Pole and tower illumination optimization control method based on intelligent strategy

Abstract

The invention discloses a tower illumination optimization control method based on an intelligent strategy, and relates to the technical field of intelligent illumination. The method mainly comprises the steps that a sensor array is installed on a tower, and multi-source data are collected in real time through the sensor array; the data is transmitted to an intelligent edge collaboration controller for real-time digital twinning situation. On the basis of a pre-trained health model, equipment state online diagnosis and early fault prediction are realized, and a multi-agent collaborative decision-making method is adopted to generate an abnormal alarm and a pre-control instruction through a dynamic scene and a preset rule. According to the method, an operation intention is understood and recognized through an intelligent scene, lighting strategy simulation optimization is carried out in a digital twin environment by taking a task as a guide and taking energy conservation as a target, and finally, a low-power-consumption lighting control scheme is formed and equipment linkage is triggered by combining collaborative scheduling of adjacent towers, so that intelligent perception, collaborative decision and regulation of tower lighting are realized.

Description

Technical Field

[0001] This invention relates to the field of smart lighting technology, specifically to a method for optimizing control of pole lighting based on intelligent strategies. Background Technology

[0002] Pole lighting is crucial in power, communication, and transportation sectors. It primarily provides functional or warning lighting for power transmission towers, communication towers, streetlights, monitoring poles, and high-mast lights. Currently, pole lighting control relies on timed switches or simple photosensitive sensor controls, lacking the perception and analysis of multi-dimensional information such as the on-site environment, equipment status, and operational tasks. While a flagpole lighting control method and flagpole light (publication number CN121126634A) can accurately distinguish between day and night states under artificial light interference and environmental obstructions, avoiding misjudgments and flickering, and ensuring stable visibility of the flag at night, it suffers from several drawbacks. These include a single lighting strategy, an inability to dynamically adjust to actual scenarios, leading to energy waste or insufficient lighting; a lack of real-time monitoring and health management of the lights and poles themselves, resulting in delayed fault response; and a lack of collaborative scheduling capabilities across poles and multiple devices, making it difficult to handle complex tasks such as mobile inspections and emergency repairs.

[0003] Therefore, there is an urgent need for a tower lighting management method that can achieve multi-source information fusion, intelligent analysis and decision-making, and collaborative optimization control. Summary of the Invention

[0004] The purpose of this invention is to solve the problems mentioned in the background art and to propose a pole lighting optimization control method based on intelligent strategies.

[0005] The objective of this invention can be achieved through the following technical solutions: A method for optimizing control of tower lighting based on intelligent strategies, comprising: Step 1: Install a sensor array on the pole and collect multi-source data in real time through the sensor array. The multi-source data includes lighting status data, pole status data, and environmental event data. Step Two: Transmit multi-source data in real time to the intelligent edge collaborative controller deployed on poles or regional gateways. The intelligent edge collaborative controller analyzes the data, including: S201: Real-time fusion of multi-source data from this pole and adjacent towers to construct a digital situational awareness of the site; S202: Call the digital twin lightweight model existing in the intelligent edge collaborative controller, compare the real-time data with the model's preset health data, obtain the online health diagnosis results and early fault prediction risk level, and inject them into the unified situation file; S203: Based on the fused situational awareness file and diagnostic results, combined with a preset rule base, a multi-agent collaborative decision-making algorithm is used for preliminary analysis to obtain alarm commands and predictive dynamic commands, and the commands are executed; the command generation results are written back to the unified situational awareness file; Step 3: Perform in-depth scene analysis based on alarm commands and predictive commands to generate a collaborative lighting control scheme.

[0006] Furthermore, the specific process of constructing a digital situational awareness on-site is as follows: Clock synchronization is performed on the sensor array on each tower to obtain the corresponding timing data; Perform unified time series analysis on time series data; Map events in environmental event data to a unified timeline; After completing the time unification, the intelligent edge collaborative controller performs spatial reference unification on multi-source data; and establishes a spatial pose model for each tower and its various sensors. The intelligent edge collaborative controller transforms the raw coordinate data from different sensors into the global coordinate system based on the spatial pose model; After the transformation is completed, features are extracted from the multi-source data and uniformly packaged into a feature set.

[0007] Based on spatiotemporal consistency, the system performs association calculations on targets detected by different sensors and outputs the association confidence score. When the association confidence score is greater than a preset threshold, the two targets are considered to be the same entity, and a unique entity identifier is generated. Based on the correlation results, a dynamically updated list of on-site entities is constructed, and the data in the entity-level list is mapped to the digital twin model to build a dynamic situation and obtain a unified structured situation file.

[0008] Furthermore, the specific process of constructing a dynamic situation is as follows: Load the three-dimensional model of the tower body, line structure and topography to form a static base model for situation calculation; inject each entity in the entity list into the digital twin model in real time, perform position mapping and state synchronization for each entity, and perform three-dimensional visualization rendering at the corresponding spatial location. In the digital twin model, physical fields are calculated and visualized in real time, including light field distribution, safety risk field, and equipment stress field; Encode the current digital twin situation into a unified structured situation document.

