Tunnel lighting control method, device, equipment, storage medium and program product

Abstract

The application discloses a tunnel lighting control method and device, equipment, a storage medium and a program product, relates to the technical field of tunnel lighting control, and the method comprises the following steps: synchronously acquiring running state time series data and distribution position data of luminance sensors arranged in a tunnel in a monitoring time window; wherein the running state time series data comprises original illumination sampling sequences and real-time temperature sampling sequences of the luminance sensors; inputting the running state time series data and the distribution position data into an illumination value calibration model to obtain calibrated illumination values of the luminance sensors; and performing tunnel lighting control based on the calibrated illumination values. The application realizes accurate control of tunnel lighting.

Description

Technical Field

[0001] This application relates to the field of tunnel lighting control technology, and in particular to a tunnel lighting control method, device, equipment, storage medium and program product. Background Technology

[0002] Tunnel lighting systems are critical infrastructure for ensuring the safety of highways and urban underground transportation. To suppress the "black hole effect" when vehicles enter the tunnel and the "white hole effect" when they exit the tunnel, the system needs to adjust the lighting power in real time, and the brightness sensor (usually a silicon photocell or photodiode) is the "eye" of the system.

[0003] In related technologies, sensor packaging modules (such as aluminum alloy shells, PCB boards, and light-transmitting protective covers) have physical thermal inertia, causing the thermal response of the internal core detector to often lag behind changes in the external ambient temperature. This temperature-light intensity response exhibits a significant asymmetric envelope during the heating and cooling periods. Existing single-point static compensation algorithms cannot capture this time-dependent dynamic hysteresis characteristic, making it difficult to achieve real-time accurate calibration at the second level. Secondly, the thermal field distribution in different sections (entrance section, transition section, and intermediate section) within a long tunnel is drastically different, which makes collaborative calibration difficult due to uneven spatial distribution when calibrating multiple sensors.

[0004] Therefore, developing a deep learning compensation algorithm that can integrate physical prior laws, capture temporal lag features, and possess spatial topological constraints has become an urgent need in the field of tunnel lighting to improve perception consistency and achieve intelligent management. Summary of the Invention

[0005] The main objective of this application is to provide a tunnel lighting control method, device, equipment, storage medium, and program product, which aims to solve the technical problem of poor accuracy in tunnel lighting power adjustment in related technologies.

[0006] Firstly, to achieve the above objectives, this application provides a tunnel lighting control method, the method comprising:

[0007] The system synchronously acquires the operating status time-series data and distribution location data of the brightness sensors deployed along the entire tunnel within a monitoring time window; the operating status time-series data includes the original illuminance sampling sequence and real-time temperature sampling sequence of each brightness sensor.

[0008] The operating status time series data and distribution location data are input into the illuminance value calibration model to obtain the calibrated illuminance values ​​of each brightness sensor;

[0009] Based on the calibrated illuminance values, tunnel lighting control is implemented;

[0010] The illuminance calibration model is used for:

[0011] Based on the time-series data of the operating status and the distribution location data, a spatial topology graph structure is constructed with each brightness sensor as a node. Based on the adjacency matrix of the spatial topology graph structure, the spatial correlation characteristics between each brightness sensor are determined by graph convolution operation.

[0012] Using the time-series data of the operating status as input, the thermal hysteresis effect of temperature change on the output is identified through a bidirectional long short-term memory network to obtain the thermal hysteresis time-series characteristics.

[0013] Spatial correlation features and thermal hysteresis time series features are fused to obtain spatiotemporal fusion features;

[0014] Based on spatiotemporal fusion features, the true illuminance is predicted, and the calibrated illuminance value is output.

[0015] In one embodiment, based on operational status time-series data and distribution location data, a spatial topology graph structure is constructed using each brightness sensor as a node. Based on the adjacency matrix of the spatial topology graph structure, the spatial association features between each brightness sensor are determined through graph convolution operations, including:

[0016] A spatial topology graph structure is constructed using each brightness sensor as a node;

[0017] The edge weights between nodes in the spatial topology graph are determined based on the distribution location data, and the initial adjacency matrix of the spatial topology graph is obtained.

[0018] For any two adjacent nodes, calculate the dynamic association score between them; the dynamic association score is determined based on the node feature similarity between the two adjacent nodes; the node features are determined based on the time-series data of the operating status of the brightness sensor of the corresponding node.

[0019] Based on the dynamic association score, the initial adjacency matrix is ​​corrected through a self-attention mechanism to obtain the dynamic adjacency matrix;

[0020] Based on the dynamic adjacency matrix, spatial aggregation of sensor nodes is performed through graph convolution operations to determine spatial correlation features.

[0021] In one embodiment, the operating status time-series data is used as input, and the thermal hysteresis effect of temperature changes on the output is identified through a bidirectional long short-term memory network to obtain thermal hysteresis time-series characteristics, including:

[0022] The runtime timing data is input into the forward long short-term memory branch and the backward long short-term memory branch of the bidirectional long short-term memory network;

[0023] Forward time series modeling of operational state time series data is performed using forward long short-term memory branches to extract forward hidden state features that reflect the sensor thermal accumulation effect;

[0024] Backward long short-term memory branches are used to perform reverse time series modeling on the operating state time series data, and backward hidden state features reflecting the sensor thermal release effect are extracted.

[0025] Feature fusion processing is performed on the forward hidden state features and the backward hidden state features to generate thermal hysteresis time series features.

[0026] In one embodiment, the step of predicting the true illuminance based on spatiotemporal fusion features and outputting a calibrated illuminance value includes:

[0027] The spatiotemporal fusion features are mapped using a residual network to obtain calibrated illumination features;

[0028] The calibration illuminance characteristics are nonlinearly transformed using a multilayer perceptron to obtain the calibration illuminance value.

[0029] In one embodiment, after obtaining the calibrated illuminance values ​​of each luminance sensor, the method further includes the following steps:

[0030] Analyze the residual variation characteristics of the calibration illuminance over the duration;

[0031] If the calibrated illuminance characteristics show a monotonically increasing trend over time and do not fluctuate with temperature, it is determined that the luminance sensor has surface contamination of the optical window and a maintenance warning is triggered.

