LED intelligent street lamp real-time control system based on edge calculation

Abstract

The invention discloses an LED intelligent street lamp real-time control system based on edge calculation, and the system comprises an edge data processing module which is used for collecting brightness feedback and multi-source input preprocessing in a control box to obtain a sample window set; the space cooperation feature construction module is used for generating an adjacent matrix and calculating a space lag brightness sequence and a diffusion input sequence; the exogenous function self-organization generation module is used for constructing GMDH nodes, performing square parameterization on strong influence coefficients and obtaining an exogenous function sequence according to the sequence of interpretation and combination; the space-time prediction and structure updating module is used for constructing space-time SARIMAX and writing back a foreign function structure according to an information criterion; and the dimming control module is used for issuing a dimming instruction according to the target brightness reference value and returning feedback. Prospective dimming of the edge side street lamp is achieved, the fluctuation and instruction frequency is reduced, and the consistency is improved.

Description

Technical Field

[0001] This invention relates to the field of smart city lighting control, and more particularly to a real-time control system for LED smart streetlights based on edge computing. Background Technology

[0002] With the advancement of smart city construction, road lighting systems are gradually evolving from traditional timed control methods to intelligent and refined control methods. Existing LED street light systems typically obtain brightness feedback information through centralized control or remote platforms, and combine this with parameters such as ambient light and time information to perform dimming control, in order to achieve a balance between energy-saving operation and basic lighting needs.

[0003] In existing technologies, one type of solution relies on real-time brightness feedback, directly generating dimming commands based on the deviation between the currently collected brightness value and a preset threshold. This type of solution depends on instantaneous feedback and does not model the trend of brightness changes. In scenarios with rapid environmental changes or multi-segment linkage, it is prone to problems such as frequent dimming or control lag. Another type of solution introduces a time series prediction model to model the historical brightness data of a single street light or a single road segment. However, it usually only uses local time information and does not fully consider the spatial correlation between adjacent road segments. The prediction results are difficult to reflect the overall evolution characteristics of road lighting.

[0004] Existing technologies often directly incorporate environmental and traffic features as exogenous variables into prediction models. The structure of these exogenous variables is usually fixed before modeling, lacking a mechanism for adaptive adjustment to different input feature structures. As the number of exogenous inputs increases, the complexity of the model structure also increases. The parameter selection process relies on manual experience or simple screening rules, making it difficult to achieve stable operation under edge computing conditions.

[0005] Therefore, how to provide a real-time control system for LED smart streetlights based on edge computing is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a real-time control system for LED smart streetlights based on edge computing. This invention constructs a spatiotemporal SARIMAX structure that integrates spatial collaborative relationships, and combines an exogenous function self-organization generation and an information criterion-driven structure update mechanism to achieve forward-looking adjustment and control of streetlight brightness. It has the advantages of high control stability, strong structural adaptability, and low edge operation burden.

[0007] According to an embodiment of the present invention, a real-time control system for LED smart streetlights based on edge computing includes:

[0008] The edge data processing module is used to collect street light group brightness feedback time-series data and multi-source input time-series data in the street light centralized control box, perform preprocessing on the brightness feedback time-series data and multi-source input time-series data and generate a sample window set;

[0009] The spatial collaborative feature construction module is used to generate an adjacency matrix based on the street light grouping topology, calculate the spatial lag brightness sequence for the brightness feedback time series data in the sample window set, and calculate the spatial diffusion input sequence for the multi-source input time series data.

[0010] The exogenous function self-organizing generation module is used to construct GMDH binary quadratic polynomial nodes, perform square parameterization on the coefficients corresponding to the strong influence features to generate non-negative coefficients to form basic exogenous functions, and introduce weak influence features in the order of interpretation and combination to generate low-order interactive exogenous functions, forming an exogenous function sequence.

[0011] The spatiotemporal prediction and structure update module is used to construct a spatiotemporal SARIMAX structure that incorporates spatial coordination and exogenous functions. It writes spatial lag brightness sequences, exogenous function sequences, and spatial diffusion input sequences. Based on information criteria, it selects exogenous function sequences to form retained exogenous function sequences and writes them into the spatiotemporal SARIMAX structure to form a spatiotemporal prediction model for target brightness.

[0012] The dimming control module is used to output the target brightness reference value according to the target brightness spatiotemporal prediction model and generate dimming commands to be sent to the street light drive device, collect execution feedback and send it back to the edge data processing module.

[0013] Optionally, modules can be integrated using the following methods:

[0014] S1. Obtain the street light group brightness feedback time series and multi-source input time series. The multi-source input time series consists of strong influence features and weak influence features. Perform preprocessing on the brightness feedback time series and multi-source input time series to generate a sample window set.

[0015] S2. Generate an adjacency matrix based on the street light grouping topology, calculate the spatially delayed luminance sequence based on the luminance feedback time series based on the adjacency matrix, and calculate the spatially diffused input sequence based on the multi-source input time series based on the adjacency matrix.

[0016] S3. Construct the first layer of GMDH binary quadratic polynomial nodes. The coefficients in the polynomial nodes that are related to the strong influence features are generated into non-negative coefficients through square parameterization. Output the basic exogenous function sequence.

