A highway along illumination uniformity section compensation control system

Abstract

The application discloses a highway along illumination uniformity segmented compensation control system and relates to the technical field of lighting. The data acquisition module collects road condition data of a target lighting section; the estimation module combines a control illumination mapping matrix and an illumination state transition matrix, and calculates an illumination estimation vector and a measurement noise diagonal covariance matrix by using an adaptive Kalman filtering algorithm; the optimization module determines a prediction time domain according to vehicle coordinate data, constructs a target illumination distribution, constructs an interference feedback matrix based on the measurement noise diagonal covariance matrix, constructs a control cost function containing the target illumination distribution, the interference feedback matrix and a penalty weight matrix in a virtual feature space according to the illumination estimation vector, and solves the function to obtain an optimal control increment sequence to generate a control instruction; and the execution module adjusts according to the instruction. The application realizes high-precision smooth compensation of road illumination and coordinated adjustment of control variables.

Description

Technical Field

[0001] This invention relates to the field of lighting technology, and in particular to a segmented compensation control system for illuminance uniformity along a highway. Background Technology

[0002] Highway lighting systems not only need to provide drivers with a clear and uniform field of vision within their line of sight to ensure driving safety, but also need to dynamically reduce energy consumption when there are no vehicles or the traffic flow is low in order to achieve the goal of green and low-carbon operation. In order to achieve automatic adjustment of highway street lights, the existing technology CN118921808A discloses a highway lighting adjustment method, system, terminal and storage medium. This scheme divides the highway into several detection areas, collects the ambient light intensity, LED light brightness, traffic flow and visibility of each area, uses the analytic hierarchy process to determine the weight coefficient of each environmental factor, and calculates the target brightness required for each detection area according to a static lighting adjustment formula. Finally, the LED lights in the area are adjusted to the target brightness. The aforementioned existing technologies typically divide the highway into independent discrete regions and assign static brightness values ​​without considering the spatial crosstalk effect caused by the physical overlap of adjacent streetlight spots on the road surface. This makes it difficult to achieve a smooth transition in the illuminance distribution across the entire road section, and easily leads to uneven illuminance at the junctions of different road sections. The adjustment mechanism of the existing technology is based solely on the collected data at the current moment for formulaic static calculations, lacking the time-domain planning for predicting the future continuous trajectory of high-speed vehicles. This adjustment mechanism makes it easy for light compensation to lag behind the actual moving speed of vehicles, creating dangerous visual dark areas. Summary of the Invention

[0003] The technical problem solved by this invention is that existing technologies are difficult to achieve high-precision, low-delay smooth compensation for the illuminance distribution along highways under complex dynamic conditions such as spatial crosstalk between adjacent light spots, physical response delay, and strong light interference from oncoming vehicle lights. Furthermore, it is difficult to simultaneously take into account the optimal coordination and adjustment of vehicle dynamic tracking and multi-dimensional control variables.

[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a segmented compensation control system for illumination uniformity along a highway, comprising: The data acquisition module is used to collect road condition data for the target lighting section; The estimation module is used to calculate the illuminance estimation vector and the measurement noise diagonal covariance matrix based on the illuminance data in the road condition data, combined with the preset control illuminance mapping matrix and illuminance state transition matrix, using the adaptive Kalman filter algorithm. The optimization module is used to determine the prediction time domain and construct the target illuminance distribution based on vehicle coordinate data, construct the disturbance feedback matrix based on the measurement noise diagonal covariance matrix, and construct a control cost function containing the target illuminance distribution, the disturbance feedback matrix and the preset penalty weight matrix in the virtual feature space based on the illuminance estimation vector. The control cost function is solved to obtain the optimal control increment sequence and generate the control command for the current control cycle. The execution module is used to adjust according to control commands.

[0005] Preferably, the road condition data includes illuminance data, vehicle coordinate data, vehicle speed data, and geometric data; Illuminance data is collected through an illuminance sensor; The geometric data includes the three-dimensional spatial coordinates of all street light nodes and the two-dimensional ground coordinates of all illuminance sensors.

[0006] Preferably, the process of constructing the control illuminance mapping matrix specifically includes: Obtain the luminous intensity data of the street light nodes, and calculate the incident angle and azimuth angle of the light reaching the illuminance sensor based on the three-dimensional spatial coordinates of the street light nodes and the two-dimensional ground coordinates of the illuminance sensor. Based on the Lambert point light source formula, and combined with the incident angle, azimuth angle, luminaire photometric data and preset current, a mapping equation from street light node control parameters to illuminance is constructed. Using the preset rated operating point as the expansion center, a small perturbation is applied to the control parameters. The partial derivatives of each control parameter and each illuminance sensor reading are calculated using the numerical difference method. The partial derivatives are then combined to form a control illuminance mapping matrix.

[0007] Preferably, the process of constructing the illuminance state transition matrix specifically includes: Set all street light nodes to the preset rated operating point, and sequentially add a step pulse of preset amplitude to the control current of each street light node; Acquire illuminance data at multiple time steps after the superimposed step pulse, and construct an illuminance response dataset; Construct a discrete linear state-space equation, which contains the initial illuminance state transition matrix to be identified; Using the illuminance response dataset, the discrete linear state-space equations are solved by the least squares method with regularization terms to obtain the initial illuminance state transition matrix. The initial illuminance state transition matrix is ​​symmetricized to obtain the illuminance state transition matrix.

[0008] Preferably, the process of calculating the illuminance estimation vector and the measurement noise diagonal covariance matrix specifically includes: Based on the illuminance state transition matrix, the illuminance estimation vector of the previous control cycle, the control illuminance mapping matrix, and historical control commands, calculate the prior illuminance prediction vector for the current control cycle. Calculate the prediction residual vector between the illuminance data collected in the current control cycle and the prior illuminance prediction vector; Based on the historical prediction residual vector within a preset sliding window length, the measurement noise variance estimate of each illuminance sensor is calculated, and the measurement noise diagonal covariance matrix is ​​constructed. Calculate the Kalman gain matrix based on the diagonal covariance matrix of the measurement noise and the covariance matrix of the process noise; Using the Kalman gain matrix, the prior illuminance prediction vector, and the prediction residual vector, the illuminance estimation vector for the current control cycle is calculated, and the estimation error covariance matrix is ​​updated.

[0009] Preferably, the process of determining the prediction time domain specifically includes: Obtain vehicle speed data from road condition data. Based on the vehicle speed data, the preset safe lighting distance, and the control cycle time interval, calculate the theoretical number of time steps required for the vehicle to travel the safe lighting distance as the length of the prediction time domain. The process of constructing a virtual target illuminance distribution specifically includes: Determine the illuminance sensor index corresponding to the front end of the vehicle based on vehicle coordinate data; Calculate the start and end indices of the safe lighting zone based on the illuminance sensor index, the preset safe lighting distance, and the preset total number of sensors; The target illuminance of the illuminance sensor located within the safe lighting zone is set to the first preset illuminance value, and the target illuminance of the illuminance sensor located outside the safe lighting zone is set to the second preset illuminance value, thereby generating a target illuminance distribution.

[0010] Preferably, the main diagonal elements of the interference feedback matrix are calculated based on the reciprocal of the main diagonal elements of the measurement noise diagonal covariance matrix and a preset reference variance constant; The process of constructing the penalty weight matrix specifically includes: The maximum permissible increment of the preset current in a single step is the first permissible increment, and the maximum permissible increment of the yaw angle in a single step is the second permissible increment. The first penalty coefficient is obtained by dividing the preset penalty parameter by the square of the first allowed increment, and the second penalty coefficient is obtained by dividing the penalty parameter by the square of the second allowed increment. The product of the first penalty coefficient and the identity matrix of the preset dimension is used as the first block, and the product of the second penalty coefficient and the identity matrix of the preset dimension is used as the second block. The first block and the second block are arranged diagonally to form a block diagonal matrix, which is used as the penalty weight matrix.