[0009] Furthermore, the specific process for obtaining online health diagnosis results and early fault prediction risk levels is as follows: The intelligent edge collaborative controller pre-builds and stores lightweight digital twin health models for different types of devices, constructing a health baseline model for the devices in a healthy state. The intelligent edge collaborative controller performs time-series modeling of device health deviation indicators, including: acquiring the health deviation indicator sequence for the past W time windows; analyzing the health deviation indicator sequence for the past W time windows based on a lightweight long short-term memory network prediction model built into the intelligent edge collaborative controller; and outputting predicted deviation indicator values ​​for future times. If the predicted deviation indicator value is greater than or equal to a preset threshold δ2, an early fault warning is triggered. The intelligent edge collaborative controller injects online health diagnostic results and early fault prediction risk levels into a unified situation file and updates the corresponding status fields.

[0010] Furthermore, the specific process for obtaining the health deviation indicator sequence over the past W time windows is as follows: The intelligent edge collaborative controller extracts the real-time status feature vector corresponding to the target device from the list of field entities; The real-time state feature vector is mapped to the digital twin lightweight model to generate the virtual state of the corresponding twin device at time t; the deviation is calculated based on the real-time state feature vector and the health baseline model, and then the device health deviation index is constructed. The calculation is performed using eigenvalue decomposition, and the output is a scalar deviation index in practice. The current health status of the device is determined based on the scalar deviation index and the preset threshold range (δ1, δ2); the current health status includes healthy status, sub-healthy status and abnormal status. Within a continuous time window, the intelligent edge collaborative controller performs time series modeling on the scalar deviation index to obtain the scalar deviation index sequence for the past W time windows.

[0011] Furthermore, the process of obtaining alarm commands and predictive dynamic commands is as follows: The intelligent edge collaborative controller extracts the decision input set I from the updated unified situation file and calls the preset rule base; The intelligent edge collaborative controller internally deploys a multi-agent collaborative decision-making algorithm model, which consists of multiple functional agents. Each agent analyzes the decision input set based on its interested state subspace and outputs a local decision mapping value. The intelligent edge collaborative controller fuses the local decision mapping values ​​output by each agent through collaborative game and consistency constraints to obtain a global decision mapping set. When a situation that meets the abnormal handling rules is detected in the global decision mapping set, an alarm instruction is generated. When there are potential events or scenarios that require early response in the global decision mapping set, the intelligent edge collaborative controller generates predictive dynamic instructions. The intelligent edge collaborative controller outputs the generated alarm commands and predictive dynamic commands to the corresponding intelligent terminals for execution, and writes the command generation results back to the unified situation file.

[0012] Furthermore, the specific process of fusing the local decision mapping values ​​output by each agent is as follows: Construct decision variables for the objective-constrained optimization problem, where the objective function is a weighted sum form; For each control cycle, which is 5 seconds, the intelligent edge collaborative controller uses a sequential quadratic programming algorithm to obtain the globally optimal control command; and constructs a global decision mapping set based on the globally optimal control command.

[0013] Furthermore, the process of generating a collaborative lighting control scheme is as follows: The intelligent edge collaborative controller receives data from various real-time sensors, preset inspection operation plan information, and external alarm information; and encapsulates them into a unified scene input set I. The intelligent edge collaborative controller calls its built-in AI scene understanding model to perform intent analysis on the scene input set and output the core intent type of the current scene; After completing the core intent recognition, the intelligent edge collaborative controller calls the digital twin model built in step two to perform simulation calculations on the lighting adjustment scheme: acquiring element data for calculation, including core intent type, current unified situation file, lamp status and spatial distribution information; A set of candidate lighting strategies is generated through twin simulation; The intelligent edge collaborative controller performs cross-tower collaborative scheduling based on a set of candidate lighting strategies, obtains the spatial distribution and coverage relationship of adjacent tower lights, and generates a collaborative lighting control scheme.

[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention transmits multi-source data in real time to the intelligent edge collaborative controller of the pole or area gateway. Through analysis by the intelligent edge collaborative controller, intelligent sensing and control of pole lighting is realized. By using a multi-source sensor array and digital twin model, a real-time and accurate on-site digital situation is constructed, thereby providing basic data for pole lighting management.

[0015] 2. The intelligent edge collaborative controller of the present invention is based on edge intelligence for device health diagnosis and early fault prediction, thereby assessing the device status in real time and providing early warning of potential faults.

[0016] 3. This invention generates a safe, low-energy, comfortable, and device-healthy lighting strategy based on dynamic scenarios and task intentions through multi-agent collaborative decision-making and multi-objective optimization methods.

[0017] 4. Based on digital twin simulation and cross-tower scheduling, this invention realizes on-demand lighting and seamless optical path functions, ensuring operational safety and quality while minimizing energy consumption. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method steps of the present invention. Detailed Implementation

[0019] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0020] It should be understood that the terms “comprising” and “including” used in this disclosure and claims indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0021] It should also be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this disclosure. As used in this disclosure and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this disclosure and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations.

[0022] Please see Figure 1 As shown, a method for optimizing tower lighting control based on intelligent strategies is implemented using several sensor devices. All sensor devices are independently powered by solar energy and have backups to ensure 24 / 7 data operation. Specifically, it includes the following steps: Step 1: Install a sensor array on the pole and collect multi-source data in real time through the sensor array. The multi-source data includes lighting status data, pole status data, and environmental event data.