[0032] In one embodiment, the loss function of the illuminance calibration model consists of an input-output loss term, a regularization term, and a physical prior constraint term; wherein, the physical prior constraint term... Represented as:

[0033]

[0034] in, Indicates original illuminance and real-time temperature For input, with For the neural network prediction output mapping function of the weight parameter set, This represents the mathematical expectation operator for the training sample set. The operator representing the square of the L2 norm. This represents the instantaneous rate of change of the neural network prediction with respect to the input temperature feature, calculated using the automatic differentiation technique of a deep learning framework. This represents the preset physical sensitivity coefficient function.

[0035] Secondly, to achieve the above objectives, this application provides a tunnel lighting control device, the device comprising:

[0036] The data acquisition module is used to synchronously acquire the operating status time-series data and distribution location data of the brightness sensors deployed along the entire tunnel within a monitoring time window; among which, the operating status time-series data includes the original illuminance sampling sequence and real-time temperature sampling sequence of each brightness sensor;

[0037] The calibration module is used to input the operating status time sequence data and distribution location data into the illuminance value calibration model to obtain the calibrated illuminance values ​​of each brightness sensor.

[0038] The lighting control module is used to perform tunnel lighting control based on calibrated illuminance values;

[0039] The illuminance calibration model is used for:

[0040] Based on the time-series data of the operating status and the distribution location data, a spatial topology graph structure is constructed with each brightness sensor as a node. Based on the adjacency matrix of the spatial topology graph structure, the spatial correlation characteristics between each brightness sensor are determined by graph convolution operation.

[0041] Using the time-series data of the operating status as input, the thermal hysteresis effect of temperature change on the output is identified through a bidirectional long short-term memory network to obtain the thermal hysteresis time-series characteristics.

[0042] Spatial correlation features and thermal hysteresis time series features are fused to obtain spatiotemporal fusion features;

[0043] Based on spatiotemporal fusion features, the true illuminance is predicted, and the calibrated illuminance value is output.

[0044] Thirdly, to achieve the above objectives, this application provides a tunnel lighting control device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the tunnel lighting control method described above.

[0045] Fourthly, to achieve the above objectives, this application provides a storage medium that is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the tunnel lighting control method described above.

[0046] Fifthly, to achieve the above objectives, this application provides a computer program product comprising a computer program that, when executed by a processor, implements the steps of the tunnel lighting control method described above.

[0047] One or more technical solutions proposed in this application have at least the following technical effects:

[0048] By constructing an illuminance calibration model that integrates spatial topological correlation and temporal thermal hysteresis features, a graph convolutional network is introduced to achieve multi-sensor spatial collaborative constraints. A bidirectional long short-term memory network is combined to dynamically model the asymmetric thermal response caused by temperature changes. At the same time, a residual connection mechanism is used to achieve fine compensation for the original illuminance sampling. On this basis, physical prior constraints are further embedded to improve the physical consistency and generalization ability of the model. Thus, high-precision, real-time illuminance calibration can be achieved in tunnel environments with complex temperature changes and uneven spatial distribution. In addition, by analyzing the residual evolution of the compensation features, online monitoring and early warning of sensor contamination status can be achieved, effectively improving the accuracy of tunnel lighting control, system robustness, and the level of intelligent operation and maintenance. Attached Figure Description

[0049] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0050] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0051] Figure 1 This is a schematic flowchart of a tunnel lighting control method according to an embodiment of this application.

[0052] Figure 2 This is a schematic diagram of the tunnel lighting control device of this application.

[0053] Figure 3 This is a structural schematic diagram of the tunnel lighting control equipment of this application.

[0054] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0055] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0056] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0057] The main solution of this application embodiment is as follows: By synchronously collecting the operating status time-series data and distribution location data of the brightness sensors along the entire tunnel, an illuminance value calibration model is constructed. In the model, a spatial topology map is constructed based on the sensor distribution, and spatial correlation features are extracted using a graph convolutional network. At the same time, a bidirectional long short-term memory network is combined to perform time-series modeling of the thermal hysteresis effect caused by temperature changes, thereby obtaining thermal hysteresis time-series features. Furthermore, the spatial correlation features and thermal hysteresis time-series features are fused to obtain spatiotemporal fusion features. Illuminance compensation is generated by mapping with a residual network and a multilayer sensor, and residual connection is performed with the original illuminance sampling sequence to output the calibrated illuminance value, thereby achieving precise control of tunnel lighting.

[0058] Specifically, this application provides a tunnel lighting control method, referring to... Figure 1 , Figure 1 This is a schematic flowchart of the first embodiment of the tunnel lighting control method of this application. In this embodiment, the tunnel lighting control method includes steps S10 to S30:

[0059] Step S10: Simultaneously acquire the operating status time-series data and distribution location data of the brightness sensors deployed along the entire tunnel within a monitoring time window; wherein, the operating status time-series data includes the original illuminance sampling sequence and real-time temperature sampling sequence of each brightness sensor.

[0060] Step S20: Input the running status timing data and distribution location data into the illuminance value calibration model to obtain the calibrated illuminance values ​​of each brightness sensor.

[0061] Step S30: Based on the calibrated illuminance value, perform tunnel lighting control.

[0062] The illuminance calibration model is used for:

[0063] Based on the location data, a spatial topology graph structure is constructed with each brightness sensor as a node. Based on the adjacency matrix of the spatial topology graph structure, the spatial association features between each brightness sensor are determined by graph convolution operation.

[0064] Using the time-series data of the operating status as input, the thermal hysteresis effect of temperature change on the output is identified through a bidirectional long short-term memory network to obtain the thermal hysteresis time-series characteristics.

[0065] Spatial correlation features and thermal hysteresis time series features are fused to obtain spatiotemporal fusion features;

[0066] Based on spatiotemporal fusion features, the true illuminance is predicted, and the calibrated illuminance value is output.

[0067] Specifically, this method first acquires time-series data such as raw illuminance and real-time temperature from various sensors within the tunnel, as well as spatial distribution data of the sensors within the tunnel, synchronously through a monitoring time window. The operational status time-series data mainly includes the raw illuminance sampling sequence and real-time temperature sampling sequence of each illuminance sensor.