[0011] Preferably, the process of constructing and solving the control cost function includes: The illumination estimation vector is mapped to the virtual feature space through the left eigenvector matrix to obtain the current virtual state vector; Using the decoupled scalar difference equation, starting from the current virtual state vector, and combining the control command of the previous control cycle and the control increment sequence to be solved, the virtual state prediction value is obtained by recursively predicting multiple future control cycles. The virtual state prediction values ​​in the future prediction time domain are expressed as a linear function of the control increment sequence to be solved through the coefficient matrix, and the stacked vector of virtual state prediction values ​​is obtained. The target illuminance distribution is mapped to a virtual feature space to obtain the virtual target illuminance distribution; Subtract the stacked vector of virtual state prediction values ​​from the virtual target sequence formed by the vertical repetition of virtual target illumination distribution, and multiply it by the inverse mapping matrix formed by the block diagonal expansion of the right feature vector matrix to obtain the prediction error vector of physical space.

[0012] Preferably, the process of constructing and solving the control cost function further includes: Construct a control cost function, which is a quadratic form of the block diagonal extension of the prediction error vector and the disturbance feedback matrix in the physical space, and a quadratic form of the block diagonal extension of the control increment sequence and the penalty weight matrix. The control cost function is rearranged into a standard quadratic form with respect to the control increment sequence. Combined with the first and second constraints, the control cost function is solved using the activity set method to obtain the optimal control increment sequence. The component corresponding to the current control cycle in the optimal control increment sequence is extracted as the optimal control increment sequence.

[0013] A segmented compensation control system for illumination uniformity along a highway, wherein the first constraint is that each control increment sequence does not exceed a preset maximum allowable rate of change per step; and the second constraint is that the current and yaw angle of the street light nodes in any control cycle within the prediction time domain are within a preset allowable range.

[0014] The beneficial effects of this invention are as follows: This invention maps the complex coupling relationship of adjacent light spots overlapping in physical space to an orthogonal virtual feature space, transforming it into mutually independent first-order scalar difference equations, thereby reducing the computational complexity of multi-step predictive control. It uses an adaptive Kalman filter algorithm to estimate the diagonal covariance matrix of measurement noise online, and constructs an interference feedback matrix based on this. This allows for adaptive adjustment of feedback weights based on real-time data from various road surface sensors, effectively suppressing transient environmental interference such as direct glare from oncoming headlights. This invention introduces a penalty weight matrix based on the maximum permissible increment of the control quantity in a single step into the control cost function, coordinating the drive current and yaw angle control quantities with differences in magnitude and dimension. Attached Figure Description

[0015] Figure 1 This is a basic flowchart of a segmented compensation control system for illumination uniformity along a highway, provided as an embodiment of the present invention. Detailed Implementation

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0017] Example, refer to Figure 1 A segmented compensation control system for illumination uniformity along a highway is provided, comprising: The data acquisition module is used to collect road condition data for the target lighting section; The estimation module is used to calculate the illuminance estimation vector and the measurement noise diagonal covariance matrix based on the illuminance data in the road condition data, combined with the preset control illuminance mapping matrix and illuminance state transition matrix, using the adaptive Kalman filter algorithm. The optimization module is used to determine the prediction time domain and construct the target illuminance distribution based on vehicle coordinate data, construct the disturbance feedback matrix based on the measurement noise diagonal covariance matrix, and construct a control cost function containing the target illuminance distribution, the disturbance feedback matrix and the preset penalty weight matrix in the virtual feature space based on the illuminance estimation vector. The control cost function is solved to obtain the optimal control increment sequence and generate the control command for the current control cycle. The execution module is used to adjust according to control commands.

[0018] This invention overcomes the computational bottleneck of traditional control systems when processing multidimensional coupled illuminance data by introducing an adaptive Kalman filter algorithm and a virtual feature space optimization mechanism. While ensuring optimal estimation of road illuminance, it constructs a control cost function using an interference feedback matrix and a penalty weight matrix, which can output the optimal control increment sequence with low computational delay. This achieves adaptive suppression of strong ambient light interference and smooth coordination of multidimensional control variables.

[0019] Road condition data includes illuminance data, vehicle coordinate data, vehicle speed data, and geometric data; Illuminance data is collected through an illuminance sensor; The geometric data includes the three-dimensional spatial coordinates of all street light nodes and the two-dimensional ground coordinates of all illuminance sensors.

[0020] Illuminance data was collected by illuminance sensors at a fixed sampling frequency of 100Hz. The illuminance sensors were installed at the bottom of the roadside guardrails, deployed at 5-meter intervals along the longitudinal direction of the road, for a total of 100 sensors, completely covering the target illuminated section and extending 125 meters to each end. The instantaneous illuminance values ​​collected by each sensor were arranged into a 100-row × 1-column column vector to obtain the illuminance data. ; Vehicle coordinate and speed data are collected by a millimeter-wave radar mounted on a roadside pole, which acquires the longitudinal coordinates of the target vehicle relative to the first illumination sensor in real time at a frequency of 100Hz. (Unit: meters, m) serves as vehicle coordinate data; The vehicle coordinate data is stored in a ground-based two-dimensional rectangular coordinate system, with the ground projection point of the location of the first illuminance sensor as its origin. The direction of vehicle travel along the longitudinal direction of the road is the positive X-axis, and the direction perpendicular to the longitudinal direction of the road and pointing to the inside of the lane is the positive Y-axis. The positions of all street light nodes and illuminance sensors are mapped to the ground two-dimensional rectangular coordinate system. The geometric data consists of the three-dimensional spatial coordinates of all 50 streetlights. Ground two-dimensional coordinates of all 100 sensors ,in The vertical height of the light center of the street light node above the ground, in meters (m). The coordinates of the street light nodes mapped to the ground in a two-dimensional Cartesian coordinate system; This invention divides the continuously running time axis into discrete control cycles at fixed intervals. Each control cycle corresponds to a time step, denoted as k, where k = 1, 2, 3, ..., and the physical time interval between adjacent time steps is fixed. This interval is the control cycle of the present invention; In each control cycle Within this process, the online calculation steps of the present invention are completed using the historical control instructions from the previous control cycle, and the optimal control instruction for the current control cycle is output. The historical control commands are the current value and yaw angle of each street light node in the target lighting section within the previous control cycle; The target lighting section in this embodiment is a continuous lighting section of a long straight highway, with 50 LED street light nodes installed at equal intervals along the direction of vehicle travel, and the longitudinal spacing between adjacent street light nodes is 5 meters. Each LED street light node has an independently adjustable drive circuit and an electrically adjustable angle bracket, whose adjustable control parameters include current and yaw angle. All 50 streetlights have 100 independent control parameters, arranged in order of current first, then angle, in each discrete control cycle. splicing together to form a dimension Control command column vector The unit for the first 50 components of the control command column vector is mA, and the unit for the last 50 components is rad.