[0023] The lighting status data includes luminaire current, voltage, temperature, and actual luminous flux output; the tower status data includes tower tilt angle, vibration frequency and amplitude, and temperature and humidity at the corresponding tower location; the environmental event data includes images acquired by high-definition cameras, sound waves acquired by radar, and ambient sounds acquired by microphones.

[0024] It should be noted that the multi-source sensing sensor array includes lighting status sensors, tower status sensors, and environmental event sensors. The lighting status sensors are installed within the drivers of each lamp, used to collect real-time current, voltage, temperature, and luminous flux output values, and report them via Modbus or DALI protocols. The tower status sensors include tilt sensors, triaxial vibration sensors, and temperature and humidity sensors installed at corresponding locations on the tower. Specifically, the tilt sensor has a range of ±15° and an accuracy of 0.1°; the triaxial vibration sensor has a sampling frequency of 100Hz and a range of ±8g; and the tower status sensors are networked via an RS485 bus to upload tower status data. The environmental event sensors include a high-definition starlight-level camera, millimeter-wave radar, and a high-sensitivity microphone, transmitting data via Ethernet or a wireless link.

[0025] Step Two: Transmit multi-source data in real time to the intelligent edge collaborative controller deployed on poles or regional gateways. The intelligent edge collaborative controller analyzes the data, including the following steps: S201: Real-time fusion of multi-source data from this pole and adjacent towers to construct a digital situational awareness of the site, specifically: For each sensor array on each tower, a network time synchronization module is set up, and the clock is synchronized using a network time protocol or a precision time protocol, so that each piece of collected data is assigned a unique timestamp T. i Thus, the corresponding time series data is obtained; For time series data, a unified time series X(t) = {x1(t), x2(t), ..., x...} n (t)}, where x n (t) represents the sampled value of the nth sensor at time t, specifically: Time series alignment is achieved through interpolation, including: for sensors with different sampling frequencies, such as a vibration sensor with a sampling frequency of 100Hz and a temperature sensor with a sampling frequency of 1Hz, time alignment is performed within the intelligent edge collaborative controller. Specifically, this involves obtaining the original discrete sequence X of the nth sensor. n raw ={(t n，1 x n，1 ), (t n，2 x n，2 ...}; Select a uniform timestamp sequence T sync ={t1, t2...t} k}, using the highest sampling frequency as the reference. For each t∈T sync If t matches the original sampling point, then the value is taken directly; otherwise, linear interpolation is used for calculation. Output the resampled data value of the nth sensor at a unified time point t. ;tn，j t represents the j-th sampling time point of the n-th sensor in the original discrete-time series; n，j+1 For adjacent to t n，j The next sampling time point; x n，j+1 The nth sensor at time t n，j The sampled value at x; n，j+1 For the nth sensor at time t n，j+1 The sampled values ​​at each location; i and j are the indexes of the original discrete sampling points; after aligning all sensor data, the unified time series X(t) = {x1(t), x2(t)...x...} is obtained. n (t)}.

[0026] For environmental event data, such as captured images and alarm signals, record the time E of the event occurrence. j = t j type j , source j And map the events to a unified timeline; where t j The timestamp of the event; type j For event type identification; source j This serves as an identifier for the source of the event.

[0027] After completing the time unification, the intelligent edge collaborative controller performs a spatial reference unification operation on multi-source data, specifically including: using the Geographic Information System (GIS) coordinate system as the global spatial reference coordinate system: G={X, Y, Z}.

[0028] Spatial pose models of each tower and its various sensors are established using laser mapping or RTK technology: P s ={x s y s , z s θ s , s , ψ s}, where x s y s , z s Let θ be the spatial position of the sensor. s , s , ψ s Its attitude angles include pitch, roll, and yaw.

[0029] The intelligent edge collaborative controller transforms the raw coordinate data of different sensors into the global coordinate system based on the spatial pose model.

[0030] Taking camera target localization as an example: Obtain the known camera intrinsic parameter matrix K and extrinsic parameter matrix [R|T], and obtain them through calibration. For the target pixel coordinates (u, v, 1) detected in the image... T Assuming the target is located on the ground plane Z=0; then its world coordinates (X... w Y w ,0) T Calculated using the following formula: First, project back onto the camera coordinate system: (x c y c ,1) T =K 1 (u, v, 1) T .

[0031] Due to Z w =0, based on the transformation relationship between world coordinates and camera coordinates: By expanding and solving the system of linear equations, we can obtain (X) w Y w The target's coordinates p in the global GIS coordinate system. i (t)=(X w Y w ,0).

[0032] After the transformation is complete, features are extracted from the multi-source data, and entity relationships are constructed. Specifically: Parallel feature extraction processing is performed on various types of multi-source data, including: Feature extraction from image data includes: extracting target category C using a lightweight convolutional neural network model. k Bounding box B k and behavioral semantic features A k It should be noted that the lightweight convolutional neural network model can be MobileNetV3-SSD.