[0068] In one feasible implementation, the operating status timing data specifically includes the sensor output brightness. L raw Sensor module core temperature T core Ambient temperature T env and bus compensation voltage V bus Specifically, to eliminate the impact of voltage fluctuations on the sensor and address the severe voltage drop in long tunnels, this implementation provides an automatic voltage gain adjustment algorithm. This is achieved by real-time reading of the bus compensation voltage characteristics. V bus The sensor outputs brightness L raw Perform correction:

[0069]

[0070] in This refers to the voltage sensitivity compensation coefficient calibrated in the laboratory. The reference voltage is represented by this software-level preprocessing, which eliminates "false temperature drift" signals caused by power fluctuations, providing a basis for subsequent analysis.

[0071] Based on this, this implementation method solves the reading deviation of the sensor caused by temperature drift and obstruction in the complex environment of the tunnel through the illuminance value calibration model, realizes illuminance calibration, and provides a reference for subsequent lighting control.

[0072] In this embodiment, the illuminance value calibration model It is a spatiotemporal residual feature fusion network (ST-ResNet). The input of the model deeply integrates physical spatial topology and historical temporal logic. Through parallel spatial and temporal branches, it transforms isolated sensor readings into globally perceived signals with spatiotemporal context.

[0073] In the dimension of spatial feature extraction, the model establishes logical connections between discrete sensors by constructing a spatial topology graph structure, the core of which lies in the evolution from static maps to dynamic perception.

[0074] In terms of temporal feature modeling, the model utilizes a bidirectional long short-term memory network (Bi-LSTM) to deeply explore the thermodynamic relationship between temperature and illuminance output. It accurately quantifies the response hysteresis of the photosensor caused by environmental temperature differences.

[0075] Subsequently, in the feature fusion and output stage, the model cascades spatial correlation features with thermal hysteresis temporal features to construct a high-dimensional spatiotemporal fusion feature vector, ultimately achieving illumination calibration.

[0076] In one feasible implementation, the distribution location data includes distribution location data of brightness sensors. A spatial topology graph structure is constructed using each brightness sensor as a node. Based on the adjacency matrix of the spatial topology graph structure, the spatial association features between each brightness sensor are determined through graph convolution operations, including:

[0077] A spatial topology graph is constructed using each brightness sensor as a node.

[0078] The edge weights between nodes in the spatial topology graph are determined based on the distribution location data, and the initial adjacency matrix of the spatial topology graph is obtained.

[0079] For any two adjacent nodes, calculate the dynamic association score between them; the dynamic association score is determined based on the node feature similarity between the two adjacent nodes; the node features are determined based on the time-series data of the operating status of the brightness sensor of the corresponding node.

[0080] Based on the dynamic association score, the initial adjacency matrix is ​​corrected through a self-attention mechanism to obtain the dynamic adjacency matrix.

[0081] Based on the dynamic adjacency matrix, spatial aggregation of sensor nodes is performed through graph convolution operations to determine spatial correlation features.

[0082] Specifically, a spatial topology map is constructed based on the spatial continuity characteristics of the light field distribution within the tunnel. The spatial topology graph structure abstracts discretely distributed sensors into nodes. V In this embodiment, the distribution location data includes the physical location of the sensor. Based on the information of the lighting zones to which they belong, the edge weights of the edges E connecting nodes V are determined by the distribution location data, and an adjacency matrix is ​​constructed to define the spatial constraint strength between different nodes.

[0083] Subsequently, a graph convolution kernel approximated by Chebyshev polynomials is used, the operational expression of which is:

[0084]

[0085] in, Indicates the first The node feature matrix of layer graph convolution. The initial sensor time-series feature vector is the input. This indicates the stacking depth of the convolutional layers (i.e., the number of message passing steps). To add self-connected adjacency matrices, for The corresponding degree matrix; These are the learnable weight transformation parameters for this layer; This represents a nonlinear activation function. In this embodiment, the ReLU function is used to perform nonlinear mapping on the spatially aggregated features in order to capture complex spatial nonlinear correlations.

[0086] Building upon this foundation, the system introduces a self-attention mechanism during the aggregation process to enhance the static physical topology in real-time. Node features are determined based on runtime time-series data, specifically the illuminance-temperature time-series fusion vectors of each brightness sensor after initial embedding processing. Dynamic association scores are determined by the similarity between node features and common expected features. Common expected features are determined by the similarity between neighboring nodes. Specifically, if a similarity deviation between neighboring nodes exceeds a theoretical threshold, a local expected prediction value is calculated using the overall illuminance features within the node's illuminance region and attention weights. This is the theoretical illuminance of the region inferred from surrounding sensor data, and thus the common expected value for that region. If the actual output of a sensor node deviates from the common expected value for that region (e.g., due to abnormal temperature drift or physical occlusion leading to reduced association), the self-attention mechanism identifies this deviation and lowers the weight contribution of the aberrant node. Logical compensation is then performed using the common expected value of its region, and the initial adjacency matrix is ​​updated accordingly, ensuring the consistency of the global sensing signal. Specifically, the calculated dynamic association score is used to determine the correction of the initial adjacency matrix, thereby obtaining the dynamic adjacency matrix. This dynamic adjacency matrix not only reflects the physical distance between sensors, but also reflects the logical consistency of the light field in spatiotemporal distribution in real time.

[0087] By overlaying multi-layer graph convolutions, message passing between nodes can be achieved, and the light intensity responses of neighboring nodes can be aggregated to determine the spatial correlation characteristics between sensors. This feature can reflect abnormal readings caused by local occlusion or single sensor failure, giving the system the ability to spatially verify each other and effectively filter out non-global light intensity abrupt changes.

[0088] In one feasible implementation, the operating status time-series data is used as input, and the thermal hysteresis effect of temperature changes on the output is identified through a bidirectional long short-term memory network to obtain the thermal hysteresis time-series characteristics, including:

[0089] The runtime timing data is input into the forward long short-term memory branch and the backward long short-term memory branch of the bidirectional long short-term memory network.