[0021] The process of constructing the control illuminance mapping matrix specifically includes: Obtain the luminous intensity data of the street light nodes, and calculate the incident angle and azimuth angle of the light reaching the illuminance sensor based on the three-dimensional spatial coordinates of the street light nodes and the two-dimensional ground coordinates of the illuminance sensor. The process of constructing the control illuminance mapping matrix is ​​the offline calibration stage; Photometric data of LED streetlights, specifically IES files (all-space luminous intensity data files), are collected from lighting manufacturers. These IES files are indexed by the zenith and azimuth angles of light emission, providing the luminous intensity value per unit current. (candela / milliampere) In this embodiment, all street light nodes are installed with the optical axis pointing vertically downwards, that is, the fixed vertical pitch angle is 0 radians, and the horizontal yaw angle is adjustable. Three-dimensional spatial coordinates based on streetlights and the ground two-dimensional coordinates of the illuminance sensor Calculate the angle of incidence of the light rays at the ground sensor. and azimuth ; Angle of incidence The mathematical expression is: ; in, For the first The first illuminance sensor to the first The straight-line horizontal distance between the center of each street light and its projection point on the ground; It is the first The installation height of each street light; Azimuth The mathematical expression is: ; in, The relative azimuth angle from the nth street light to the mth sensor is expressed in rad. and These are the two-dimensional ground coordinates of the sensor and the center of the street light, respectively. Let yaw angle be the yaw angle of the nth street light; Based on the Lambert point light source formula, and combined with the incident angle, azimuth angle, luminaire photometric data and preset current, a mapping equation from street light node control parameters to illuminance is constructed. Based on the Lambert point light source formula, the incident angle Azimuth By combining luminaire photometric data and current, a mapping equation from control parameters to illuminance is constructed, that is, the calculation of the first... The streetlight was on the first Illuminance generated at each illuminance sensor ; In this embodiment, all street light nodes are installed with the optical axis pointing vertically downwards, and the slant distance from the light source to the illuminated point is... and In addition to the ratio, this embodiment also introduces the influence of current linearity and yaw angle; Therefore, the first The streetlight was on the first Illuminance generated at each illuminance sensor The mathematical expression is: ; in, Let be the theoretical illuminance value generated by n streetlights at the m-th sensor, in lx. It is the first The installation height of each street light; To retrieve based on the index Luminous intensity per unit current obtained from the IES file via bilinear interpolation; It is the angle of incidence of light reaching the ground sensor. This represents the driving current of the nth street light, in mA. It is the azimuth angle. Angle of incidence Since all street light nodes in this embodiment are installed with the optical axis pointing vertically downwards, the zenith angle at which the light rays exit and the angle of incidence at which the light rays reach the ground are... They are equal, therefore they can be used directly. Used as a retrieval index for the zenith angle in the IES file.

[0022] Using the preset rated operating point as the expansion center, a small perturbation is applied to the control parameters. The partial derivatives of each control parameter and each illuminance sensor reading are calculated using the numerical difference method. The partial derivatives are then combined to form a control illuminance mapping matrix.

[0023] The rated operating point of all streetlights was taken as the center of the deployment. The rated operating point current was 700mA and the yaw angle was 0rad. Control parameters such as current, yaw angle, and illuminance near the rated operating point. It exhibits linear characteristics. Small perturbations are applied to each control variable one by one, and the partial derivatives of each pair of control variables and sensor readings are calculated using forward difference to construct the Jacobian matrix. Specifically, for the first The current of a street lamp Given a small increment Keeping other control variables unchanged, recalculate the illuminance of all illuminance sensors; Subtract the illuminance value at the rated operating point from the recalculated illuminance and then divide by . 100 partial derivatives were obtained. ,in traversal 50 sets of current partial derivatives were obtained; yaw angle Apply small increments Similarly, calculate 100 partial derivatives. ; Adjusting the horizontal yaw angle only causes the luminaire to rotate around its vertical central axis; it does not change the height of the luminaire's optical center. The spatial slant distance between the light source and the sensor, and the angle of incidence of the light source reaching the ground. Therefore, when calculating the change in illuminance caused by the change in yaw angle, the spatial geometric path remains unchanged, and the change in illuminance originates only from the azimuth index. The resulting change in luminous intensity Nonlinear changes; All partial derivatives are combined column-wise to form a 100-row, 100-column real matrix as the control illumination mapping matrix. Control the row number of the illuminance mapping matrix Corresponding to the road surface sensor numbers, the column numbers sequentially correspond to 50 current control quantities and 50 yaw angle control quantities, controlling each element in the illumination mapping matrix B. , indicating when the When a control parameter changes by 1 unit (1mA or 1rad), the first... The absolute change in illuminance value produced by each sensor (unit: lx / mA or lx / rad) in this invention The unit of illuminance is lux. In this embodiment, lx refers to the unit of illuminance.

[0024] In constructing the control illuminance mapping matrix, this invention combines full-space luminous intensity data with spatial geometric angles. By applying a small perturbation at the rated operating point and performing numerical difference calculations, the complex nonlinear spatial optical relationship is quantified into a linear Jacobian matrix, reducing the complexity of online prediction calculations. At the same time, it retains the photometric mapping characteristics with high fidelity, enabling the algorithm to quickly and accurately deduce the absolute physical impact of changes in driving parameters on the illuminance distribution.

[0025] The process of constructing the illuminance state transition matrix specifically includes: Set all street light nodes to the preset rated operating point, and sequentially add a step pulse of preset amplitude to the control current of each street light node; The process of constructing the initial illuminance state transition matrix is ​​the offline calibration stage; The purpose of constructing the initial illuminance state transition matrix is ​​to quantify the time inertia caused by the capacitor charging and discharging hysteresis of the LED driver power supply, the response delay of the sensor analog-to-digital conversion and the signal conditioning circuit, as well as the spatial crosstalk caused by the physical overlap of adjacent light spots on the road surface, into a constant matrix, namely the initial illuminance state transition matrix. Specifically, under the same no-load nighttime conditions, all streetlights were set and stabilized at their rated operating points, i.e., current 700mA and yaw angle 0rad. Starting with the first street light, a positive step pulse with an amplitude of 100mA is superimposed on the control current of each street light node in turn, that is, the current suddenly changes to 800mA, while keeping the current of all other street lights at 700mA and the yaw angle at 0rad. Acquire illuminance data at multiple time steps after the superimposed step pulse, and construct an illuminance response dataset; After each positive step pulse is superimposed, the illuminance is recorded at 100 time steps with a period of 10ms using an illuminance sensor to construct an illuminance state column vector. For the When a positive step pulse is superimposed on a street light, 100 time steps will be recorded. The illuminance data matrix of the sensors is denoted as follows: Illuminance data matrix Each row corresponds to a sampling control cycle, and each column corresponds to an illuminance sensor. After superimposing positive step pulses on all streetlights in sequence, 50 illuminance data matrices are obtained. An illuminance response dataset is constructed using illuminance data matrices collected by individually superimposing positive step pulses on all 50 streetlights. Construct a discrete linear state-space equation, which contains the initial illuminance state transition matrix to be identified; A discrete linear state-space equation is constructed to represent the dynamic change of illuminance under the combined effects of electrical delay and optical spatial overlap. The mathematical expression of the discrete linear state-space equation is as follows: ; in, For the first A column vector of illumination states at each time step. The 100×100 dimensional initial illuminance state transition matrix to be identified is... In the first A column vector of illumination states at each time step. To control the illuminance mapping matrix, This is a step control input; The illuminance state column vector is the first... After a positive step pulse is superimposed on a street light, at the 1st... At each time step, a column vector consisting of illuminance values ​​from 100 illuminance sensors is generated. for The The transpose column vector of each row vector; The step control input is the first The control commands applied to 50 streetlights at each time step, including current and yaw angle, are column vectors with a dimension of 100×1. Using the illuminance response dataset, the discrete linear state-space equations are solved by the least squares method with regularization terms to obtain the initial illuminance state transition matrix. The initial illuminance state transition matrix is ​​solved using the least squares method with ridge regression regularization. The ridge regression regularization parameter is set to 0.001. The initial illuminance state transition matrix is ​​symmetricized to obtain the illuminance state transition matrix.

[0026] Considering that the sensors and streetlights on the long straight highway are symmetrically distributed at equal intervals, the initial illuminance state transition matrix is ​​obtained. Then, it is symmetricized to obtain the illumination state transition matrix. The mathematical expression for the illuminance state transition matrix is: ; in, This is the initial illuminance state transition matrix. This is the transpose of the initial illuminance state transition matrix.