[0017] S4. Take the basic exogenous function sequence as input and introduce weak influence features in the order of interpretation followed by combination to generate low-order interactive polynomial nodes, and output the exogenous function sequence.

[0018] S5. Construct a spatiotemporal SARIMAX structure that incorporates spatial coordination and exogenous functions, and write the spatial lag brightness sequence, exogenous function sequence and spatial diffusion input sequence into the spatiotemporal SARIMAX structure to form a spatiotemporal prediction model for target brightness.

[0019] S6. Perform parameter estimation on the spatiotemporal SARIMAX structure based on the sample window set, and select the exogenous function sequence according to the information criterion to form a retained exogenous function sequence and write it into the spatiotemporal SARIMAX structure.

[0020] S7. Based on the spatiotemporal SARIMAX structure and the retained exogenous function sequence, output the target brightness reference value, generate a dimming command based on the target brightness reference value and send it to the street light driver, collect and execute feedback to update the sample window set.

[0021] Optionally, S2 specifically includes:

[0022] Based on the physical arrangement and connection relationships of street light groups in the road, the topology information of the street light groups is obtained. The topology information of the street light groups includes the adjacency relationships between each street light group. An adjacency matrix is ​​constructed according to the number of groups.

[0023] The number of elements with a value of 1 in the corresponding row of the adjacency matrix for each street light group is counted to obtain the number of adjacent groups. For each sampling time and each street light group in the sample window set, the brightness feedback values ​​of the adjacent groups at the sampling time are read and summed. The summation result is divided by the number of adjacent groups to obtain the spatial lag brightness value. When the number of adjacent groups is 0, the spatial lag brightness value is set as the brightness feedback value of the street light group at the sampling time. The spatial lag brightness values ​​of each sampling time are arranged in chronological order to obtain the spatial lag brightness sequence.

[0024] For each sampling time, each street light group, and each input feature in the multi-source input time series in the sample window set, the input feature values ​​of adjacent groups at the sampling time are read and summed. The summation result is divided by the number of adjacent groups to obtain the spatial diffusion input value. When the number of adjacent groups is 0, the spatial diffusion input value is set as the input feature value of the street light group at the sampling time. The spatial diffusion input values ​​of each sampling time are arranged in chronological order to obtain the spatial diffusion input sequence.

[0025] Optionally, S3 specifically includes:

[0026] Extract strong influence feature sequences from the sample window set. The strong influence feature sequences include ambient lighting data and time-stamped data.

[0027] Construct a set of binary quadratic polynomial nodes in the first layer of GMDH, and select two strongly influential features as input variables for each binary quadratic polynomial node to generate a polynomial expression containing constant terms, linear terms, square terms and cross terms;

[0028] Perform quadratic parameterization on the coefficients of the first-order term, the quadratic term, and the cross term corresponding to the strong influence features in the polynomial expression, express the corresponding coefficients in quadratic form, and use the squared result as the coefficient value to form a bivariate quadratic polynomial node with non-negative coefficient constraints.

[0029] Two strongly influential features at each sampling time in the sample window set are input into the binary quadratic polynomial node to calculate the node output value sequence; the output value sequences of each binary quadratic polynomial node are aggregated according to the node number to form the basic exogenous function sequence.

[0030] Optionally, S4 specifically includes:

[0031] Using the sequence of basic exogenous functions as input, the exogenous function expansion operation is performed according to the node generation order constraint rule; a single basic exogenous function is selected from the sequence of basic exogenous functions as the main input term, and weakly influential features in the sample window set are introduced as auxiliary input terms to construct a low-order polynomial expression containing only first-order terms and cross terms, and generate low-order interactive polynomial nodes.

[0032] For low-order interactive polynomial nodes, no squared terms or higher-order cross terms are introduced; only the interaction relationship between the basic exogenous function and the weakly influential feature is retained. The values ​​of the basic exogenous function and the weakly influential feature corresponding to each sampling time in the sample window set are input into the low-order interactive polynomial node to calculate the node output value sequence. The output value sequence of each low-order interactive polynomial node is merged with the basic exogenous function sequence to form the exogenous function sequence.

[0033] Optionally, the exogenous function self-organization generation process adopts a conditional triggering approach of interpretation followed by combination, specifically including:

[0034] Using the sequence of basic exogenous functions as the output of the interpretation layer, the monotonic direction of the output sequence corresponding to each basic exogenous function is determined in the sample window set according to the time order, and the sequence of changes in the direction of the basic exogenous function between adjacent sampling times is recorded.

[0035] A consistency check is performed on the sequence of changing directions within a preset continuous sampling window. When the changing direction of the basic exogenous function remains consistent within the continuous sampling window, the basic exogenous function is determined to meet the structural consistency condition. When the basic exogenous function reverses its direction within the continuous sampling window, the basic exogenous function is determined not to meet the structural consistency condition.

[0036] Combination operations are performed only on the basic exogenous functions that satisfy the structural consistency condition. The basic exogenous functions are used as the main input terms, and weakly influential features are introduced as auxiliary input terms to construct low-order interactive polynomial nodes containing only first-order terms and cross terms, thereby generating low-order interactive exogenous functions.

[0037] For basic exogenous functions that do not meet the structural consistency condition, it is forbidden to introduce weak influence features and the original basic exogenous function structure remains unchanged.