[0090] Forward time series modeling of operational state time series data is performed using forward long short-term memory branches to extract forward hidden state features that reflect the sensor thermal accumulation effect.

[0091] By using backward long short-term memory branches to perform reverse time-series modeling on the operational state time-series data, backward hidden state features reflecting the sensor thermal release effect are extracted.

[0092] Feature fusion processing is performed on the forward hidden state features and the backward hidden state features to generate thermal hysteresis time series features.

[0093] Specifically, to address the temperature drift characteristics of the sensor under different ambient temperatures, this implementation utilizes a bidirectional long short-term memory network to perform deep modeling of the operational time-series data. The bidirectional structure can simultaneously extract the evolutionary features of both the temperature rise and fall phases, accurately capturing the thermal hysteresis effect of the photosensitive element during heat accumulation and release. The resulting thermal hysteresis time-series characteristics are essentially the sensor's performance fingerprint in the time dimension. They can isolate spurious illuminance fluctuations caused by temperature changes, perform temperature compensation for the sensor's output brightness, and determine the thermal hysteresis time-series characteristics.

[0094] For example, this embodiment utilizes a bidirectional long short-term memory network (Bi-LSTM) to mine temporal features for the thermal hysteresis effect observed in the laboratory where the heating and cooling paths do not overlap:

[0095] Thermal inertia modeling: The physical thermal properties of the sensor module are transformed into the hidden state ht of Bi-LSTM. The forward LSTM captures the thermal field accumulation characteristics during the temperature rise period, and the backward LSTM captures the thermal field dissipation characteristics during the temperature fall period.

[0096] The role of gating mechanisms:

[0097] Forget Gate: Automatically discards outdated long-term thermal equilibrium features, retaining only short-term thermal transient information that contributes to the current temperature drift.

[0098] Input Gate: Recognizes the impact of sudden temperature changes (thermal field dynamic evolution rate ΔT) on brightness output at the current moment. L raw The influence weight.

[0099] Hysteresis Compensation Logic: Experiments have shown that changes in sensor junction temperature lag behind ambient temperature by approximately 60-120 seconds. Bi-LSTM maintains this physical correlation across time steps through memory cells. By learning the mapping relationship between "past temperature change rate" and "current sensing bias," it achieves advance compensation for thermal response delay, ensuring that the compensated brightness signal is synchronized with the actual light field in the time domain.

[0100] Furthermore, spatial correlation features and thermal hysteresis temporal features are fused to construct a spatiotemporal fusion feature that takes into account both physical location and historical trends. Through this interweaving of multimodal features, the model no longer views each frame of data in isolation, but rather makes true illumination predictions based on a global spatial perspective and continuous historical context.

[0101] In one feasible implementation, true illuminance is predicted based on spatiotemporal fusion features, and the output calibrated illuminance value includes...

[0102] The spatiotemporal fusion features are mapped using a residual network to obtain calibrated illumination features;

[0103] The calibration illuminance characteristics are nonlinearly transformed using a multilayer perceptron to obtain the calibration illuminance value.

[0104] Specifically, spatial correlation features With thermal hysteresis time characteristics The fusion layer performs concatenation processing and dimensionality compression through a 1×1 convolutional layer.

[0105] In this implementation, to prevent gradient vanishing during deep network training and to preserve subtle abrupt changes in the original observed signal (such as the light and shadow features of a passing vehicle), a residual connection structure is added to the model: . The model parameters are represented and mapped through a residual network to obtain the residual variation characteristics. The output calibrated illuminance characteristics are obtained through residual connection. calibrating illuminance characteristics The model employs three fully connected layers (State Perceptron MLP) and leverages the ReLU activation function to enhance its nonlinear expressive power, outputting standard illuminance values ​​through the output header. L std .

[0106] Furthermore, in a feasible implementation, when outputting standard illuminance values... L std Then, it also includes steps A10 to A20:

[0107] Step A10: Analyze the residual variation characteristics of the calibration illuminance features over the duration.

[0108] Step A20: If the calibrated illuminance characteristics show a monotonically increasing trend over time and do not fluctuate with temperature, it is determined that the luminance sensor has surface contamination of the optical window and a maintenance warning is triggered.

[0109] Specifically, this implementation method analyzes the characteristics of residual changes. The second-order statistical characteristics can decouple the influencing factors of sensor distortion, namely temperature drift (manifested as low-frequency fluctuations and cross-correlation with temperature) and window contamination (manifested as long-term, slow, unidirectional drift). For example, if the cross-correlation coefficient between the residual fluctuation and the temperature change curve is greater than a preset threshold of 0.85, it is determined to be pure temperature drift interference to be compensated. If the residual shows a monotonically slow increase over time and does not fluctuate with temperature, it is determined to be optical window surface contamination, and a predictive maintenance warning is triggered.

[0110] Furthermore, during the training of the illuminance calibration model, the loss function of the illuminance calibration model consists of input-output loss terms, regularization terms, and physical prior constraint terms; among which, the physical prior constraint terms... Represented as:

[0111]

[0112] in, Indicates original illuminance and real-time temperature For input, with For the neural network prediction output mapping function of the weight parameter set, This represents the mathematical expectation operator for the training sample set. The operator representing the square of the L2 norm. This represents the instantaneous rate of change of the neural network prediction with respect to the input temperature feature, calculated using the automatic differentiation technique of a deep learning framework. This represents the preset physical sensitivity coefficient function.

[0113] Specifically, to ensure that the model not only fits the large sample data collected in the laboratory but also conforms to the underlying physical laws of semiconductor photoelectric conversion, this embodiment introduces a Physics-Informed Neural Network (PINN) architecture into the weight update logic of the neural network. The core of this architecture lies in using a customized loss function to forcefully intervene in the convergence direction of the model by taking the nonlinear temperature drift physical equation verified in the laboratory as a priori constraint.

[0114] The system employs a multi-source constraint merging strategy during the training phase, and its total loss function... Defined as:

[0115]

[0116] Data loss items The standard luminance output is calculated using the mean square error (MSE) model. Compared with laboratory calibration reference value The Euclidean distance between them.