[0027] This invention identifies the initial illuminance state transition matrix by actually superimposing step pulses and combining it with the least squares method with ridge regression regularization. This matrix includes the comprehensive physical effects such as drive electrical hysteresis, signal delay, and spatial crosstalk between adjacent light spots. Further matrix symmetry processing fits the physical reality of the equidistant symmetrical distribution of equipment on long straight highways, providing a reliable mathematical model that fits the actual dynamic time inertia for subsequent Kalman filtering and prediction optimization. The process of calculating the illuminance estimation vector and the measurement noise diagonal covariance matrix specifically includes: Based on the illuminance state transition matrix, the illuminance estimation vector of the previous control cycle, the control illuminance mapping matrix, and historical control commands, calculate the prior illuminance prediction vector for the current control cycle. This module uses an adaptive Kalman filter algorithm to make the optimal estimate of road surface illuminance by using the illuminance data collected in the current control cycle, the control commands executed in the previous control cycle, and the system matrix obtained in the offline stage. At the same time, it obtains the measurement noise diagonal covariance matrix of each sensor channel online. The input data for the adaptive Kalman filter algorithm includes illuminance data. Control instruction column vector This refers to the control commands that were issued and executed in the previous control cycle, with the dimension being... The first 50 components are current, and the last 50 components are yaw angle. Illuminance state transition matrix. , dimension The control illuminance mapping matrix has been acquired in the offline phase. , dimension It has already been obtained in the offline phase; Preset parameters include: initial illuminance estimation The value is Zero-dimensional vector; initial historical control commands The first 50 current components are all set to The last 50 yaw angle components are all set to This initial value is only used for the instantaneous state prediction calculation in the first control cycle; the initial estimate error covariance matrix The value is 1-dimensional identity matrix Process noise covariance matrix The value is a diagonal matrix. ; Sliding window length The value is Regularized variance coefficient The value is Initial measurement noise diagonal covariance matrix The value is , It is a unit for the square of illuminance; After completing the input data and preset parameter settings, in each control cycle Within this process, the following online recursive calculation procedure is executed; Using the illuminance estimation vector from the previous control cycle and the executed historical control commands, the prior prediction value of the illuminance for the current control cycle, i.e., the prior illuminance prediction vector, is calculated. The mathematical expression for the prior illuminance prediction vector is: ; in, Let be the prior illumination prediction vector, with dimension . ; Here is the illumination state transition matrix; This is the illuminance estimation vector for the previous control cycle; To control the illumination mapping matrix; For control instructions that were executed in the previous control cycle, This is a constant bias term vector used to represent fixed deviations such as ambient background illumination, with a dimension of 100×1; The control input is in incremental form, and the prediction process satisfies ; The measurement equation for the adaptive Kalman filter algorithm is: ; Where H is the measurement matrix. Since the sensor directly measures the illuminance at each corresponding node, H is a 100×100 dimensional identity matrix. , To measure the noise vector.

[0028] Calculate the prediction residual vector between the illuminance data collected in the current control cycle and the prior illuminance prediction vector; After obtaining the prior prediction value, the real-time collected illuminance data is subtracted from the prior prediction value. ), to obtain the predicted residual vector Since the measurement matrix H is an identity matrix, the prediction residuals are directly derived from... and Obtain by subtraction; Based on the historical prediction residual vector within a preset sliding window length, the measurement noise variance estimate of each illuminance sensor is calculated, and the measurement noise diagonal covariance matrix is ​​constructed. Using the predicted residual vector Online estimation of the diagonal covariance matrix of the measurement noise is performed. Specifically, for the first... One sensor, of which Extract the sensor's most recent Each predicted residual value The variance is calculated as the estimated measurement noise variance of the sensor channel. The mathematical expression for the estimated measurement noise variance is: ; in, For the first One sensor in Estimated variance of measurement noise during the control cycle; The length of the sliding window; For the first The sensor at the first The predicted residual value of the control period; The regularization variance coefficient is 10 in this embodiment. −4 lx²; For the first The sensor recently The mean of the predicted residuals; Arrange the estimated variance values ​​of the measurement noise from all illuminance sensors diagonally to form the measurement noise diagonal covariance matrix. The mathematical expression for the measurement noise diagonal covariance matrix is: ; in, To measure the noise diagonal covariance matrix, with dimension 1 The first one on its main diagonal The element is the th element. Measurement noise variance estimate of each sensor All elements not on the main diagonal are zero; For the special case where the sliding window is not yet full, the following handling is performed: If The preset initial measurement noise diagonal covariance matrix is ​​used directly. ;like Use existing Each predicted residual value is calculated using the above formula, and the denominator in the variance estimation formula is corrected to... ; Calculate the Kalman gain matrix based on the diagonal covariance matrix of the measurement noise and the covariance matrix of the process noise; After obtaining the diagonal covariance matrix of the measurement noise, the Kalman gain is calculated. Specifically, the prior estimation error covariance matrix is ​​calculated, and its mathematical expression is as follows: ; in, Let be the prior estimation error covariance matrix, with dimension . ; Here is the illumination state transition matrix; This is the posterior estimation error covariance matrix of the previous control cycle; The process noise covariance matrix; Based on the prior estimation error covariance matrix and the measurement noise diagonal covariance matrix, the Kalman gain matrix is ​​calculated. The mathematical expression for the Kalman gain matrix is: ; in, The Kalman gain matrix has a dimension of . ; The prior estimate is the error covariance matrix; To measure the diagonal covariance matrix of the noise; Using the Kalman gain matrix, the prior illuminance prediction vector, and the prediction residual vector, the illuminance estimation vector for the current control cycle is calculated, and the estimation error covariance matrix is ​​updated.

[0029] After obtaining the Kalman gain, the state estimate is updated posteriorly using the prediction residuals. The optimal illuminance estimation vector for the current control cycle is then calculated. The mathematical expression for the optimal illuminance estimation vector is: ; in, This is the illuminance estimation vector for the current control cycle, with dimension . ; This is the prior illumination prediction vector; The Kalman gain matrix; To predict the residual vector.

[0030] Update the posterior estimation error covariance matrix for use in the next control cycle. The mathematical expression for the posterior estimation error covariance matrix is: ; in, The posterior estimation error covariance matrix for the current control cycle; for 3D identity matrix; The Kalman gain matrix; This is the prior estimate of the error covariance matrix.

[0031] The purpose of the optimization module is to find a set of optimal control increment sequences within the current control cycle, so that the road illuminance distribution is as close as possible to the desired target illuminance distribution in a future prediction time domain, i.e., multiple future control cycles. In other words, minimize the error between the target illuminance distribution and the road illuminance distribution, while keeping the adjustment range of the control quantity as small as possible, and adaptively suppressing strong light interference from oncoming vehicle lights during the solution process. The control increment sequence includes changes in current and changes in deflection angle; The optimization objective of this embodiment is to make the illuminance distribution as close as possible to the target illuminance distribution, and to minimize the adjustment range of the control parameters; Since there are 100 control parameters for 50 streetlights, and there is a contradiction between the two optimization objectives, namely, to make the illuminance closer to the target, a larger control adjustment is usually required. In order to achieve the optimal balance between the two conflicting optimization objectives, this embodiment constructs a control cost function, and uses the weighted sum based on the illuminance error and the control increment sequence penalty as the optimized control cost function. The optimal control increment sequence is solved by minimizing this control cost function. To construct the aforementioned control cost function, this embodiment needs to determine the length of the prediction time domain; construct the desired target illuminance distribution; construct adaptive feedback weights to suppress strong light interference; and construct penalty weights for the control increment sequence to coordinate the adjustment ratio of the control parameter current and yaw angle. This invention utilizes historical prediction residuals within a preset sliding window to update the measurement noise variance estimate in real time, constructing a dynamic measurement noise diagonal covariance matrix. This online recursive update mechanism can quantify and capture measurement distortion caused by sudden interference in each sensor channel in real time, thereby adaptively adjusting the confidence level in the calculation of Kalman gain and posterior update, significantly improving the anti-interference robustness of the optimal estimation of road illumination state.