[0033] Feature extraction from radar data includes: extracting the target range D. k Radial velocity V k Azimuth α k and reflection intensity R k ; Feature extraction from acoustic data includes: analyzing the voiceprint feature vector S kThe matching degree with abnormal sounds is calculated by preprocessing the acquired raw sound signal, including denoising, framing, and windowing; extracting acoustic low-level feature parameters in each frame, including at least Mel-frequency cepstral coefficients (MFCC), short-time energy, zero-crossing rate, and spectral entropy; then aggregating the low-level feature parameters based on time series statistics and feature splicing to form a voiceprint feature vector characterizing the target acoustic properties. The voiceprint feature vector is input into a pre-constructed abnormal sound feature template library or abnormal acoustic model, and comparison analysis is performed through similarity calculation to obtain the corresponding abnormal sound matching degree.

[0034] Feature extraction is performed on the state-time data, including time-domain and frequency-domain analysis of current and vibration signals to extract the health feature vector H. k The time domain parameters include mean, variance, and peak-to-peak value; the frequency domain parameters include FFT transform and characteristic frequency amplitude. The health feature vector performs time synchronization and standardization processing on the collected current and vibration signals. In the time domain, statistical characteristic parameters of the current and vibration signals within a set time window are calculated, including mean, variance, root mean square, peak-to-peak value, and kurtosis. In the frequency domain, the signal is analyzed using Fast Fourier Transform to extract the dominant frequency component, characteristic frequency amplitude, and frequency band energy distribution parameters. The time-domain and frequency-domain features are then concatenated and normalized to construct a health feature vector characterizing the equipment's operating status and health level.

[0035] The above features are uniformly encapsulated into a feature set F. k ={C k B k V k D k α k R k S k H k}

[0036] Based on spatiotemporal consistency, association calculations are performed on targets detected by different sensors to construct entity association functions. The association confidence score A(i,j) is output through the entity association function; where σ t , σ d These are the time and space correlation threshold parameters, respectively, where γ is the motion weighting factor; Δt ij Let Δd be the time difference between the detection of two targets. ij For spatial distance, Δv ij This is a motion consistency index; when the association confidence L(i,j) is greater than the preset threshold A... th If the target i and j are considered to be the same entity, a unique entity identifier ID will be generated.

[0037] Based on the above association results, a dynamically updated list of on-site entities is constructed: E={ID} k Type k P k (t), V k (t), State k Source k}, where ID k Type is the unique identifier for the k-th entity. k The type label for the k-th entity includes, but is not limited to, people, vehicles, drones, equipment, or unknown targets; P k (t) is the spatial position vector of the k-th entity at time t; V k (t) is the velocity vector of the k-th entity at time t; State k Provides current state description information for the k-th entity, including operational status, risk status, or health status; Source k This is the source information for the k-th entity, used to record the entity's perceived source or data fusion source.

[0038] Mapping data within an entity-level list to a digital twin model to build a dynamic situation, including: A three-dimensional model of the tower body, line structure, and terrain is loaded to form a static base model M for situational assessment. s Each entity in the entity list is injected into the digital twin model in real time, and position mapping and state synchronization are performed on each entity, and 3D visualization rendering is performed at the corresponding spatial location.

[0039] Building upon this, physical fields are calculated and visualized in real time within the digital twin model, including: Through the light field distribution model The output illumination intensity distribution is L(x,y), where L(x,y) represents the illumination intensity at the spatial coordinate point (x,y); I i1 g represents the brightness parameter of the i1th luminaire; i1 This is the light attenuation function for the corresponding luminaire, used to describe the attenuation characteristics of light as distance, angle and shading conditions change; Safety risk field: For personnel, equipment, or mobile vehicles within the target space, obtain their spatial coordinates P in the digital twin model in real time. i1 =(x i1 ,y i1 and the corresponding velocity vector V i1 =(v xi1 ,v yi1 ).

[0040] Based on the spatial distance, relative speed, and motion trend between entities, a risk impact function is constructed. When the distance between entities is less than a preset safety distance threshold, the risk weight of the corresponding spatial area is increased. By spatially superimposing and normalizing the risk impacts of multiple entities, a safety risk probability distribution field R(x,y) covering the target space is generated; R(x,y) represents the comprehensive risk probability of safety events such as collisions, intrusions, or operational conflicts occurring at spatial coordinate points.

[0041] Equipment stress field: The distribution of vibration energy and the trend of tilt change are analyzed. Abnormal vibration enhancement, high-frequency vibration or continuous tilt change are mapped into structural stress increment factors. The stress increment factors are then mapped into the digital twin geometric model of the equipment or structure. The stress level of each structural unit is analyzed to form the overall stress distribution field S(x,y) of the equipment. S(x,y) represents the relative stress level of the equipment or structure at the corresponding position and is used to characterize the stress concentration area.

[0042] Encode the current digital twin potential into a unified structured potential file S= T, P, O, E Where T is the time when the state occurs; P represents the three-dimensional spatial coordinates where the state occurred; O represents the corresponding object being identified and tracked in the scene; and E represents the list of entities in the scene.

[0043] S202: Invoke the built-in lightweight digital twin model of the intelligent edge collaborative controller to compare real-time data with preset health data, specifically: The intelligent edge collaborative controller pre-builds and stores lightweight digital twin health models M for different types of devices. h Different types of equipment, such as lighting fixtures and sensors, are included. This lightweight digital twin health model is generated based on historical normal operation data of the equipment, which includes at least one full year. Through statistical learning methods, a baseline feature space for the equipment in its healthy state is constructed, resulting in the health baseline model B. h ={μ f , σ f Ω f}, where μ f Let σ be the mean vector of each feature dimension under the health state. f Ω represents the characteristic fluctuation range. f Let be the covariance matrix among features, representing the constraints between feature associations.