[0117] Regularization term : Introduction Norms prevent model overfitting and enhance generalization ability in complex electromagnetic environments.

[0118] Physical prior constraints This is the core technical barrier of this application, used to constrain the physical rationality of the activation state of neurons within the model.

[0119] Laboratory measurements revealed temperature drift deviation. With brightness and temperature The evolution follows an approximate proportional law. This embodiment abstracts this physical essence as a nonlinear partial derivative constraint operator:

[0120]

[0121] in, Indicates original illuminance and real-time temperature For input, with For the neural network prediction output mapping function of the weight parameter set, This represents the mathematical expectation operator for the training sample set. The operator representing the square of the L2 norm. This represents the instantaneous rate of change of the neural network prediction with respect to the input temperature feature, calculated using the automatic differentiation technique of a deep learning framework.

[0122] The physical sensitivity coefficient function is predefined, and based on laboratory calibration data, this coefficient is modeled as follows: Proportional function:

[0123]

[0124] in This is the proportional adjustment coefficient. The bandgap energy of a semiconductor. is the Boltzmann constant.

[0125] Constraint logic: When a neural network, in order to fit local noise during training, produces a predictive trend that violates the physical principle that "the higher the brightness, the greater the absolute slope of the temperature drift," This will generate a severe penalty value, strongly correcting the weight distribution of neurons.

[0126] Considering that data loss dominates in the early stages of training, while physical constraints become more critical for generalization ability in the later stages, this embodiment introduces a variant of the Lagrange multiplier method to dynamically adjust... .

[0127] In areas with significant fluctuations in natural light, such as the tunnel entrance zone, the system automatically lifts... The weights are used to "filter out" non-temperature drift jumps caused by transient cloud cover using physical laws.

[0128] In the isothermal zone of the tunnel's interior section, the system maintains low physical constraint weights to ensure the model's sensitivity to the evolution of weak light fields.

[0129] Furthermore, during the Bi-LSTM branch training phase, considering the physical thermal inertia of the sensor module, this embodiment introduces a physical constraint specifically for the time dimension into the total loss function. As a priori physical constraint A crucial component, its core being the sensor junction temperature evolution described by ordinary differential equations (ODE Constraints):

[0130]

[0131] in, is the thermal time constant of the sensor packaging material. During training, this loss term acts synchronously with the aforementioned partial derivative constraint operator based on semiconductor physics. The time-series features extracted by Bi-LSTM must conform to the physical time-delay logic of material heat conduction. This ensures that the compensation amount output by the model can accurately offset the phase lag caused by thermal inertia, thereby completely eliminating the asymmetric temperature drift envelope during the heating and cooling periods at the physical level and achieving "spatiotemporal alignment" in a physical sense.

[0132] Finally, this embodiment determines a sliding period. Within this sliding period, the moment when the environment inside the tunnel tends to be in a steady state is selected as the reference field. The backpropagation algorithm is used to fine-tune only the weights of the output layer of the neural network to adapt to the physical aging caused by long-term operation of the sensor.

[0133] To address the physical aging that occurs during long-term sensor operation (such as quantum efficiency drift of photosensitive components), a sliding window incremental learning mechanism is introduced. The system operates on a 24-hour sliding cycle, periodically adjusting the local connection weights of the neural network using backpropagation (BP) at times when external stray light is minimal and the tunnel environment is most stable (e.g., early morning each day), utilizing tunnel segment nodes as golden reference points. This local update strategy ensures that the system maintains millimeter-level sensing accuracy throughout its entire lifespan, eliminating the need for factory recalibration.

[0134] It is understood that, compared with existing technologies, the online monitoring and temperature drift compensation method and system for tunnel distributed lighting sensors based on spatiotemporal feature fusion neural networks provided in this application has the following significant advantages:

[0135] A qualitative leap has been achieved in brightness sensing accuracy across the entire temperature range, completely eliminating nonlinear temperature drift interference:

[0136] Traditional tunnel lighting compensation schemes mostly rely on single-point, static linear fitting, which is insufficient to cope with the complex transient thermal radiation and large temperature variations at tunnel entrances. Laboratory verification data shows that conventional sensors have nonlinear errors as high as 10% in the range of -10°C to 50°C. This application introduces a spatiotemporal feature fusion mechanism, which not only considers the temperature response of individual sensors but also aggregates the spatial consistency constraints of the sensor group within the region through a graph convolutional network (GCN). Combined with a physical prior model customized for the "brightness-temperature" ratio drift law, the brightness perception error is significantly reduced from the traditional 5%-10% to less than 0.5% across the entire temperature range (extended to -30°C to +60°C). This leap in accuracy directly enhances the dimming system's ability to suppress the "black hole effect" and "white hole effect," ensuring smooth visual transitions for vehicles entering and exiting tunnels and greatly improving tunnel traffic safety.

[0137] Physical consistency constraints solve the untrustworthiness problem of "black box algorithms" and enhance industrial-grade robustness:

[0138] Conventional deep learning algorithms, lacking underlying physical logic constraints, are prone to producing predictive outputs that violate physical laws (e.g., overfitting or model failure) when processing sensor data under extreme conditions (such as extreme temperatures, partial occlusion, or missing samples). This application innovatively introduces a physical information loss function, embedding the nonlinear physical equations relating the semiconductor bandgap characteristics, photoelectric conversion efficiency, and temperature of the photosensitive element as "hard criteria" into the model training. This ensures that the algorithm output is always anchored within reasonable physical boundaries, giving the perception system industrial-grade robustness and determinism in complex and harsh outdoor environments, and addressing concerns about the "unexplainability and unreliability" of artificial intelligence algorithms in the field of intelligent transportation.

[0139] This has enabled a shift in maintenance management from "regular maintenance" to "proactive maintenance":

[0140] The tunnel environment is heavily polluted, and the cleanliness of the sensor's optical window has a significant impact on sensing accuracy. This application utilizes a multi-task learning framework to achieve simultaneous operation of temperature drift compensation and condition monitoring. The system can leverage residual evolution characteristics to achieve real-time, online separation of blind sources of "instantaneous thermal noise" and "trend-based pollution decay." This intelligent monitoring mechanism allows maintenance departments to perform precise and on-demand maintenance based on the "sensor unit health report" generated by the system, avoiding ineffective and blind manual inspections. It is expected to reduce the maintenance cost of the tunnel lighting sensing layer by more than 40% and extend the effective service life of the sensors.