[0032] The process of determining the prediction time domain specifically includes: Obtain vehicle speed data from road condition data. Based on the vehicle speed data, the preset safe lighting distance, and the control cycle time interval, calculate the theoretical number of time steps required for the vehicle to travel the safe lighting distance as the length of the prediction time domain. The length of the prediction time domain is represented by the number of future control cycles. Too few control cycles will cause illumination compensation to lag behind vehicle movement, while too many will increase unnecessary computational burden. The mathematical expression for the length of the prediction time domain is: ; in, To predict the length of the time domain, The preset safe lighting distance is set at 5m in this embodiment, which is the longitudinal spacing of the streetlights; Current vehicle speed The physical time interval between adjacent time steps, i.e., the control period. The maximum predicted time domain length is preset; in this embodiment, the value is set to 25. The minimum predicted time domain length is preset; in this embodiment, the value is 5. The rounding up symbol; When there are no vehicles in the section or When it is 0, The default value is ; In order to determine the safety boundary of the prediction time domain length under extreme conditions (high speed and low speed), ensure that the algorithm's computation time does not exceed the hard real-time constraint of a single-step control cycle, and ensure uniform road illumination, this embodiment conducted online optimization performance tests under different fixed prediction time domain lengths. The test conditions were selected by vehicles passing through the target illuminated road section at different speeds. The average computation time of solving the control cost function and the mean square error of illumination in front of the vehicle were recorded. The experimental data results are shown in Table 1. Table 1

[0033] As shown in Table 1, since the complexity of the quadratic programming problem used by the optimization module of this invention increases non-linearly with the number of prediction steps, when the length of the prediction time domain is greater than 25, the single calculation time of the algorithm will exceed the preset control period (10ms), causing the instruction to be unable to be issued within the current period. When the length of the prediction time domain is less than 25, the illuminance mean square error has basically converged to a minimum value (2.9 lx²), and further increasing the number of steps will have a negligible effect on the improvement of illuminance uniformity. Therefore, in this embodiment, the preset maximum length of the prediction time domain is limited to 25. When the length of the prediction time domain is less than 5, the predicted future time span is less than the physical lag time of LED power supply capacitor charging and discharging and communication network (30-40m), which will cause the lamp to be unable to respond to the movement of the vehicle, resulting in a dangerous visual dark area lag phenomenon. In this embodiment, the preset minimum prediction time domain length is set to 5.

[0034] The process of constructing a virtual target illuminance distribution specifically includes: Determine the illuminance sensor index corresponding to the front end of the vehicle based on vehicle coordinate data; The target illuminance distribution is determined based on the real-time position of the vehicle. In this embodiment, the road surface within the safe line of sight in front of the vehicle needs higher illuminance to ensure driving safety, while the area behind and far from the vehicle only needs to maintain background illuminance to save energy. Specifically, using vehicle coordinate data Determine the illuminance sensor index corresponding to the current position of the front of the vehicle. The mathematical expression for the illuminance sensor index is: ; in, This is the index of the discrete illuminance sensor corresponding to the current position of the front of the vehicle. For vehicle coordinate data, The longitudinal spacing between adjacent illuminance sensors. rice; in, The floor symbol is used for rounding down. Negative numbers or numbers greater than 100 are allowed, indicating that the vehicle has not yet entered or has already left the zone; Calculate the start and end indices of the safe lighting zone based on the illuminance sensor index, the preset safe lighting distance, and the preset total number of sensors; like If the vehicle is within the target lighting section, the safe lighting zone is calculated; otherwise, if there is no vehicle within the target lighting section, the vehicle enters standby lighting mode. The start and end indices of the safety lighting zone are: , where the starting index Terminate index 100 is the maximum number of sensors; The target illuminance of the illuminance sensor located within the safe lighting zone is set to the first preset illuminance value, and the target illuminance of the illuminance sensor located outside the safe lighting zone is set to the second preset illuminance value, thereby generating a target illuminance distribution.

[0035] Set the target illuminance of the illuminance sensor within the safe lighting zone to... , The unit of illuminance is lux. The target illuminance for the illuminance sensor outside the safe lighting zone is set to... The vector formed by the sequential arrangement of the target illuminance values ​​from the illuminance sensors within the target road segment constitutes the target illuminance distribution. ; This module calculates the safe lighting range and sets high illuminance only in the distance in front of the vehicle, while maintaining low illuminance behind and far away from the vehicle, thus ensuring both safety and energy saving. because Negative numbers or numbers greater than 100 are allowed, indicating that the vehicle has not yet entered or has already left the zone, which will result in... or If there are no vehicles in the target road segment, the entire segment will be assigned a default value. Enter standby lighting mode, that is, all street light nodes uniformly execute a current of 100mA and a yaw angle of 0rad; The dimension of the target illumination vector is , is a vector composed of the target illuminance values ​​of 100 illuminance sensors, where the sensor numbers are located at the start and end indices of the safe lighting zone. Within the target illuminance vector, the target illuminance of the illuminance sensor is... The target illuminance of the remaining illuminance sensors is If there are no vehicles in the target road segment, all elements in the target illumination vector are... .

[0036] This invention links the predicted time domain length with the real-time speed of the vehicle. Under the premise of ensuring that the computation time of the optimization algorithm does not exceed the hard real-time constraint of the single-step control cycle, it eliminates the risk of low-speed computational redundancy and high-speed lighting compensation lag causing visual dark areas. By defining safe lighting intervals and assigning differentiated values, it achieves the dynamic segmented lighting goal of high illumination in front of the vehicle to ensure driving safety and low illumination standby in areas behind the vehicle and without vehicles, thus achieving the optimal balance between traffic safety and green energy saving and consumption reduction.

[0037] The main diagonal elements of the interference feedback matrix are calculated based on the reciprocal of the main diagonal elements of the measurement noise diagonal covariance matrix and a preset reference variance constant. When a certain illuminance sensor is directly illuminated by the strong light of an oncoming vehicle headlight, its illuminance will deviate significantly from the true illuminance of the local road section, causing the measurement noise of the sensor to increase sharply. In this embodiment, the sharp increase in sensor measurement noise is reflected in the measurement noise diagonal covariance matrix. In, corresponding to As the diagonal element values ​​increase, this embodiment achieves this by adjusting the values ​​of the diagonal elements. The interference feedback matrix is ​​constructed by inverting the values. The inversion operation makes the diagonal elements with larger values ​​become smaller weights, and the diagonal elements with smaller values ​​become larger weights. The feedback weight of the sensor affected by strong light interference is reduced in the optimization algorithm. The mathematical expression for the interference feedback matrix is: ; in, For adaptive interference feedback weight matrix, for The identity matrix, To measure the diagonal covariance matrix of the noise; The preset baseline variance constant is taken as follows: lx², calculated for A diagonal positive definite matrix of dimension 1; The process of constructing the penalty weight matrix specifically includes: The maximum permissible increment of the preset current in a single step is the first permissible increment, and the maximum permissible increment of the yaw angle in a single step is the second permissible increment. The control command includes both current and yaw angle, which differ greatly in size and magnitude. If a uniform penalty weight is used in the cost function, the penalty for the yaw angle increment is negligible due to its extremely small value, causing the algorithm to tend to adjust the yaw angle frequently. To address this issue, this embodiment presets the maximum permissible increment of the current control parameter in a single step to be the first permissible increment. The maximum permissible increment of yaw angle is the second permissible increment. In this embodiment, the first allowable increment is 50mA and the second allowable increment is 0.001rad, representing the current of each street light, which can only change by ±50mA in a single control cycle, and the yaw angle of each street light, which can only change by ±0.001rad in a single control cycle. The first penalty coefficient is obtained by dividing the preset penalty parameter by the square of the first allowed increment, and the second penalty coefficient is obtained by dividing the penalty parameter by the square of the second allowed increment. The product of the first penalty coefficient and the identity matrix of the preset dimension is used as the first block, and the product of the second penalty coefficient and the identity matrix of the preset dimension is used as the second block. The first block and the second block are arranged diagonally to form a block diagonal matrix, which is used as the penalty weight matrix.