[0044] Digital Twin Lightweight Health Model M hBecause different types of devices have different working mechanisms, operating parameters, and failure modes, the intelligent edge collaborative controller builds and stores corresponding digital twin lightweight health models M for different device types such as lighting fixtures and sensors. h .

[0045] The lightweight health model of digital twin is not a complete high-precision simulation model, but a lightweight health representation model formed by screening, compressing and abstracting key state features during equipment operation, in order to adapt to application scenarios with limited computing power on the edge and high real-time requirements.

[0046] The digital twin lightweight health model M h It includes the following: a set of key operating characteristics of the equipment in a healthy state; the statistical distribution characteristics of each operating characteristic in a healthy state; the inherent correlation constraints between different operating characteristics; and a health reference benchmark for comparative analysis with real-time operating data.

[0047] It should be understood that the digital twin lightweight health model M h It can digitally map and determine the health status of equipment without relying on complex physical simulation calculations.

[0048] Obtain the health baseline model B h The process is as follows: The intelligent edge collaborative controller acquires the operational dataset generated during the device's historical normal operation. This historical data covers at least one full year to include the device's normal operating status under different seasons, loads, and environmental conditions, avoiding distortion of the health baseline due to short-term sample bias. After historical data acquisition, the device operational data is cleaned, denoised, and normalized, and multidimensional feature data reflecting the device's operating status is extracted. These multidimensional features include, but are not limited to, power characteristics, operating duration characteristics, response latency characteristics, environmental adaptability characteristics, and stability characteristics. Based on the processed multidimensional feature data, statistical learning methods are used to model the feature distribution of the device in a healthy state, constructing the device's health baseline model B. h The health baseline model describes three types of information: the typical value level of each operational feature under healthy conditions; the range of variation of each operational feature within the normal fluctuation range; and the correlation between each operational feature, which is used to constrain the rationality of feature combinations and avoid misjudgment caused by isolated judgment of a single feature.

[0049] The intelligent edge collaborative controller extracts the real-time state feature vector corresponding to the target device (such as a lamp) from the list of field entities E: H(t) = {h1(t), h2(t)...h m (t)}, where the real-time state feature vector includes, but is not limited to, vibration features, current features, temperature rise features, structural attitude change features, etc.; The real-time state feature vector is mapped to the lightweight digital twin model to generate the virtual state of the corresponding twin device at time t. : Through formula We obtain Φ(·) as the state synchronization function between the feature map and Φ.

[0050] The deviation between the real-time status feature vector and the health baseline model is calculated to construct an equipment health deviation index. Specifically through formula get; The covariance matrix is ​​calculated using eigenvalue decomposition, Ω f =UΛU Т Where U is the real-time state feature vector matrix, and Λ is the diagonal eigenvalue matrix; through Output the scalar deviation index D(t) in practice; it should be noted that the larger the value of the scalar deviation index D(t), the further the health status deviates from the normal baseline. In this application, T is represented as a matrix transpose.

[0051] The scalar deviation index D(t) is compared with the preset threshold interval (δ1, δ2) to determine the current health status level of the equipment, including healthy, sub-healthy, and abnormal. Specifically: If D(t) < δ1, then the equipment is determined to be in a healthy state; If δ1≤D(t)<δ2, then the equipment is determined to be in a sub-healthy state; If D(t)≥δ2, then the equipment is determined to be in an abnormal state.

[0052] It should be noted that the thresholds δ1 and δ2 are set by those skilled in the art based on the importance of the equipment and historical fault data, for example, δ1=1.5 and δ2=3.0.

[0053] Within a continuous time window, the intelligent edge collaborative controller performs time-series modeling of the device's scalar deviation indicators to predict early faults, including: Obtain the scalar deviation index sequence [D(t) over the past W time windows] W+1), D(t) [W+2)...D(t)]; A lightweight long short-term memory network prediction model is built into the intelligent edge collaborative controller. The structure of the lightweight long short-term memory network prediction model is: 1 LSTM layer with 32 hidden units H, connected to a fully connected layer to output the prediction value.

[0054] LSTM layer computation, for a time step k1: , Where σ is the sigmoid function, It is Hadamaji, f k1 Let C be a vector between 0 and 1, used to determine the cell state C at the previous time step. k-1 How much information is retained; k1 is the k1th discrete time step in the time series, used to identify the current moment in the health deviation index sequence; i k1 Let tildeC be a vector between 0 and 1, which determines the current candidate information. k1 How much is stored in the cellular state; tildeC k1 To determine the input based on the current input D(k1) and the previous state h k1-1 New information that is generated and remembered; C k1 As the long-term memory of LSTM, it carries the core information accumulated from the beginning of the sequence; k1 Let C be a vector between 0 and 1, which determines the current cell state. k1 How much information is output to the hidden state h? k1 W f W i W c and W o Their respective weight matrices; b f b i b c and b o These are the bias vectors for each corresponding gate; [h k1-1 [D(k1)] is the vector concatenation operation. The previous hidden state h is concatenated... k1-1 The input D(k1) is concatenated with the current input D(k1) to form a long vector.