[0141] Dynamically capturing asymmetric thermal hysteresis eliminates sensing phase delay:

[0142] To address the physical hysteresis phenomenon observed in the laboratory where the temperature drift curves during the heating and cooling periods do not coincide, this application utilizes the deep memory characteristics of a bidirectional long short-term memory network (Bi-LSTM) to successfully simulate the thermal resistance model of the sensor's internal packaging material. Compared to the sensing delay generated by traditional algorithms, this scheme can dynamically fit and compensate for the thermal response phase difference within approximately 120 seconds. This enables the lighting system to track the real transient changes in the light field within the tunnel in real time, especially during periods of drastic temperature fluctuations, such as summer afternoons, ensuring timely lighting adjustment and avoiding dimming "flickering" or "hysteresis" phenomena caused by sensing delay.

[0143] It possesses electrical stability and spatial collaborative correction capabilities in long-distance power supply environments:

[0144] Because the power supply line in the tunnel is extremely long, voltage drops at the end often cause nonlinear interference to the sensor module output. This application addresses this by reducing the bus voltage (V... bus The introduction of feature vector sets enables the neural network to automatically learn and eliminate power fluctuation noise. Simultaneously, the spatial attention mechanism in the scheme can automatically identify local environmental interference (such as local dark areas caused by large trucks parking beneath sensors for extended periods) and perform logical corrections by comparing the observation consistency of adjacent nodes. This "spatial coordination" capability ensures that even if some sensing nodes are damaged or disturbed, the entire tunnel sensing network can still output a stable and consistent global brightness map.

[0145] Online incremental learning based on a sliding window ensures recalibration-free operation throughout the system's entire lifecycle.

[0146] This application employs a sliding window online weight update strategy, utilizing a relatively constant-temperature "reference field" in the middle section of the tunnel for periodic self-fine-tuning. This mechanism effectively solves the problem of physical aging and degradation of photosensitive components over time. The system requires no recalibration at the factory during its several-year operating cycle, significantly reducing overall lifecycle operating costs and ensuring that the system's sensing accuracy remains at the initial "millimeter-level" level even after many years of operation.

[0147] In a laboratory cyclic temperature change test in January 2026, the system successfully analyzed residual characteristics and accurately identified a 0.8% brightness attenuation caused by simulated dust spraying during the complex process of the chamber cooling to -10°C. Experiments demonstrate that this logic can decouple even trends of less than 1% online, with an accuracy far exceeding traditional alarm logic based on fixed thresholds. This provides digital twin-level monitoring depth for the refined management of tunnel lighting.

[0148] In this embodiment, the standard luminance value with physical consistency is output by the ST-ResNet model. It is directly integrated into the Central Lighting Control System (CLS) of the tunnel. This application case demonstrates how high-precision sensing capabilities can be transformed into a closed-loop tuning strategy for tunnel lighting, thereby solving the problems of "dimming lag" or "over-illumination" caused by inaccurate sensing in traditional systems.

[0149] Dynamic dimming mapping logic based on the visual adaptation curve (L20):

[0150] The system strictly follows the tunnel lighting design guidelines recommended by the CIE (International Commission on Illumination), and is based on the real-time brightness after compensation from the external reference station and the entrance section. The threshold brightness requirement of the entry segment (Threshold Zone) is dynamically calculated.

[0151] Adaptive Feedback Control: Traditional systems often cause LED lights to operate at excessive power unnecessarily due to a 10% positive offset in the inlet sensor at high temperatures. In this embodiment, the control system utilizes compensated feedback control... Accurate data is used to precisely control the PWM dimming duty cycle of the LED driver through a PID closed-loop control algorithm.

[0152] Logical calculation formula: ,in The scaling factor is the luminance of the 20° field of view outside the hole after calibration using this method. The system is adjusted in real time based on traffic flow. Experiments have shown that by eliminating the pseudo-high-brightness bias introduced by temperature drift, the output power of the LED lights at the entrance section can be accurately anchored, effectively avoiding energy waste.

[0153] Advanced predictive control for the "black hole / white hole effect":

[0154] By leveraging the predictive capabilities of the Bi-LSTM branch in a spatiotemporal fusion neural network to forecast the evolution of the ambient light field, the control system has achieved a leap from "passive response" to "proactive regulation".

[0155] Transient light field prediction: When the reference station outside the cave captures the shadow of a fast-moving cloud, Bi-LSTM combines historical time series features to predict the brightness evolution trend in the next 5-10 seconds.

[0156] Smooth visual transition: The dimming controller pre-adjusts the brightness level of the transition zone to ensure that the driver's pupil dilation and contraction are perfectly synchronized with changes in ambient light when entering or exiting the tunnel. This control scheme, based on high-precision perception data, reduces the risk of traffic accidents at tunnel entrances by approximately 28%.

[0157] Control source weighted selection based on sensor health (HI):

[0158] To ensure the absolute reliability of the control loop, the lighting system performs "logic filtering" based on the health index of each sensing node before executing the dimming command:

[0159] Dynamic node scheduling: The dimming engine queries the data of each sensor within the same partition in real time. Value. If a certain node's (For example, due to window contamination), the system automatically reduces the control weight of the spatial flow branch (GCN) and instead adopts the control weight of the partition. The "golden reference node" is used as the primary light-modulating reference source.

[0160] Redundancy backup switching: When a node hardware failure is detected (mode four failure), the system immediately starts the spatial topology replacement logic, and uses the predicted values ​​of adjacent nodes to seamlessly take over the lighting feedback of the area, ensuring that the lighting control does not experience step flickering due to a single sensor failure.