[0038] Constructing the penalty weight matrix The mathematical expression for the penalty matrix is: ; in, For the penalty matrix, As the first allowed increment, For the second allowed increment, It is a 50×50 identity matrix, and r is a preset penalty parameter; Indicates two The matrix is ​​arranged diagonally to generate a The block diagonal matrix R; In the penalty weight matrix Below, when the current increment reaches the first allowable increment, the value of the current penalty term in the cost function is: The penalty value for the yaw angle when the yaw angle increment reaches the second permissible increment is also [value missing]. In this embodiment The value of is 1; This invention generates an adaptive interference feedback weight matrix by inverting the diagonal covariance matrix of the measurement noise. During the optimization process, it directly reduces the prediction error weight of sensors exposed to strong direct light, such as oncoming headlights, from a mathematical perspective, thus achieving adaptive immunity against local illumination distortion. By introducing a penalty weight matrix based on the maximum permissible increment of different physical quantities in a single step, it smooths out the difference in dimensions and orders of magnitude between current and yaw angle, effectively curbing the algorithm's tendency to make frequent, minor, or excessive adjustments to the yaw angle, and reducing the adjustment losses of mechanical components.

[0039] The process of constructing and solving the control cost function includes: The illumination estimation vector is mapped to the virtual feature space through the left eigenvector matrix to obtain the current virtual state vector; Based on predicted time steps Virtual target illuminance Interference feedback matrix Penalty weight matrix The process of constructing the control cost function includes: This embodiment will solve the future The sequence consisting of the physical control increment sequences at each time step is used as the control increment sequence. ,in, That is, the optimal control increment sequence to be solved in the current control cycle; The dimension is ; To predict the future The traditional method for determining the illuminance distribution across the entire road segment at each time step is an iterative approach, which utilizes the illuminance distribution during the current control cycle. Calculate the illuminance distribution for the next control cycle. Because the light spots overlap, the illuminance at any sensor location... It depends not only on the illuminance of the previous control cycle at that location, but also on the combined influence of all other position sensors. If subsequent rolling iterative optimization is performed directly in physical space, each prediction step requires the inversion of a 100-dimensional coupling matrix, resulting in high computational complexity. To address the aforementioned issues, this embodiment will use the optimal illuminance estimation vector. Mapping to a virtual space to obtain a virtual state vector, and using this virtual state vector as a starting point, prediction is performed in the virtual space. The specific process includes: The illumination state transition matrix is ​​decomposed into eigenvectors to obtain an eigenvector matrix. The optimal illumination estimate vector is then projected onto an orthogonal virtual feature space to achieve decoupling and dimensionality reduction. The feature space projection process includes: Eigenvalue decomposition of the state transition matrix yields the diagonal feature matrix. Right eigenvector matrix and left eigenvector matrix ,because It is a real symmetric matrix, and its eigenvector matrices satisfy orthogonality; left eigenvector matrix The product of the illumination state transition matrix B and the illuminance matrix is ​​used as the virtual input matrix. ; Using the left eigenvector matrix The optimal illuminance estimation vector Mapping to the virtual feature space yields the virtual state vector. The mathematical expression for the virtual state vector is: ; in, For virtual state vectors, The left eigenvector matrix, This is the optimal illuminance estimation vector; yes A column vector of dimension, whose ? element An eigenvector corresponding to the illuminance state transition matrix A represents the projected intensity of the illuminance distribution across the entire road segment in the direction of that eigenspace. Using the decoupled scalar difference equation, starting from the current virtual state vector, and combining the control command of the previous control cycle and the control increment sequence to be solved, the virtual state prediction value is obtained by recursively predicting multiple future control cycles. In the virtual feature space, originally by The described coupling relationship is equivalently transformed into 100 mutually independent first-order scalar difference equations, the mathematical expressions of which are: ; in, Let be the predicted intensity value of the m-th independent mode in the virtual space in the next control cycle. Let m be the m-th diagonal element of the diagonal characteristic matrix. Let m be the intensity value of the m-th independent mode in the virtual space during the current control cycle. Virtual input matrix The m-th line, This is the column vector of control commands for the current control cycle; The m-th independent mode refers to the m-th virtual state component in the virtual feature space. The corresponding illuminance distribution pattern; The value of is 1, 2, ..., 100. It is the predicted value of the m-th virtual state component at the next time step. The m-th element; Due to the illumination state transition matrix The feature vectors are mutually orthogonal, and each virtual state component The corresponding illuminance distribution of the entire road section in the matrix The projection intensity along an eigenvector direction, the evolution of each component is independently described by the aforementioned first-order scalar difference equation, and they are uncoupled. The value depends only on the previous control cycle. and ; Through the above transformation, the prediction is transformed from a complex 100-dimensional coupled matrix operation into 100 one-dimensional scalar recursions. Based on this independent structure, multi-step prediction can be expressed as a simple linear function. It is the starting point for future predictions. , ,…, Based on The virtual state predictions, along with the predicted values ​​of future control inputs, are stacked vertically to form a... ; ; The virtual state prediction values ​​in the future prediction time domain are expressed as a linear function of the control increment sequence to be solved through the coefficient matrix, and the stacked vector of virtual state prediction values ​​is obtained. The mathematical expression for the stacked vector of virtual state predictions is: ; in, This is a stacked vector of virtual state predictions. , , Both are coefficient matrices. This is the virtual state vector for the current control cycle. This is a column vector of historical control instructions, i.e., control instructions executed in the previous control cycle. The control increment sequence to be solved; , , Depend on and ; It is The coefficient matrix mapped to the predicted values ​​at each future step has dimensions of . ,Depend on It is formed by vertically stacking successive powers; ; in It is a diagonal matrix; It is The coefficient matrix mapped to the predicted values ​​at each future step has dimensions of . , its first Each row block (corresponding to the control cycle) ) is defined as: ; The mathematical expression for the overall structure is: ; It is to control the increment sequence The coefficient matrix mapped to the predicted values ​​at each future step has dimensions of . is a lower triangular block matrix, whose first... The first row block, the first Column blocks ( )for: ; when When this occurs, the block is a zero matrix; The mathematical expression for the overall structure is: ; The target illuminance distribution is mapped to a virtual feature space to obtain the virtual target illuminance distribution; Due to the optimal illuminance estimation vector The target illuminance distribution defined in the physical space has already been mapped to the virtual feature space. It is necessary to also map the target illuminance distribution defined in the physical space to the virtual space through the same left eigenvector matrix. Only when the two are in the same coordinate space can they be subtracted to obtain the error between the target illuminance distribution and the road surface illuminance distribution required by the control cost function. Otherwise, the calculation cannot be performed. The target illuminance distribution is mapped to a virtual space to obtain the virtual target illuminance distribution. The mathematical expression for the virtual target illuminance distribution is: ; in, Let be the illuminance distribution vector of the virtual target. The left eigenvector matrix, The target illuminance vector; Subtract the stacked vector of virtual state prediction values ​​from the virtual target sequence formed by the vertical repetition of virtual target illumination distribution, and multiply it by the inverse mapping matrix formed by the block diagonal expansion of the right feature vector matrix to obtain the prediction error vector of physical space.