[0038] The generated low-order interactive exogenous functions are merged with the basic exogenous functions that have not undergone combination operations to form a sequence of exogenous functions.

[0039] Optionally, S5 specifically includes:

[0040] The exogenous function output sequence corresponding to each street light group in the sample window set is aligned with the exogenous function output sequence corresponding to the adjacent street light group according to the time order.

[0041] Time offset determination is performed on the output sequence of the same exogenous function in the current street light group and the adjacent street light group. The time offset determination determines the time lead or time lag relationship of the exogenous function in the current street light group relative to the adjacent street light group by comparing the consistency order of the output change direction of the exogenous function at adjacent sampling times.

[0042] After determining the time lead relationship or time lag relationship, a time remapping operation is performed on the exogenous function output sequence corresponding to the current street light group. The time remapping operation shifts the time index of the exogenous function output sequence forward or backward according to the determined time offset relationship. The exogenous function output sequence of the current street light group is consistent with the exogenous function output sequence of the adjacent street light group in terms of time evolution order.

[0043] After completing the time remapping operation, the time-aligned exogenous function sequence is written into the spatiotemporal SARIMAX structure, and the street light group brightness feedback sequence is written into the spatiotemporal SARIMAX structure at the same time. Joint modeling is performed on the time-aligned exogenous function sequence and the brightness feedback sequence to form a spatiotemporal prediction model of the target brightness.

[0044] Optionally, S6 specifically includes:

[0045] Perform parameter estimation on the spatiotemporal SARIMAX structure of the written exogenous function sequence on the sample window set;

[0046] During parameter estimation, the structure of the exogenous functions involved in the modeling is selected based on the information criterion.

[0047] Based on the comparison results of the information criteria, a set of retained exogenous functions is determined, and the set of retained exogenous functions is written back into the spatiotemporal SARIMAX structure as the exogenous function structure;

[0048] Based on the rewritten exogenous function structure, parameter estimation is re-performed to form a spatiotemporal SARIMAX model.

[0049] The beneficial effects of this invention are:

[0050] (1) By constructing a spatiotemporal prediction structure that integrates the topological relationship of street lamp groups on the edge side, the brightness modeling process can simultaneously consider the temporal evolution relationship and the spatial correlation relationship, avoid relying solely on historical data of a single street lamp or a single road segment for control, reduce the dimming inconsistency problem in multi-road segment linkage scenarios, and improve the coordination and stability of overall lighting control.

[0051] (2) By using the self-organizing generation of exogenous functions and the structure writing back mechanism based on information criteria, the structure of exogenous functions is dynamically updated during the parameter estimation process, avoiding the redundancy and mismatch problems caused by fixing the exogenous input structure before modeling. Even when the number of multi-source inputs changes, the controllability and scalability of the model structure can still be maintained, making it suitable for application environments with limited edge computing resources.

[0052] (3) By processing the time alignment of exogenous functions between different street light groups, the exogenous information can be spatially consistent with the time remapping before entering the spatiotemporal prediction model, reducing the impact of the temporal misalignment of exogenous information on the prediction results, making the generated target brightness reference value more consistent with the actual evolution process of road lighting, and supporting data updates and model iterations within the continuous control cycle. Attached Figure Description

[0053] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0054] Figure 1 This is a flowchart of a real-time control system for LED smart streetlights based on edge computing proposed in this invention.

[0055] Figure 2 This is a schematic diagram of the combination of GMDH and spatiotemporal SARIMAX in a real-time control system for LED smart streetlights based on edge computing proposed in this invention.

[0056] Figure 3 This is a schematic diagram of the exogenous function time alignment of a real-time control system for LED smart streetlights based on edge computing proposed in this invention. Detailed Implementation

[0057] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0058] refer to Figure 1-3 A real-time control system for LED smart streetlights based on edge computing, comprising:

[0059] The edge data processing module is used to collect street light group brightness feedback time-series data and multi-source input time-series data in the street light centralized control box, perform preprocessing on the brightness feedback time-series data and multi-source input time-series data and generate a sample window set;

[0060] The spatial collaborative feature construction module is used to generate an adjacency matrix based on the street light grouping topology, calculate the spatial lag brightness sequence for the brightness feedback time series data in the sample window set, and calculate the spatial diffusion input sequence for the multi-source input time series data.

[0061] The exogenous function self-organizing generation module is used to construct GMDH binary quadratic polynomial nodes, perform square parameterization on the coefficients corresponding to the strong influence features to generate non-negative coefficients to form basic exogenous functions, and introduce weak influence features in the order of interpretation and combination to generate low-order interactive exogenous functions, forming an exogenous function sequence.

[0062] The spatiotemporal prediction and structure update module is used to construct a spatiotemporal SARIMAX structure that incorporates spatial coordination and exogenous functions. It writes spatial lag brightness sequences, exogenous function sequences, and spatial diffusion input sequences. Based on information criteria, it selects exogenous function sequences to form retained exogenous function sequences and writes them into the spatiotemporal SARIMAX structure to form a spatiotemporal prediction model for target brightness.

[0063] The dimming control module is used to output the target brightness reference value according to the target brightness spatiotemporal prediction model and generate dimming commands to be sent to the street light drive device, collect execution feedback and send it back to the edge data processing module.