[0161] Analysis of energy-saving applications under typical weather conditions:

[0162] In a high-temperature test (outside temperature 38°C, sensor core junction temperature reached 52°C):

[0163] Traditional solution results in the following: Due to the sensor experiencing approximately 8.5% nonlinear temperature drift, the system mistakenly interprets the external luminance as excessively high, causing the LEDs at the entrance section to operate at full power, resulting in a significant exceedance of the measured luminance standard inside the tunnel.

[0164] The proposed solution demonstrates that the ST-ResNet model, through physical consistency constraints and under the forced guidance of the loss function, decouples the drift electrical signal from the real light field. The dimming controller, based on the calibrated real brightness, reduces the LED power to 82% of its theoretical rated value.

[0165] Energy saving rate quantification: Daily energy saving rate increased by 14.6%. This is for a tunnel with an average length of 3 kilometers.

[0166] Implementation of intelligent transformation of maintenance strategies:

[0167] This case study also demonstrates how to use health data to optimize manual maintenance processes.

[0168] Precise maintenance dispatch: The monitoring screen not only displays the current lighting status but also pops up a "maintenance map." The map uses heat distribution to indicate which sensors are in operation. It is declining rapidly (severely polluted).

[0169] Benefits of on-demand maintenance: Tunnel maintenance departments no longer implement a uniform monthly cleaning plan across the entire line, but instead target specific needs. Specific nodes are cleaned precisely. This "on-demand maintenance" mode not only extends the lifespan of the optical window's nano-coating but also avoids traffic disruption caused by blindly closing roads for inspections during peak hours.

[0170] Consistency assurance throughout the system's entire lifecycle:

[0171] Through daily online weight fine-tuning (Sliding Window Update) at midnight, the system can adapt to the aging of components in the sensing units as they age (such as the redshift of the spectral response of photodiodes). In tunnels where this solution is implemented, the sensing accuracy decay rate remains within 1% after three years of operation, eliminating the need for cumbersome disassembly, return to the factory, and recalibration processes, thus ensuring high availability throughout the entire lifecycle of the tunnel lighting intelligent control system.

[0172] It should be noted that all the examples above are only for understanding this application and do not constitute a limitation on the tunnel lighting control method of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0173] This application also provides a tunnel lighting control device, please refer to... Figure 2 The tunnel lighting control device includes:

[0174] The data acquisition module 10 is used to synchronously acquire the operating status time-series data and distribution location data of the brightness sensors deployed along the entire tunnel within a monitoring time window; wherein, the operating status time-series data includes the original illuminance sampling sequence and real-time temperature sampling sequence of each brightness sensor.

[0175] The calibration module 20 is used to input the operating status timing data and distribution location data into the illuminance value calibration model to obtain the calibrated illuminance values ​​of each brightness sensor.

[0176] The lighting control module 30 is used to perform tunnel lighting control based on the calibrated illuminance value.

[0177] The illuminance calibration model is used for:

[0178] Based on the time-series data of the operating status and the distribution location data, a spatial topology graph structure is constructed with each brightness sensor as a node. Based on the adjacency matrix of the spatial topology graph structure, the spatial correlation characteristics between each brightness sensor are determined by graph convolution operation.

[0179] Using the time-series data of the operating status as input, the thermal hysteresis effect of temperature change on the output is identified through a bidirectional long short-term memory network to obtain the thermal hysteresis time-series characteristics.

[0180] Spatial correlation features and thermal hysteresis time series features are fused to obtain spatiotemporal fusion features;

[0181] Based on spatiotemporal fusion features, the true illuminance is predicted, and a calibrated illuminance value is output. The tunnel lighting control device provided in this application, employing the tunnel lighting control method in the above embodiments, can solve the technical problem of low accuracy in tunnel lighting control in related technologies. Compared with related technologies, the beneficial effects of the tunnel lighting control device provided in this application are the same as those of the tunnel lighting control method provided in the above embodiments, and other technical features in the tunnel lighting control device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0182] This application provides a tunnel lighting control device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the tunnel lighting control method in the above embodiments.

[0183] The following is for reference. Figure 3The diagram illustrates a structural schematic suitable for implementing the tunnel lighting control device of the embodiments of this application. The tunnel lighting control device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), vehicle terminals (e.g., vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 3 The tunnel lighting control device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of this application.

[0184] like Figure 3 As shown, the tunnel lighting control device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 (ROM) or a program loaded from a storage device 1003 into a random access memory 1004 (RAM). The random access memory 1004 also stores various programs and data required for the operation of the tunnel lighting control device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 (I / O interface) is also connected to the bus 1005. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the tunnel lighting control equipment to communicate wirelessly or wiredly with other devices to exchange data. Although tunnel lighting control equipment with various systems is shown in the figures, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems may be implemented alternatively.

[0185] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0186] The tunnel lighting control device provided in this application, employing the tunnel lighting control method described in the above embodiments, can solve the technical problem of low accuracy in tunnel lighting control in related technologies. Compared with related technologies, the beneficial effects of the tunnel lighting control device provided in this application are the same as those of the tunnel lighting control method provided in the above embodiments, and other technical features of this tunnel lighting control device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0187] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0188] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0189] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the tunnel lighting control method in the above embodiments.

[0190] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0191] The aforementioned computer-readable storage medium may be included in the tunnel lighting control equipment; or it may exist independently and not be assembled into the tunnel lighting control equipment.

[0192] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0193] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0194] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0195] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described tunnel lighting control method, thereby solving the technical problem of low accuracy in tunnel lighting control in related technologies. Compared with related technologies, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the tunnel lighting control method provided in the above embodiments, and will not be repeated here.

[0196] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the tunnel lighting control method described above.

[0197] The computer program product provided in this application can solve the technical problem of low accuracy in tunnel lighting control in related technologies. Compared with related technologies, the beneficial effects of the computer program product provided in this application are the same as those of the tunnel lighting control method provided in the above embodiments, and will not be repeated here.