[0040] Since the above transformations are all performed in virtual space, it is necessary to inversely map the prediction error in virtual space back to physical space so that adaptive weights from the physical sensor space can be used for weighting. The prediction error in virtual space is taken as the first error, and the prediction error in physical space is taken as the second error. The mathematical expression for the second error is: ; in, This represents the prediction error vector in physical space. It is a block diagonal extension matrix of the right eigenvector matrix Ψ. For the future A stacked vector of virtual state predictions for each control cycle. Virtual reference vector Vertical repeated stacking The vector obtained this time; This invention maps the highly coupled physical space illumination state transition matrix to an orthogonal virtual feature space through eigenvalue decomposition, effectively decoupling the coupled prediction operation into multiple independent first-order scalar difference equations. This enables long-term rolling prediction derivation to be completed in a short time, and finally seamlessly connects back to the physical space through inverse mapping for adaptive weighting.

[0041] The process of constructing and solving the control cost function also includes: Construct a control cost function, which is a quadratic form of the block diagonal extension of the prediction error vector and the disturbance feedback matrix in the physical space, and a quadratic form of the block diagonal extension of the control increment sequence and the penalty weight matrix. The mathematical expression for the control cost function is: ; in, To control the cost function, the objective of this embodiment is to find a set of control increment sequences such that... The value is the smallest; The second error has the following dimensions: By using the predicted state error in virtual space The result obtained after inverse mapping back to physical space is that each element represents the deviation between the predicted illuminance and the target illuminance at a certain sensor location and in a certain control cycle in the future, with the unit being lx; This is a block diagonal extension of the interference feedback matrix, with dimensions of... ;Depend on Copy along diagonal This shows that when a sensor is subjected to strong light interference, causing an increase in measurement noise variance... The corresponding diagonal elements become smaller, thereby reducing the weight of the sensor's prediction error in the control cost function; Let be the control increment sequence to be solved, with dimension . It is the future Control increment sequence at each time step The vector formed by vertical stacking is the variable that needs to be solved in the quadratic programming problem of this embodiment; Penalty weight matrix The block-diagonal expansion form, with dimensions of Depend on Copy along diagonal This time it was obtained; Considering the weight matrix Given a diagonal matrix in the physical sensor space, the virtual state prediction error needs to be inversely mapped back to the physical space. This requires defining a block-diagonal extended matrix: for A block diagonal matrix, where each block on the diagonal is ,because For offline pre-stored constant matrices, In each cycle, only needs to be according to It can be generated simply by copying and arranging; the computational cost is negligible. for The first block diagonal matrix, each block is ; for The second block diagonal matrix, each block is ; To control the current cycle Vertical repeated stacking This time it was obtained vector; The control cost function is rearranged into a standard quadratic form with respect to the control increment sequence. Combined with the first and second constraints, the control cost function is solved using the activity set method to obtain the optimal control increment sequence. The component corresponding to the current control cycle in the optimal control increment sequence is extracted as the optimal control increment sequence.

[0042] Substituting the second error and the virtual target illuminance distribution into the control cost function, we can reorganize it into the following: The standard quadratic form of the control cost function, after simplification, is expressed as follows: ; Where T is the transpose of a matrix or vector. The control cost function is in standard quadratic form. The control increment sequence to be solved , , Both are coefficient matrices. It is a block diagonal extension matrix of the right eigenvector matrix Ψ. This is the first block diagonal matrix. This is the second block diagonal matrix; For virtual state vectors, To control the current cycle Vertical repeated stacking This time it was obtained vector; Historical control commands; Based on the control cost function, the first constraint, and the second constraint, the solution is obtained using the activity set method. An all-zero vector is used as the initial search point, the maximum number of iterations is set to 100, and the tolerance error is [missing value]. The iteration terminates when the change in the objective function value between two consecutive iterations is less than the tolerance error, and the output is the optimal control increment sequence. Only the first 100 components of this sequence are extracted as the control increment sequence issued in the current control cycle. The optimal control increment sequence is added to the actual command executed in the previous cycle to obtain the control command vector for the current cycle; this vector is then encoded and sent to each street light node for execution via the communication bus. The control command vector issued is sent back to the estimation module and used to calculate the prior state prediction in the next control cycle.

[0043] This invention integrates the physical space prediction error and control increment penalty into a standard quadratic control cost function in a block-diagonal extended form, and solves it efficiently using the active set method. This transforms the contradictory objectives of minimizing illuminance error and stabilizing control amplitude into a rigorous mathematical optimization process, ensuring that the optimal control increment sequence can both make the future road illuminance approach the target distribution and ensure a smooth transition of control command output, thus avoiding high-frequency flickering and abrupt changes in lighting.

[0044] The first constraint is that the current and yaw angle of each control increment sequence do not exceed the preset maximum allowable rate of change per step; the second constraint is that the current and yaw angle of the street light node in any control cycle within the prediction time domain are within the preset allowable range.

[0045] The mathematical expression for the first constraint is: ; in, The lower bound vector of the rate of change The control increment sequence to be solved is... This is the upper bound vector of the rate of change; and It is the lower bound of the single-step rate of change. and the upper bound of the rate of change in one step Stack vertically repeatedly The dimensions obtained are all... ; In this embodiment, the lower bound of the single-step rate of change The first 50 components are The last 50 components are Upper bound of single-step rate of change The first 50 components are The last 50 components are ; That is, the lower bound vector of the single-step rate of change constraint is The upper bound vector is ; The second constraint is used to apply an absolute amplitude constraint on the control quantity. The constraint starts from the current moment and the total sum of the control quantity accumulated to any future control cycle cannot exceed the physical allowable range. Specifically, in any control cycle within the prediction time domain, the current of the 50 streetlights must not exceed [a certain value]. The yaw angle must not exceed ; For the h-th step in the prediction time domain The cumulative control amount is The cumulative control quantity must satisfy the second constraint, the mathematical expression of which is: ; in, This is the column vector of control instructions executed in the previous control cycle. This is the cumulative control increment sequence from the current moment to the h-th future step; This is the lower bound vector for the absolute amplitude constraint of the single-step control quantity. This is the upper bound vector for the absolute amplitude constraint of the single-step control quantity; In this embodiment ; ; express A vector consisting entirely of zeros. express A vector consisting entirely of 1s; The second constraint is about Linear inequality constraints are determined by constructing a constraint matrix. and constant vector It can be converted into its standard form, and the standard mathematical expression is: ; in, The inequality constraint matrix is... The inequality constraint vector; It is a lower triangular block matrix, where each row corresponds to a constraint. Its structure ensures... The elements in the time sequence are added sequentially. Then by and Subtract respectively get.

[0046] This invention maps the complex coupling relationship of adjacent light spots overlapping in physical space to an orthogonal virtual feature space, transforming it into mutually independent first-order scalar difference equations. This reduces the computational complexity of multi-step predictive control, enabling complex optimization algorithms to be solved efficiently under hard real-time constraints of single-step control cycles. As a result, it achieves high-precision, low-latency smooth tracking compensation for high-speed vehicles, effectively eliminating the dangerous visual dark area phenomenon caused by system response lag. This invention utilizes an adaptive Kalman filter algorithm to estimate the diagonal covariance matrix of measurement noise online, and constructs an interference feedback matrix based on this matrix. It can adaptively adjust the feedback weights based on the real-time data of each road surface sensor. This mechanism effectively suppresses transient environmental interference such as direct glare from oncoming headlights, avoids abnormal adjustment commands due to local measurement distortion, and significantly improves the control robustness and illuminance distribution stability under complex road conditions. This invention introduces a penalty weight matrix based on the maximum permissible increment of the control quantity in the control cost function, coordinating the driving current and horizontal yaw angle control quantities with differences in magnitude and dimension. This invention not only effectively avoids frequent and extreme adjustments of sensitive angle parameters by the algorithm, reducing the mechanical and electrical losses of physical street light equipment, but also achieves on-demand segmented dynamic lighting by combining the real-time coordinates and speed of the target vehicle to dynamically calculate and predict the time domain and safe lighting range, maintaining high illumination in front of vehicles and low illumination in areas without vehicles. This maximizes energy saving while comprehensively ensuring the safety of nighttime driving on highways.