[0064] In this embodiment, the modules are interconnected using the following method:

[0065] S1. Obtain the street light group brightness feedback time series and multi-source input time series. The multi-source input time series consists of strong influence features and weak influence features. Perform preprocessing on the brightness feedback time series and multi-source input time series to generate a sample window set.

[0066] S2. Generate an adjacency matrix based on the street light grouping topology, calculate the spatially delayed luminance sequence based on the luminance feedback time series based on the adjacency matrix, and calculate the spatially diffused input sequence based on the multi-source input time series based on the adjacency matrix.

[0067] S3. Construct the first layer of GMDH binary quadratic polynomial nodes. The coefficients in the polynomial nodes that are related to the strong influence features are generated into non-negative coefficients through square parameterization. Output the basic exogenous function sequence.

[0068] S4. Take the basic exogenous function sequence as input and introduce weak influence features in the order of interpretation followed by combination to generate low-order interactive polynomial nodes, and output the exogenous function sequence.

[0069] S5. Construct a spatiotemporal SARIMAX structure that incorporates spatial coordination and exogenous functions, and write the spatial lag brightness sequence, exogenous function sequence and spatial diffusion input sequence into the spatiotemporal SARIMAX structure to form a spatiotemporal prediction model for target brightness.

[0070] S6. Perform parameter estimation on the spatiotemporal SARIMAX structure based on the sample window set, and select the exogenous function sequence according to the information criterion to form a retained exogenous function sequence and write it into the spatiotemporal SARIMAX structure.

[0071] S7. Based on the spatiotemporal SARIMAX structure and the retained exogenous function sequence, output the target brightness reference value, generate a dimming command based on the target brightness reference value and send it to the street light driver, collect and execute feedback to update the sample window set.

[0072] In this embodiment, S1 specifically includes:

[0073] Inside the street light centralized control box, according to the preset sampling period, brightness feedback data is read from the lighting drive interface corresponding to each street light group, and a street light group brightness feedback time sequence is formed in chronological order; multi-source input data is read from the sensing interface associated with the street light group, and a multi-source input time sequence is formed in chronological order. The multi-source input time sequence is divided into strong influence feature sequence and weak influence feature sequence according to the degree of feature influence. The strong influence feature sequence includes ambient light data and time stamp data, and the weak influence sequence includes traffic feature data, visibility feature data and equipment status data.

[0074] The street light group brightness feedback time series and the multi-source input time series are subjected to unified time alignment processing, and data with different sampling frequencies are resampled to a consistent time axis; interpolation is performed to complete missing time positions, and anomaly removal is performed on data that exceeds the preset physical range; after completing time alignment and anomaly processing, normalization processing is performed on the street light group brightness feedback time series, strong influence feature series and weak influence feature series respectively; the normalization of the street light brightness feedback time series is complementary normalization, which is positively correlated with the brightness that needs to be supplemented;

[0075] Based on the preset window length and sliding step size, a sliding window truncation operation is performed on the brightness feedback time series, strong influence feature series and weak influence feature series. Within each time window, the brightness feedback subsequence, strong influence feature subsequence and weak influence feature subsequence of the corresponding time period are extracted, and the brightness feedback subsequence and multi-source input subsequence within the same time window are combined to form a sample unit. Several sample units generated in chronological order are collected to form a sample window set.

[0076] In this embodiment, S2 specifically includes:

[0077] Based on the physical arrangement and connection relationships of street light groups in the road, the topology information of the street light groups is obtained. The topology information of the street light groups includes the adjacency relationships between each street light group. An adjacency matrix is ​​constructed according to the number of groups. The matrix elements corresponding to the adjacency relationship are assigned a value of 1, and the matrix elements corresponding to the non-adjacency relationship are assigned a value of 0. The diagonal elements of the matrix are assigned a value of 0.

[0078] The number of elements with a value of 1 in the corresponding row of the adjacency matrix for each street light group is counted to obtain the number of adjacent groups. For each sampling time and each street light group in the sample window set, the brightness feedback values ​​of the adjacent groups at the sampling time are read and summed. The summation result is divided by the number of adjacent groups to obtain the spatial lag brightness value. When the number of adjacent groups is 0, the spatial lag brightness value is set as the brightness feedback value of the street light group at the sampling time. The spatial lag brightness values ​​of each sampling time are arranged in chronological order to obtain the spatial lag brightness sequence.

[0079] For each sampling time, each street light group, and each input feature in the multi-source input time series in the sample window set, the input feature values ​​of adjacent groups at the sampling time are read and summed. The summation result is divided by the number of adjacent groups to obtain the spatial diffusion input value. When the number of adjacent groups is 0, the spatial diffusion input value is set as the input feature value of the street light group at the sampling time. The spatial diffusion input values ​​of each sampling time are arranged in chronological order to obtain the spatial diffusion input sequence.

[0080] In this embodiment, S3 specifically includes:

[0081] Extract strong influence feature sequences from the sample window set. The strong influence feature sequences include ambient lighting data and time-stamped data.