[0055] Finally, a lightweight long short-term memory network prediction model is used to output the predicted value of the deviation index after a future time Δt. .like If this occurs, an early fault warning will be triggered; and based on the deviation from the predicted index value... The predicted risk level is determined by analyzing the numerical values, which can be categorized as low, medium, or high.

[0056] The intelligent edge collaborative controller injects the device's current health status level and early fault prediction risk level into the unified situation file S, updating the corresponding status fields: State←State∪{HealthLevel, FaultRisk, PredictionTime}. Where HealthLevel is the device's current health status level; FaultRisk is the early fault prediction risk level; and PredictionTime is the time window for predicting the occurrence of a fault. S203: Based on the unified situation report and the current health status level of the equipment, combined with a preset rule base, a preliminary analysis is performed using a multi-agent collaborative decision-making algorithm, specifically: The intelligent edge collaborative controller extracts the decision input set I={S, HealthLevel, FaultRisk, PredictionTime} from the updated unified situation file and calls the preset rule base R. The rule base includes at least: device anomaly handling rules, scene response rules, linkage control constraint rules, and energy consumption safety priority rules.

[0057] Rules are defined by mapping conditions to actions: r k Condition k →Actionkr k Among them, r k For the k-th rule in rule base R; Condition k For example, the trigger condition for the k-th rule: HealthLevel = Abnormal and FaultRisk = High; Actionkr k For when Condition k When the conditions are met, the action to be performed should be, for example, generating the highest priority alarm.

[0058] The intelligent edge collaborative controller internally deploys a multi-agent collaborative decision-making algorithm model M. MA It consists of multiple functional intelligent agents, including but not limited to: an equipment health assessment agent, a scene recognition agent, a lighting control agent, and a linkage coordination agent. Each agent analyzes the decision input set based on its interested state subspace and outputs a local decision mapping value D. i2 :D i2 =f i2 (I, R).

[0059] The intelligent edge collaborative controller fuses the local decision mapping values ​​output by each agent through collaborative game theory and consistency constraints, specifically as follows: The global decision fusion is formalized as a multi-objective constrained optimization problem; where the decision variables are: u = [u1, u2, ..., u] L ] Т And represents the set of actions of all controllable devices; the objective function of the problem is in the form of a weighted sum: ; where J health For health goals, such as minimizing health deviations, J safety For safety objectives, such as maximizing illuminance in high-risk areas, J energy For energy consumption target, J is the total power consumption. comfort For comfort targets, such as anti-glare index; λ h , λs , λ e and λ c All are dynamic weights, which are dynamically adjusted according to the intent of the current scenario and the alarm level. For example, in a security intrusion scenario, λ s The weight is the highest. The intent of the current scenario includes: inspection, emergency repair, etc. Controllable devices include lighting brightness, angle, etc.

[0060] Constraints include physical limitations of equipment, linkage logic constraints, and resource contention constraints. Linkage logic constraints include, but are not limited to, brightness gradient limits between adjacent lamps; resource contention constraints include, but are not limited to, total current not exceeding the line limit.

[0061] For each control cycle, which is 5 seconds, the intelligent edge collaborative controller uses a sequential quadratic programming algorithm to obtain the globally optimal control command. Based on globally optimal control commands And construct a global decision mapping set: D global =G(D1, D2…D n ).

[0062] When a situation that meets the anomaly handling rules is detected in the global decision mapping set, an anomaly data alarm instruction is generated. The alarm instruction includes {ID, anomaly type, severity, location, timestamp}. Among them, the anomaly type includes equipment failure, abnormal operating status, or security risk anomaly. The alarm instruction is used to trigger operation and maintenance early warning or security handling process.

[0063] When there are potential events or scenarios requiring advance response in the global decision mapping set, the intelligent edge collaborative controller generates predictive dynamic instructions. These predictive dynamic instructions are used to build a response strategy framework in advance, which includes: {target area, device list, strategy template, trigger condition}. The target area indicates the lighting area that needs to be adjusted, the device list indicates the relevant devices that need to be linked, and the strategy template describes the preliminary control strategy to be executed, such as gradual brightening of the area.

[0064] It should be noted that the potential events refer to events identified based on predictive analysis results that may occur within a preset time range in the future, but have not yet actually occurred and have not met the conditions for triggering an abnormal alarm.

[0065] The intelligent edge collaborative controller outputs the generated alarm commands and predictive dynamic commands to the corresponding intelligent terminals for execution, and writes the command generation results back to the unified situation file to update the state: State←State∪{alarm commands, predictive commands}.

[0066] Step 3: Conduct in-depth scenario analysis based on alarm commands and predictive commands: The intelligent edge collaborative controller receives data from various real-time sensors, preset inspection operation plan information, and external alarm information. This information is uniformly encapsulated into a scene input set: Iscene={S2, Plan, Alarm}; where S2 is the unified situation file generated and updated in step two, Plan is the inspection operation plan information, and Alarm is the external alarm information.