[0047] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0048] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. A segmented compensation control system for illumination uniformity along a highway, characterized in that, include: The data acquisition module is used to collect road condition data for the target lighting section; The estimation module is used to calculate the illuminance estimation vector and the measurement noise diagonal covariance matrix based on the illuminance data in the road condition data, combined with the preset control illuminance mapping matrix and illuminance state transition matrix, using the adaptive Kalman filter algorithm. The optimization module is used to determine the prediction time domain and construct the target illuminance distribution based on the vehicle coordinate data in the road condition data, construct the interference feedback matrix based on the measurement noise diagonal covariance matrix, and construct a control cost function in the virtual feature space based on the illuminance estimation vector, which includes the target illuminance distribution, the interference feedback matrix and the preset penalty weight matrix. The module solves the control cost function to obtain the optimal control increment sequence and generates the control command for the current control cycle. An execution module is used to adjust according to the control instructions.

2. The segmented compensation control system for illuminance uniformity along a highway as described in claim 1, characterized in that, The road condition data includes illuminance data, vehicle coordinate data, vehicle speed data, and geometric data; The illuminance data is collected by an illuminance sensor; The geometric data includes the three-dimensional spatial coordinates of all street light nodes and the two-dimensional ground coordinates of all illuminance sensors.

3. The segmented compensation control system for illuminance uniformity along a highway as described in claim 2, characterized in that, The process of constructing the control illuminance mapping matrix specifically includes: Acquire the luminous intensity data of the street light nodes, and calculate the incident angle and azimuth angle of the light reaching the illuminance sensor based on the three-dimensional spatial coordinates of the street light nodes and the two-dimensional ground coordinates of the illuminance sensor. Based on the Lambert point light source formula, and combined with the incident angle, the azimuth angle, the luminaire photometric data and the preset current, a mapping equation from street light node control parameters to illuminance is constructed. Using a preset rated operating point as the expansion center, a small perturbation is applied to the control parameters. The partial derivatives of each control parameter with each illuminance sensor reading are calculated using the numerical difference method. The partial derivatives are then combined to form the control illuminance mapping matrix.

4. The segmented compensation control system for illuminance uniformity along a highway as described in claim 3, characterized in that, The process of constructing the illuminance state transition matrix specifically includes: Set all street light nodes to the preset rated operating point, and sequentially add a step pulse of preset amplitude to the control current of each street light node; Acquire illuminance data at multiple time steps after the superposition of the step pulse, and construct an illuminance response dataset; Construct a discrete linear state-space equation, which includes the initial illuminance state transition matrix to be identified; Using the illuminance response dataset, the discrete linear state-space equations are solved using the least squares method with regularization terms to obtain the initial illuminance state transition matrix; The initial illuminance state transition matrix is ​​symmetricized to obtain the illuminance state transition matrix.

5. A segmented compensation control system for illuminance uniformity along a highway as described in claim 4, characterized in that, The process of calculating the illuminance estimation vector and the measurement noise diagonal covariance matrix specifically includes: Based on the illuminance state transition matrix, the illuminance estimation vector of the previous control cycle, the control illuminance mapping matrix, and the historical control commands, the prior illuminance prediction vector of the current control cycle is calculated. Calculate the prediction residual vector between the illuminance data collected in the current control cycle and the prior illuminance prediction vector; Based on the historical prediction residual vector within a preset sliding window length, the measurement noise variance estimate of each illuminance sensor is calculated, and the measurement noise diagonal covariance matrix is ​​constructed. Calculate the Kalman gain matrix based on the measurement noise diagonal covariance matrix and the process noise covariance matrix; Using the Kalman gain matrix, the prior illuminance prediction vector, and the prediction residual vector, the illuminance estimation vector for the current control cycle is calculated, and the estimation error covariance matrix is ​​updated.

6. A segmented compensation control system for illuminance uniformity along a highway as described in claim 5, characterized in that, The process of determining the prediction time domain specifically includes: Obtain vehicle speed data from road condition data, and based on the vehicle speed data, the preset safe lighting distance, and the control cycle time interval, calculate the theoretical number of time steps required for the vehicle to travel past the safe lighting distance as the length of the prediction time domain; The process of constructing a virtual target illuminance distribution specifically includes: The illuminance sensor index corresponding to the front end of the vehicle is determined based on the vehicle coordinate data. Based on the illuminance sensor index, the preset safe lighting distance, and the preset total number of sensors, calculate the start and end indices of the safe lighting zone; The target illuminance of the illuminance sensor located within the safe lighting zone is set to a first preset illuminance value, and the target illuminance of the illuminance sensor located outside the safe lighting zone is set to a second preset illuminance value, thereby generating a target illuminance distribution.

7. A segmented compensation control system for illuminance uniformity along a highway as described in claim 6, characterized in that, The main diagonal elements of the interference feedback matrix are calculated based on the reciprocal of the main diagonal elements of the measurement noise diagonal covariance matrix and a preset reference variance constant. The process of constructing the penalty weight matrix specifically includes: The maximum permissible increment of the preset current in a single step is the first permissible increment, and the maximum permissible increment of the yaw angle in a single step is the second permissible increment. The first penalty coefficient is obtained by dividing the preset penalty parameter by the square of the first allowed increment, and the second penalty coefficient is obtained by dividing the penalty parameter by the square of the second allowed increment. The product of the first penalty coefficient and the identity matrix of the preset dimension is used as the first block, and the product of the second penalty coefficient and the identity matrix of the preset dimension is used as the second block. The first block and the second block are arranged diagonally to form a block diagonal matrix, which is used as the penalty weight matrix.

8. A segmented compensation control system for illuminance uniformity along a highway as described in claim 7, characterized in that, The process of constructing and solving the control cost function includes: The illumination estimation vector is mapped to the virtual feature space through the left eigenvector matrix to obtain the current virtual state vector; Using the decoupled scalar difference equation, starting from the current virtual state vector, and combining the control command of the previous control cycle and the control increment sequence to be solved, the virtual state prediction value is obtained by recursively predicting multiple future control cycles. The virtual state prediction values ​​in the future prediction time domain are expressed as a linear function of the control increment sequence to be solved through the coefficient matrix, and the stacked vector of virtual state prediction values ​​is obtained. The target illuminance distribution is mapped to a virtual feature space to obtain the virtual target illuminance distribution; Subtract the stacked vector of the virtual state prediction values ​​from the virtual target sequence formed by the vertical repetition of the virtual target illumination distribution, and multiply it by the inverse mapping matrix formed by the block diagonal expansion of the right feature vector matrix to obtain the prediction error vector of the physical space.

9. A segmented compensation control system for illuminance uniformity along a highway as described in claim 8, characterized in that, The process of constructing and solving the control cost function further includes: The control cost function is constructed by summing the quadratic form of the prediction error vector of the physical space and the interference feedback matrix in a block-diagonal extended form, and the quadratic form of the control increment sequence and the penalty weight matrix in a block-diagonal extended form. The control cost function is rearranged into a standard quadratic form with respect to the control increment sequence. Combined with the first and second constraints, the control cost function is solved using the activity set method to obtain the optimal control increment sequence. The component corresponding to the current control cycle in the optimal control increment sequence is extracted as the optimal control increment sequence.

10. A segmented compensation control system for illuminance uniformity along a highway as described in claim 9, characterized in that, The first constraint is that each control increment sequence does not exceed a preset maximum allowable rate of change per step; the second constraint is that the current and yaw angle of the street light node in any control cycle within the prediction time domain are within a preset allowable range.

Images