[0082] Construct a set of binary quadratic polynomial nodes in the first layer of GMDH, and select two strongly influential features as input variables for each binary quadratic polynomial node to generate a polynomial expression containing constant terms, linear terms, square terms and cross terms;

[0083] The coefficients of the first-order terms, square terms, and cross terms corresponding to the strong influence features in the polynomial expression are subjected to square parameterization. The corresponding coefficients are expressed in square form and the squared result is used as the coefficient value to form a bivariate quadratic polynomial node with non-negative coefficient constraints. The square parameterization is performed during the initialization and update of polynomial node parameters.

[0084] Two strongly influential features at each sampling time in the sample window set are input into the binary quadratic polynomial node to calculate the node output value sequence; the output value sequences of each binary quadratic polynomial node are aggregated according to the node number to form the basic exogenous function sequence.

[0085] In this embodiment, S4 specifically includes:

[0086] Using the sequence of basic exogenous functions as input, an exogenous function expansion operation is performed according to the node generation order constraint rules. A single basic exogenous function is selected from the basic exogenous function sequence as the main input term, and weakly influential features in the sample window set are introduced as auxiliary input terms to construct a low-order polynomial expression containing only first-order terms and cross terms, generating low-order interactive polynomial nodes. The node generation order constraint rules are as follows: each basic exogenous function in the basic exogenous function sequence is used as an independent main input term, and an exogenous function expansion operation is performed once with each weakly influential feature in the sample window set. The basic exogenous functions do not participate in the same round of node generation as input to each other.

[0087] For low-order interaction polynomial nodes, no squared terms or higher-order cross terms are introduced; only the interaction relationship between the basic exogenous function and the weakly influential features is retained. The values ​​of the basic exogenous function and the weakly influential features corresponding to each sampling time step in the sample window set are input into the low-order interaction polynomial node to calculate the node's output value sequence. The output value sequences of each low-order interaction polynomial node are merged with the basic exogenous function sequence to form an exogenous function sequence. The weakly influential features are all feature dimensions pre-labeled as weakly influential features in the sample window set, and the low-order interaction polynomial nodes are generated by combining the basic exogenous function with each weakly influential feature one by one.

[0088] In this embodiment, the exogenous function self-organization generation process adopts a conditional triggering approach of interpretation followed by combination, specifically including:

[0089] Using the sequence of basic exogenous functions as the output of the interpretation layer, the monotonic direction of the output sequence corresponding to each basic exogenous function is determined in the sample window set according to the time order, and the sequence of changes in the direction of the basic exogenous function between adjacent sampling times is recorded.

[0090] A consistency check is performed on the sequence of changing directions within a preset continuous sampling window. If the changing direction of the basic exogenous function remains consistent within the continuous sampling window, the basic exogenous function is determined to satisfy the structural consistency condition. If the basic exogenous function reverses its direction within the continuous sampling window, the basic exogenous function is determined to not satisfy the structural consistency condition. The continuous sampling window is formed by extracting adjacent sampling moments from the sample window set in chronological order, and the continuous sampling window slides gradually in the sample window set according to the time index.

[0091] Combination operations are performed only on the basic exogenous functions that satisfy the structural consistency condition. The basic exogenous functions are used as the main input terms, and weakly influential features are introduced as auxiliary input terms to construct low-order interactive polynomial nodes containing only first-order terms and cross terms, thereby generating low-order interactive exogenous functions.

[0092] For basic exogenous functions that do not meet the structural consistency condition, it is forbidden to introduce weak influence features and the original basic exogenous function structure remains unchanged.

[0093] The generated low-order interactive exogenous functions are merged with the basic exogenous functions that have not undergone combination operations to form a sequence of exogenous functions.

[0094] In this embodiment, S5 specifically includes:

[0095] The exogenous function output sequence corresponding to each street light group in the sample window set is aligned with the exogenous function output sequence corresponding to the adjacent street light group according to the time order. The time alignment analysis uses the sampling time axis of the current street light group in the sample window set as the reference time axis, and performs alignment processing on the exogenous function output sequences corresponding to the adjacent street light groups according to the reference time axis.

[0096] Time offset determination is performed on the output sequences of the same exogenous function in the current street light group and adjacent street light groups. The time offset determination determines the time lead or time lag relationship of the exogenous function in the current street light group relative to the adjacent street light groups by comparing the consistency order of the output change direction of the exogenous function at adjacent sampling times. The change direction is determined by comparing the magnitude of the output values ​​of the exogenous function at adjacent sampling times. When the output value at the later sampling time is greater than the output value at the earlier sampling time, it is recorded as an upward direction. When the output value at the later sampling time is less than the output value at the earlier sampling time, it is recorded as a downward direction. When the two are equal, it is recorded as a hold direction. When performing consistency order comparison, the hold direction is regarded as consistent with the previous change direction.

[0097] After determining the time lead or time lag relationship, a time remapping operation is performed on the exogenous function output sequence corresponding to the current street light group. The time remapping operation shifts the time index of the exogenous function output sequence forward or backward according to the determined time offset relationship. The exogenous function output sequence of the current street light group is consistent with the exogenous function output sequence of the adjacent street light group in terms of time evolution order. The time remapping operation is a unit step forward or backward shift operation based on the sampling time index. When a time lead relationship is determined, a single step forward shift of the time index is performed. When a time lag relationship is determined, a single step backward shift of the time index is performed.