[0067] The intelligent edge collaborative controller invokes its built-in AI scene understanding model to perform intent analysis on the scene input set, and outputs the core intent type Intent of the current scene, expressed by the formula Intent=F. scene (I scene ) to obtain, where F scene (·) is the scene understanding mapping function.

[0068] The core intents include the following types: nighttime drone inspection operation scenarios, emergency repair operation scenarios for power equipment, security perimeter intrusion handling scenarios, and scenarios to ensure protection against severe weather such as strong winds and heavy rain.

[0069] After completing the core intent recognition, the intelligent edge collaborative controller uses the minimum necessary lighting to meet the corresponding core visual task as control, and calls the digital twin model constructed in step two to simulate and calculate the lighting adjustment scheme. The simulation calculation acquires element data for calculation, including the core intent (Intent), the current unified situation file (S2), the luminaire status, and spatial distribution information.

[0070] Generate a set of candidate lighting strategies through twin simulation: L plan ={I i θ i , i CCT i Area i}; where I i θ represents the brightness of the light fixture. i , i Correspondingly, the illumination angle and beam range are represented by CCT. i Indicates the color temperature parameter, Area i Indicates the area covered by lighting.

[0071] In the twin simulation process, to meet the illumination requirements of the task area while minimizing total energy consumption and ensuring illumination uniformity: , Where P i3 Let I be the power consumption function of the i-th lamp, and let I be the power consumption function of the lamp. i3 It has an approximate square relationship, m i3τ is the beam angle index, characterizing the light intensity distribution; τ is the attenuation coefficient; U0 is the illuminance uniformity; E(x, y) is the illuminance at point (x, y); θ i3 (x, y) is the angle between the line connecting luminaire i3 to point (x, y) and the central axis of the beam; r i3 (x, y) is the distance from lamp i3 to point (x, y); P tota l represents the total power consumption; N represents the total number of lamps; E min Minimum illumination requirements, such as 20 lux for inspection scenarios; I i3 This represents the actual brightness of the i3rd lamp. This represents the maximum rated brightness of the i3rd lamp. The average brightness of the lamp; Minimum uniformity requirement; This represents the uniformity of illumination, such as 0.4.

[0072] The intelligent edge collaborative controller performs cross-tower collaborative scheduling based on a set of candidate lighting strategies, obtains the spatial distribution and coverage relationship of adjacent tower lights, and generates a collaborative lighting control scheme. The collaborative lighting control scheme includes: seamless light path lighting for continuously moving targets, uniform light area lighting for fixed or semi-fixed work surfaces, and avoidance of glare and ineffective over-lighting.

[0073] For example, when the intelligent edge collaborative controller identifies the core intent of the drone inspection operation in the current scenario's core intent type and obtains the corresponding drone flight trajectory information, it executes the following control process: Predict the drone's position for the next 30 seconds based on its flight path; Ahead of the flight path, a moving, glare-free, high-definition lighting strip is dynamically generated through multi-tower coordinated lighting to ensure clear camera footage; After the inspection is completed, the relevant lights will be automatically restored to low power consumption or standby mode.

[0074] While generating and issuing lighting control strategies, the intelligent edge collaborative controller automatically triggers multiple linkage commands to form a collaborative response, including: Initiate automatic tracking and enhanced recording of cameras in the event-related area; The control lever-mounted information screen displays safety warning information in graphic and text formats; The equipment fault warning information, current abnormal event snapshots and processing logs diagnosed in step two are automatically packaged and reported to the cloud management platform and the terminals of relevant responsible persons.

[0075] The results of the coordinated execution are synchronously written back to the unified situation file S for subsequent status updates and closed-loop optimization.

[0076] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for optimizing control of tower lighting based on intelligent strategies, characterized in that, include: Step 1: Install a sensor array on the pole and collect multi-source data in real time through the sensor array. The multi-source data includes lighting status data, pole status data, and environmental event data. Step Two: Transmit multi-source data in real time to the intelligent edge collaborative controller deployed on poles or regional gateways. The intelligent edge collaborative controller analyzes the data, including: S201: Real-time fusion of multi-source data from this pole and adjacent towers to construct a digital situational awareness of the site; S202: Call the digital twin lightweight model existing in the intelligent edge collaborative controller, compare the real-time data with the model's preset health data, obtain the online health diagnosis results and early fault prediction risk level, and inject them into the unified situation file; S203: Based on the fused situational awareness file and diagnostic results, combined with a preset rule base, a multi-agent collaborative decision-making algorithm is used for preliminary analysis to obtain alarm commands and predictive dynamic commands, and the commands are executed; the command generation results are written back to the unified situational awareness file; Step 3: Perform in-depth scene analysis based on alarm commands and predictive commands to generate a collaborative lighting control scheme.

2. The pole lighting optimization control method based on intelligent strategy according to claim 1, characterized in that, The specific process of constructing a digital situational awareness on-site is as follows: Clock synchronization is performed on the sensor array on each tower to obtain the corresponding timing data; Perform unified time series analysis on time series data; Map events in environmental event data to a unified timeline; After completing the time unification, the intelligent edge collaborative controller performs spatial reference unification on multi-source data; and establishes a spatial pose model for each tower and its various sensors. The intelligent edge collaborative controller transforms the raw coordinate data from different sensors into the global coordinate system based on the spatial pose model; After the transformation is completed, features are extracted from the multi-source data and uniformly packaged into a feature set; Based on spatiotemporal consistency, the system performs association calculations on targets detected by different sensors and outputs the association confidence score. When the association confidence score is greater than a preset threshold, the two targets are considered to be the same entity, and a unique entity identifier is generated. Based on the correlation results, a dynamically updated list of on-site entities is constructed, and the data in the entity-level list is mapped to the digital twin model to build a dynamic situation and obtain a unified structured situation file.