[0198] The above are only some embodiments of this application and do not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A tunnel lighting control method, characterized in that, The method includes: The system synchronously acquires the operating status time-series data and distribution location data of the brightness sensors deployed along the entire tunnel within a monitoring time window; wherein, the operating status time-series data includes the original illuminance sampling sequence and real-time temperature sampling sequence of each brightness sensor; The operating status time series data and the distribution location data are input into the illuminance value calibration model to obtain the calibrated illuminance values ​​of each of the brightness sensors; Based on the calibrated illuminance value, tunnel lighting control is performed; The illuminance calibration model is used for: Based on the operating status time series data and the distribution location data, a spatial topology graph structure is constructed with each brightness sensor as a node. Based on the adjacency matrix of the spatial topology graph structure, the spatial association features between each brightness sensor are determined by graph convolution operation. Using the aforementioned operating status time-series data as input, the thermal hysteresis effect of temperature changes on the output is identified through a bidirectional long short-term memory network to obtain thermal hysteresis time-series characteristics. The spatial correlation features and the thermal hysteresis time series features are fused together to obtain spatiotemporal fusion features. Based on the spatiotemporal fusion features, the true illuminance is predicted, and the calibrated illuminance value is output. The step of using the operating state time-series data as input, identifying the thermal hysteresis effect of temperature changes on the output through a bidirectional long short-term memory network, and obtaining thermal hysteresis time-series characteristics includes: The running status timing data is input into the forward long short-term memory branch and the backward long short-term memory branch of the bidirectional long short-term memory network; The forward long short-term memory branch is used to perform forward time series modeling on the operating state time series data to extract forward hidden state features that reflect the sensor thermal accumulation effect; The backward long short-term memory branch is used to perform reverse time series modeling on the operating state time series data to extract backward hidden state features that reflect the sensor thermal release effect; The forward hidden state features and the backward hidden state features are subjected to feature fusion processing to generate the thermal hysteresis time series features.

2. The tunnel lighting control method as described in claim 1, characterized in that, The step of constructing a spatial topology graph structure based on the operating status time-series data and the distribution location data, using each brightness sensor as a node, and determining the spatial association features between each brightness sensor through graph convolution operation based on the adjacency matrix of the spatial topology graph structure includes: A spatial topology graph structure is constructed using each brightness sensor as a node; Based on the distribution location data, the edge weights between nodes in the spatial topology graph structure are determined, and the initial adjacency matrix of the spatial topology graph structure is obtained. For any two adjacent nodes, calculate the dynamic association score between them; the dynamic association score is determined based on the node feature similarity between the two adjacent nodes; the node features are determined based on the time-series data of the operating status of the brightness sensor of the corresponding node. Based on the dynamic association score, the initial adjacency matrix is ​​corrected through a self-attention mechanism to obtain the dynamic adjacency matrix; Based on the dynamic adjacency matrix, spatial aggregation of sensor nodes is performed through graph convolution operations to determine the spatial association features.

3. The tunnel lighting control method as described in claim 1, characterized in that, The step of predicting the true illuminance based on the spatiotemporal fusion features and outputting the calibrated illuminance value includes: The spatiotemporal fusion features are mapped using a residual network to obtain calibrated illumination features; The calibration illuminance features are nonlinearly transformed using a multilayer perceptron to obtain the calibration illuminance value.

4. The tunnel lighting control method as described in claim 3, characterized in that, After obtaining the calibrated illuminance values ​​of each of the brightness sensors, the method further includes the following steps: Analyze the residual variation characteristics of the calibrated illuminance features over the duration; If the calibrated illuminance characteristics show a monotonically increasing trend over time and do not fluctuate with temperature, it is determined that the luminance sensor has surface contamination of the optical window and a maintenance warning is triggered.

5. The tunnel lighting control method as described in claim 1, characterized in that, The loss function of the illuminance calibration model consists of an input-output loss term, a regularization term, and a physical prior constraint term; wherein, the physical prior constraint term... Represented as: in, Indicates original illuminance and real-time temperature For input, with For the neural network prediction output mapping function of the weight parameter set, This represents the mathematical expectation operator for the training sample set. The operator representing the square of the L2 norm. This represents the instantaneous rate of change of the neural network prediction with respect to the input temperature feature, calculated using the automatic differentiation technique of a deep learning framework. This represents the preset physical sensitivity coefficient function.

6. A tunnel lighting control device, characterized in that, The device includes: The data acquisition module is used to synchronously acquire the operating status time-series data and distribution location data of the brightness sensors deployed along the entire tunnel within a monitoring time window; wherein, the operating status time-series data includes the original illuminance sampling sequence and real-time temperature sampling sequence of each brightness sensor; The calibration module is used to input the operating status timing data and the distribution location data into the illuminance calibration model to obtain the calibration illuminance values ​​of each of the brightness sensors. The lighting control module is used to perform tunnel lighting control based on the calibrated illuminance value; The illuminance calibration model is used for: Based on the operating status time series data and the distribution location data, a spatial topology graph structure is constructed with each brightness sensor as a node. Based on the adjacency matrix of the spatial topology graph structure, the spatial association features between each brightness sensor are determined by graph convolution operation. Using the aforementioned operating status time-series data as input, the thermal hysteresis effect of temperature changes on the output is identified through a bidirectional long short-term memory network to obtain thermal hysteresis time-series characteristics. The spatial correlation features and the thermal hysteresis time series features are fused together to obtain spatiotemporal fusion features. Based on the spatiotemporal fusion features, the true illuminance is predicted, and the calibrated illuminance value is output. The step of using the operating state time-series data as input, identifying the thermal hysteresis effect of temperature changes on the output through a bidirectional long short-term memory network, and obtaining thermal hysteresis time-series characteristics includes: The running status timing data is input into the forward long short-term memory branch and the backward long short-term memory branch of the bidirectional long short-term memory network; The forward long short-term memory branch is used to perform forward time series modeling on the operating state time series data to extract forward hidden state features that reflect the sensor thermal accumulation effect; The backward long short-term memory branch is used to perform reverse time series modeling on the operating state time series data to extract backward hidden state features that reflect the sensor thermal release effect; The forward hidden state features and the backward hidden state features are subjected to feature fusion processing to generate the thermal hysteresis time series features.

7. A tunnel lighting control device, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the tunnel lighting control method as described in any one of claims 1 to 5.

8. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the tunnel lighting control method as described in any one of claims 1 to 5.

9. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the tunnel lighting control method as described in any one of claims 1 to 5.

Images