[0098] After completing the time remapping operation, the time-aligned exogenous function sequence is written into the spatiotemporal SARIMAX structure, and the street light group brightness feedback sequence is written into the spatiotemporal SARIMAX structure at the same time. Joint modeling is performed on the time-aligned exogenous function sequence and the brightness feedback sequence to form a spatiotemporal prediction model of the target brightness.

[0099] In this embodiment, S6 specifically includes:

[0100] Parameter estimation is performed on the spatiotemporal SARIMAX structure with the exogenous function sequence written on the sample window set. Parameter estimation refers to the maximum likelihood estimation of the autoregressive parameters, moving average parameters and corresponding parameters of the exogenous function in the spatiotemporal SARIMAX structure based on the brightness feedback sequence and exogenous function sequence in the sample window set, given the exogenous function structure, to obtain the corresponding model parameter values.

[0101] During parameter estimation, the structure of the exogenous functions involved in the modeling is selected based on the information criterion.

[0102] Based on the comparison results of the information criteria, a set of exogenous functions to be retained is determined, and this set is written back into the spatiotemporal SARIMAX structure as the exogenous function structure. The information criterion used is the Akaike information criterion, which is calculated by combining the log-likelihood value on the sample window set with the number of parameters in the model, and is used to numerically compare models corresponding to different exogenous function structures. During parameter estimation, the corresponding Akaike information criterion value is calculated for different combinations of exogenous functions. By comparing the Akaike information criterion values, the exogenous function combination with the smallest information criterion value is selected as the set of exogenous functions to be retained. The write-back operation refers to writing the set of retained exogenous functions into the spatiotemporal SARIMAX structure as the exogenous function structure of the current model, and removing the unselected exogenous functions.

[0103] Based on the rewritten exogenous function structure, parameter estimation is re-performed to form a spatiotemporal SARIMAX model.

[0104] In this embodiment, S7 specifically includes:

[0105] Based on the spatiotemporal SARIMAX structure with completed parameter estimation and determined exogenous function structure, the target brightness reference value for the next control cycle is calculated by inputting the street light group brightness feedback sequence and the retained exogenous function sequence corresponding to the current sampling time; a corresponding dimming command is generated according to the target brightness reference value and sent to the street light drive interface to perform brightness adjustment; after the dimming command is executed, the corresponding brightness execution feedback data is collected from the street light drive interface and written into the sample window set in chronological order.

[0106] Example 1: To verify the feasibility of this invention in practice, it was applied to a road lighting operation scenario consisting of multiple groups of streetlights. In this scenario, the streetlights are divided into several groups according to the road direction. These groups are physically interconnected, and the lighting status of one group affects adjacent groups. Each group continuously collects its own brightness feedback data and simultaneously collects multi-source input data related to lighting adjustment, including ambient light change data, time stamp data, traffic change data, visibility change data, and equipment operating status data. Existing control methods mainly generate dimming commands based on current brightness feedback. Under conditions of rapid changes in external conditions or multi-group linkage, problems such as delayed brightness adjustment, excessive dimming frequency, and discontinuous brightness between adjacent groups can easily occur.

[0107] When applying the method of this invention in the above scenario, firstly, the brightness feedback sequences and multi-source input sequences of each street light group are uniformly collected in the centralized street light control box. Time alignment, outlier removal, and normalization are then performed at the edge side to form a continuously updated sample window set. Subsequently, a group topology is constructed based on the physical arrangement relationship between street light groups, and spatially delayed brightness sequences and spatially diffused input sequences are calculated on the sample window set, enabling the brightness modeling process to reflect the state changes of adjacent groups. Based on this, through the exogenous function self-organization generation process, strong and weak influence features are transformed into exogenous function structures, and time alignment processing is performed on the exogenous functions in conjunction with spatial coordination relationships. After the exogenous function generation is completed, a spatiotemporal prediction model is constructed at the edge side, and during parameter estimation, the exogenous function structure is filtered and written back according to information criteria, allowing the model structure to dynamically adjust as the running data changes.

[0108] Under continuous operation conditions, the results of using the method of this invention, the traditional control method, and the method that only introduces time prediction are compared and analyzed. During the comparison, multiple indicators such as brightness deviation, brightness fluctuation, dimming behavior, and model operation characteristics are statistically analyzed to evaluate the differences in stability, consistency, and computational load of different control methods.

[0109] Table 1: Quantitative Comparison of Lighting Control Methods

[0110] Indicator Name traditional methods Time-based prediction method Method of the present invention Improvement range Mean within the statistical period Standard deviation within the statistical period Average brightness deviation 0.41 0.27 0.15 63% 0.15 0.04 Brightness fluctuation range 0.46 0.33 0.18 61% 0.18 0.05 Dimming command count 31 21 9 71% 9 2 Brightness difference between adjacent groups 0.35 0.24 0.12 66% 0.12 0.03 Abnormal dimming ratio 0.23 0.16 0.07 70% 0.07 0.02 Edge computing load index 0.81 0.73 0.64 21% 0.64 0.06

[0111] The quantitative comparison results in Table 1 show that the traditional control method maintains high values ​​for indicators such as average brightness deviation and brightness fluctuation amplitude, indicating that its dimming decision mainly relies on real-time feedback and is difficult to reflect brightness change trends in a timely manner. After introducing only the time prediction method, some indicators decreased, but the brightness difference between adjacent groups and the proportion of abnormal dimming remained at a high level, reflecting that single-time modeling is insufficient to characterize the linkage relationship between multiple groups.