3. The pole lighting optimization control method based on intelligent strategy according to claim 2, characterized in that, The specific process of constructing a dynamic situation is as follows: Load the three-dimensional model of the tower body, line structure and topography to form a static base model for situation calculation; inject each entity in the entity list into the digital twin model in real time, perform position mapping and state synchronization for each entity, and perform three-dimensional visualization rendering at the corresponding spatial location. In the digital twin model, physical fields are calculated and visualized in real time, including light field distribution, safety risk field, and equipment stress field; Encode the current digital twin situation into a unified structured situation document.

4. The pole lighting optimization control method based on intelligent strategy according to claim 1, characterized in that, The specific process for obtaining online health diagnosis results and early fault prediction risk levels is as follows: The intelligent edge collaborative controller pre-builds and stores lightweight digital twin health models for different types of devices to build a health baseline model of the device in a healthy state; The intelligent edge collaborative controller performs time series modeling of device health deviation indicators, including: obtaining the health deviation indicator sequence of the past W time windows, analyzing the health deviation indicator sequence of the past W time windows based on a lightweight long short-term memory network prediction model built into the intelligent edge collaborative controller, outputting the deviation indicator prediction value for future time, and triggering an early fault warning if the deviation indicator prediction value is ≥ a preset threshold δ2. The intelligent edge collaborative controller injects online health diagnostic results and early fault prediction risk levels into a unified situation file and updates the corresponding status fields.

5. The pole lighting optimization control method based on intelligent strategy according to claim 4, characterized in that, The specific process for obtaining the health deviation indicator sequence over the past W time windows is as follows: The intelligent edge collaborative controller extracts the real-time status feature vector corresponding to the target device from the list of field entities; The real-time state feature vector is mapped to the digital twin lightweight model to generate the virtual state of the corresponding twin device at time t; the deviation is calculated based on the real-time state feature vector and the health baseline model, and then the device health deviation index is constructed. The calculation is performed using eigenvalue decomposition, and the output is a scalar deviation index in practice. The current health status of the device is determined based on the scalar deviation index and the preset threshold range (δ1, δ2); Current health status includes healthy status, sub-healthy status, and abnormal status; Within a continuous time window, the intelligent edge collaborative controller performs time series modeling on the scalar deviation index to obtain the scalar deviation index sequence for the past W time windows.

6. The pole lighting optimization control method based on intelligent strategy according to claim 4, characterized in that, The process of obtaining alarm commands and predictive dynamic commands is as follows: The intelligent edge collaborative controller extracts the decision input set I from the updated unified situation file and calls the preset rule base; The intelligent edge collaborative controller internally deploys a multi-agent collaborative decision-making algorithm model, which consists of multiple functional agents. Each agent analyzes the decision input set based on its interested state subspace and outputs a local decision mapping value. The intelligent edge collaborative controller fuses the local decision mapping values ​​output by each agent through collaborative game and consistency constraints to obtain a global decision mapping set. When a situation that meets the abnormal handling rules is detected in the global decision mapping set, an alarm instruction is generated. When there are potential events or scenarios that require early response in the global decision mapping set, the intelligent edge collaborative controller generates predictive dynamic instructions. The intelligent edge collaborative controller outputs the generated alarm commands and predictive dynamic commands to the corresponding intelligent terminals for execution, and writes the command generation results back to the unified situation file.

7. The pole lighting optimization control method based on intelligent strategy according to claim 6, characterized in that, The specific process of fusing the local decision mapping values ​​output by each agent is as follows: Construct decision variables for the objective-constrained optimization problem, where the objective function is a weighted sum form; For each control cycle, which is 5 seconds, the intelligent edge collaborative controller uses a sequential quadratic programming algorithm to obtain the globally optimal control command; and constructs a global decision mapping set based on the globally optimal control command.

8. The pole lighting optimization control method based on intelligent strategy according to claim 1, characterized in that, The process of generating a collaborative lighting control scheme is as follows: The intelligent edge collaborative controller receives data from various real-time sensors, preset inspection operation plan information, and external alarm information; and encapsulates them into a unified scene input set I. The intelligent edge collaborative controller calls its built-in AI scene understanding model to perform intent analysis on the scene input set and output the core intent type of the current scene; After completing the core intent recognition, the intelligent edge collaborative controller calls the digital twin model built in step two to perform simulation calculations on the lighting adjustment scheme: acquiring element data for calculation, including core intent type, current unified situation file, lamp status and spatial distribution information; A set of candidate lighting strategies is generated through twin simulation; The intelligent edge collaborative controller performs cross-tower collaborative scheduling based on a set of candidate lighting strategies, obtains the spatial distribution and coverage relationship of adjacent tower lights, and generates a collaborative lighting control scheme.

Images