the energy consumption exceeding the limit level and extract the corresponding path conduction efficiency to calculate the reduction coefficient. Based on the reduction coefficient, reduce the control amplitude of the lighting terminals within the energy consumption exceeding the limit level to obtain the hierarchical corrected control amplitude. Calculate the global cumulative energy consumption based on the hierarchical corrected control amplitude and compensate it in conjunction with the global energy consumption upper limit to obtain the corrected control amplitude. Based on the correction control amplitude and the control trigger time, the control parameters of each lighting terminal are determined and a hierarchical control parameter set is constructed according to the hierarchy. The path conduction efficiency is extracted from the pre-acquired inter-layer connection edge set and the inter-layer conduction coefficient is calculated. Based on the inter-layer conduction coefficient, the hierarchical control parameter set is subjected to conduction compensation and a balanced control command is generated. Based on the equalization control command and the conduction delay coefficient, the lighting terminals corresponding to each level node are controlled step by step along the multi-level topology.

7. A multi-level collaborative equalization system for smart city lighting systems, used to implement the method described in any one of claims 1-6, characterized in that, include: The data acquisition unit is used to collect illuminance parameters and energy consumption parameters of lighting terminals within the lighting area, perform time-domain sliding window statistics on the illuminance parameters to obtain brightness fluctuation characteristics, construct an operating state vector based on the brightness fluctuation characteristics and the energy consumption parameters, construct a region adjacency matrix based on the topological connection relationship between lighting terminals, and perform graph convolution operation on the operating state vector to obtain node evaluation values ​​and region coupling degree. The topology analysis unit is used to perform hierarchical clustering on the node evaluation values ​​to obtain the load distribution pattern, calculate the cross-regional influence factor based on the regional coupling degree and solve the hierarchical dependency relationship by combining the node set corresponding to the load distribution pattern, construct a multi-level topology structure based on the hierarchical dependency relationship and extract the inter-layer transmission path, and determine the transmission delay coefficient based on the inter-layer transmission path. The timing control unit is used to sort the inter-layer conduction paths according to the conduction delay coefficient to obtain a control priority sequence, determine the control trigger time of each level node according to the control priority sequence and determine the control amplitude of each level node in combination with the node evaluation value, compare and verify the control amplitude based on the preset energy consumption constraints to obtain a corrected control amplitude, generate a balanced control command based on the corrected control amplitude and perform step-by-step control on the lighting terminals corresponding to each level node.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images