[0112] After adopting the method of this invention, all core indicators showed a downward trend. The reduction in average brightness deviation and brightness fluctuation amplitude indicates that the predicted target brightness reference value can guide brightness changes more smoothly; the reduction in the number of dimming commands reflects that control decisions are more concentrated in the time dimension; the decrease in the brightness difference between adjacent groups indicates that the spatial coordination mechanism plays a constraining role under multi-group conditions. When multiple inputs are involved in modeling, the edge computing load index does not increase with the input scale, indicating that the exogenous function structure write-back mechanism has stability in terms of control model complexity. Overall results show that the method of this invention can achieve a lighting control process that balances stability and computational controllability under continuous operation conditions.

[0113] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A real-time control system for LED smart streetlights based on edge computing, characterized in that, include: The edge data processing module is used to collect street light group brightness feedback time-series data and multi-source input time-series data in the street light centralized control box, perform preprocessing on the brightness feedback time-series data and multi-source input time-series data and generate a sample window set; The spatial collaborative feature construction module is used to generate an adjacency matrix based on the street light grouping topology, calculate the spatial lag brightness sequence for the brightness feedback time series data in the sample window set, and calculate the spatial diffusion input sequence for the multi-source input time series data. The exogenous function self-organizing generation module is used to construct GMDH binary quadratic polynomial nodes, perform square parameterization on the coefficients corresponding to the strong influence features to generate non-negative coefficients to form basic exogenous functions, and introduce weak influence features in the order of interpretation and combination to generate low-order interactive exogenous functions, forming an exogenous function sequence. The spatiotemporal prediction and structure update module is used to construct a spatiotemporal SARIMAX structure that incorporates spatial coordination and exogenous functions. It writes spatial lag brightness sequences, exogenous function sequences, and spatial diffusion input sequences. Based on information criteria, it selects exogenous function sequences to form retained exogenous function sequences and writes them into the spatiotemporal SARIMAX structure to form a spatiotemporal prediction model for target brightness. The dimming control module is used to output the target brightness reference value according to the target brightness spatiotemporal prediction model and generate dimming commands to be sent to the street light drive device, collect execution feedback and send it back to the edge data processing module.

2. The LED smart street light real-time control system based on edge computing according to claim 1, characterized in that, The modules are connected in the following way: S1. Obtain the street light group brightness feedback time series and multi-source input time series. The multi-source input time series consists of strong influence features and weak influence features. Perform preprocessing on the brightness feedback time series and multi-source input time series to generate a sample window set. S2. Generate an adjacency matrix based on the street light grouping topology, calculate the spatially delayed luminance sequence based on the luminance feedback time series based on the adjacency matrix, and calculate the spatially diffused input sequence based on the multi-source input time series based on the adjacency matrix. S3. Construct the first layer of GMDH binary quadratic polynomial nodes. The coefficients in the polynomial nodes that are related to the strong influence features are generated into non-negative coefficients through square parameterization. Output the basic exogenous function sequence. S4. Take the basic exogenous function sequence as input and introduce weak influence features in the order of interpretation followed by combination to generate low-order interactive polynomial nodes, and output the exogenous function sequence. S5. Construct a spatiotemporal SARIMAX structure that incorporates spatial coordination and exogenous functions, and write the spatial lag brightness sequence, exogenous function sequence and spatial diffusion input sequence into the spatiotemporal SARIMAX structure to form a spatiotemporal prediction model for target brightness. S6. Perform parameter estimation on the spatiotemporal SARIMAX structure based on the sample window set, and select the exogenous function sequence according to the information criterion to form a retained exogenous function sequence and write it into the spatiotemporal SARIMAX structure. S7. Based on the spatiotemporal SARIMAX structure and the retained exogenous function sequence, output the target brightness reference value, generate a dimming command based on the target brightness reference value and send it to the street light driver, collect and execute feedback to update the sample window set.

3. The LED smart street light real-time control system based on edge computing according to claim 2, characterized in that, S2 specifically includes: Based on the physical arrangement and connection relationships of street light groups in the road, the topology information of the street light groups is obtained. The topology information of the street light groups includes the adjacency relationships between each street light group. An adjacency matrix is ​​constructed according to the number of groups. The number of elements with a value of 1 in the corresponding row of the adjacency matrix for each street light group is counted to obtain the number of adjacent groups. For each sampling time and each street light group in the sample window set, the brightness feedback values ​​of the adjacent groups at the sampling time are read and summed. The summation result is divided by the number of adjacent groups to obtain the spatial lag brightness value. When the number of adjacent groups is 0, the spatial lag brightness value is set as the brightness feedback value of the street light group at the sampling time. The spatial lag brightness values ​​of each sampling time are arranged in chronological order to obtain the spatial lag brightness sequence. For each sampling time, each street light group, and each input feature in the multi-source input time series in the sample window set, the input feature values ​​of adjacent groups at the sampling time are read and summed. The summation result is divided by the number of adjacent groups to obtain the spatial diffusion input value. When the number of adjacent groups is 0, the spatial diffusion input value is set as the input feature value of the street light group at the sampling time. The spatial diffusion input values ​​of each sampling time are arranged in chronological order to obtain the spatial diffusion input sequence.

4. The LED smart street light real-time control system based on edge computing according to claim 3, characterized in that, S3 specifically includes: Extract strong influence feature sequences from the sample window set. The strong influence feature sequences include ambient lighting data and time-stamped data. Construct a set of binary quadratic polynomial nodes in the first layer of GMDH, and select two strongly influential features as input variables for each binary quadratic polynomial node to generate a polynomial expression containing constant terms, linear terms, square terms and cross terms; Perform quadratic parameterization on the coefficients of the first-order term, the quadratic term, and the cross term corresponding to the strong influence features in the polynomial expression, express the corresponding coefficients in quadratic form, and use the squared result as the coefficient value to form a bivariate quadratic polynomial node with non-negative coefficient constraints. Two strongly influential features at each sampling time in the sample window set are input into the binary quadratic polynomial node to calculate the node output value sequence; the output value sequences of each binary quadratic polynomial node are aggregated according to the node number to form the basic exogenous function sequence.

5. The LED smart street light real-time control system based on edge computing according to claim 4, characterized in that, S4 specifically includes: Using the sequence of basic exogenous functions as input, the exogenous function expansion operation is performed according to the node generation order constraint rule; a single basic exogenous function is selected from the sequence of basic exogenous functions as the main input term, and weakly influential features in the sample window set are introduced as auxiliary input terms to construct a low-order polynomial expression containing only first-order terms and cross terms, and generate low-order interactive polynomial nodes. For low-order interactive polynomial nodes, no squared terms or higher-order cross terms are introduced; only the interaction relationship between the basic exogenous function and the weakly influential feature is retained. The values ​​of the basic exogenous function and the weakly influential feature corresponding to each sampling time in the sample window set are input into the low-order interactive polynomial node to calculate the node output value sequence. The output value sequence of each low-order interactive polynomial node is merged with the basic exogenous function sequence to form the exogenous function sequence.

6. The LED smart street light real-time control system based on edge computing according to claim 5, characterized in that, The self-organizing generation process of the exogenous function adopts a conditional triggering method of interpretation followed by combination, specifically including: Using the sequence of basic exogenous functions as the output of the interpretation layer, the monotonic direction of the output sequence corresponding to each basic exogenous function is determined in the sample window set according to the time order, and the sequence of changes in the direction of the basic exogenous function between adjacent sampling times is recorded. A consistency check is performed on the sequence of changing directions within a preset continuous sampling window. When the changing direction of the basic exogenous function remains consistent within the continuous sampling window, the basic exogenous function is determined to meet the structural consistency condition. When the basic exogenous function reverses its direction within the continuous sampling window, the basic exogenous function is determined not to meet the structural consistency condition. Combination operations are performed only on the basic exogenous functions that satisfy the structural consistency condition. The basic exogenous functions are used as the main input terms, and weakly influential features are introduced as auxiliary input terms to construct low-order interactive polynomial nodes containing only first-order terms and cross terms, thereby generating low-order interactive exogenous functions. For basic exogenous functions that do not meet the structural consistency condition, it is forbidden to introduce weak influence features and the original basic exogenous function structure remains unchanged. The generated low-order interactive exogenous functions are merged with the basic exogenous functions that have not undergone combination operations to form a sequence of exogenous functions.

7. The LED smart street light real-time control system based on edge computing according to claim 6, characterized in that, S5 specifically includes: The exogenous function output sequence corresponding to each street light group in the sample window set is aligned with the exogenous function output sequence corresponding to the adjacent street light group according to the time order. Time offset determination is performed on the output sequence of the same exogenous function in the current street light group and the adjacent street light group. The time offset determination determines the time lead or time lag relationship of the exogenous function in the current street light group relative to the adjacent street light group by comparing the consistency order of the output change direction of the exogenous function at adjacent sampling times. After determining the time lead relationship or time lag relationship, a time remapping operation is performed on the exogenous function output sequence corresponding to the current street light group. The time remapping operation shifts the time index of the exogenous function output sequence forward or backward according to the determined time offset relationship. The exogenous function output sequence of the current street light group is consistent with the exogenous function output sequence of the adjacent street light group in terms of time evolution order. After completing the time remapping operation, the time-aligned exogenous function sequence is written into the spatiotemporal SARIMAX structure, and the street light group brightness feedback sequence is written into the spatiotemporal SARIMAX structure at the same time. Joint modeling is performed on the time-aligned exogenous function sequence and the brightness feedback sequence to form a spatiotemporal prediction model of the target brightness.

8. The LED smart street light real-time control system based on edge computing according to claim 7, characterized in that, S6 specifically includes: Parameter estimation is performed on the spatiotemporal SARIMAX structure of the exogenous function sequence written on the sample window set; during the parameter estimation process, the structure of the exogenous function involved in the modeling is selected based on the information criterion. Based on the comparison results of the information criteria, a set of retained exogenous functions is determined, and the set of retained exogenous functions is written back into the spatiotemporal SARIMAX structure as the exogenous function structure; Based on the rewritten exogenous function structure, parameter estimation is re-performed to form a spatiotemporal SARIMAX model.

Images