An intelligent carbon emission analysis system for highway tunnel lighting

Abstract

The present invention discloses an intelligent carbon emission analysis system for highway tunnel lighting, comprising: a demand analysis module for calculating the required lighting brightness of a plurality of lighting devices in the tunnel according to the traffic flow, the natural light intensity outside the tunnel, and the PM<subgt;2.5< / subgt; value inside the tunnel collected in real time; a carbon efficiency analysis module for constructing a tunnel carbon emission model, quantifying the uncertainty of the tunnel carbon emission model by using Monte Carlo simulation, and obtaining a carbon emission confidence interval; and an environmental optimization module for, when the upper limit of the carbon emission confidence interval is greater than a preset threshold, optimizing the device control parameters by using a reinforcement learning framework to obtain device optimization parameters, and controlling the devices by using the device optimization parameters. The present invention relates to the technical field of carbon emission optimization, and solves the technical problem of insufficient dynamic adaptability of existing carbon emission analysis methods.

Description

Technical Field

[0001] The present invention belongs to the field of highway tunnels, relates to carbon emission optimization technology, and specifically is an intelligent carbon emission analysis system for highway tunnel lighting. Background Art

[0002] As an important part of transportation infrastructure, the energy consumption and carbon emission problems of the lighting system in highway tunnels are becoming increasingly prominent. The carbon emissions of tunnel lighting not only include the electricity consumption during the operation stage, but also involve the embodied emissions in the whole life cycle such as equipment production, transportation, and maintenance. Accurately calculating and optimizing the carbon emissions of tunnel lighting is of great significance for reducing the carbon footprint in the transportation field and promoting the development of green transportation.

[0003] However, existing methods for analyzing carbon emissions in tunnel lighting mostly calculate using fixed parameters, relying on preset lighting standards and equipment parameters, and it is difficult to dynamically adjust the lighting strategy according to the real-time environment. As a result, during peak traffic hours or under adverse weather conditions, the lighting brightness cannot meet the actual needs, which easily increases the risk of traffic accidents; when natural light is sufficient, the artificial lighting brightness cannot be reduced in time, resulting in serious energy waste and further increasing the carbon emission burden. At the same time, due to the lack of comprehensive consideration of the embodied carbon emissions in the whole life cycle such as equipment production, transportation, and maintenance, the carbon emission assessment of the tunnel lighting system is not comprehensive and accurate enough, and it cannot provide a reliable basis for formulating energy-saving and emission-reduction measures. In addition, existing methods do not fully consider the impact of lighting on the comfort of drivers and passengers, nor do they establish an effective feedback mechanism to optimize the lighting scheme, making it difficult to achieve intelligent and green management of the lighting system. Summary of the Invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art; for this purpose, the present invention proposes an intelligent carbon emission analysis system for highway tunnel lighting to solve the technical problem of insufficient dynamic adaptability of existing carbon emission analysis methods.

[0005] To achieve the above object, the first aspect of the present invention provides an intelligent carbon emission analysis system for highway tunnel lighting, including:

[0006] A demand analysis module: used to analyze and control the lighting brightness of several lighting devices in the tunnel according to the traffic flow, natural light illumination outside the tunnel, and PM 2.5 value collected in real time;

[0007] A carbon efficiency analysis module: used to construct a tunnel carbon emission model, quantify the uncertainty of the tunnel carbon emission model using Monte Carlo simulation, and obtain a carbon emission confidence interval;

[0008] Environmental optimization module: When the upper limit of the carbon emission confidence interval is greater than the preset threshold, it is used to optimize the device control parameters using a reinforcement learning framework to obtain device optimization parameters, and use the device optimization parameters to control the device; wherein, the device includes lighting devices and ventilation devices.

[0009] Further, the analysis and control of the lighting brightness of several lighting devices in the tunnel includes:

[0010] A1. Divide the tunnel into several segments according to a preset distance, and collect the traffic flow V(t), PM value PM(t) in the tunnel, and natural light illumination L(t) outside the tunnel in real time; wherein, t represents the current time; 2.5 value PM(t), and the natural light illumination L ext (t); where t represents the current time;

[0011] A2. Calculate the visibility C(t) based on the PM value PM(t) in the tunnel, and the calculation formula is: 2.5 ; ;

[0012] A3. Calculate the basic lighting brightness I(t) of several segments based on the traffic flow V(t), visibility C(t), and natural light illumination L(t) outside the tunnel, and the calculation formula is: ext value I(t), and the calculation formula is: base ; where α, β, and γ represent dynamic learning parameters, which are obtained by online updating through a neural network model, V represents the maximum traffic capacity of the tunnel, and L(x) represents the segment reference illuminance; ; where α, β, and γ represent dynamic learning parameters, which are obtained by online updating through a neural network model, V max represents the maximum traffic capacity of the tunnel, and L base (x) represents the segment reference illuminance;

[0013] A4. Set the segment correction coefficient S(x) of several segments, calculate the segment lighting brightness according to the segment correction coefficient, and control several lighting devices according to the segment lighting brightness; where x represents the segment index.

[0014] Further, the calculation of the segment lighting brightness according to the segment correction coefficient includes:

[0015] A41. Calculate the required illuminance I'(t) of several segments according to the formula I'(t)=I base (t)×S(x);

[0016] A42. Determine whether the required illuminance I'(t) of several segments is greater than or equal to the segment reference illuminance L base (x); if so, mark the required illuminance I'(t) as the segment lighting brightness I(t); if not, mark the segment reference illuminance L base (x) as the segment lighting brightness I(t).

[0017] By collecting traffic flow, PM 2.5Value and natural illumination data, combined with visibility and dynamic learning parameters, realize the adaptive adjustment of lighting brightness. And the parameters are updated online through a neural network to ensure that the lighting brightness can not only meet the safety requirements, but also make the most of natural light to reduce energy consumption, solve the problems of energy waste and insufficient adaptability in the traditional fixed illuminance mode, and achieve the balance between energy conservation and safety.

[0018] Furthermore, the dynamic learning parameters are obtained through online update by a neural network model, including:

[0019] A31. According to the data acquisition period, use the sliding window method to obtain the time series data within the recent preset time period to get the input features;

[0020] Among them, the time series data includes traffic volume V, visibility C, and natural illumination L outside the tunnel ext , traffic volume gradient , and the traffic volume gradient is obtained through the formula , represents the preset time interval;

[0021] A32. Perform standardized preprocessing on the input features to obtain preprocessed data;

[0022] A33. Build a recurrent neural network based on a deep learning algorithm, and set the dynamic loss function, online incremental training parameters, and initial dynamic parameters α0, β0, γ0 of the recurrent neural network to obtain an online update model; among them, the online incremental training parameters include an optimizer, the number of iterations, and the gradient norm threshold;

[0023] A34. According to the update period, input the preprocessed data into the online update model for parameter update to obtain dynamic parameter increments , , ;

[0024] A35. According to the parameter update formula , obtain the dynamic learning parameters; among them, represents the learning rate of parameter update.

[0025] Furthermore, the construction process of the tunnel carbon emission model includes:

[0026] B1-1. Collect the rated power P rated of the lighting equipment, the rated illuminance I rated , the carbon emission per unit production M, and the designed life L, and obtain the grid carbon emission factor λ(t) according to the real-time data interface of the State Grid;

[0027] B1-2. Calculate the real-time power P(t) of the lighting equipment according to the rated power, and the calculation formula is: ;

[0028] B1-3. Calculate the equivalent operation time T(t) of the lighting equipment based on the rated illuminance. The calculation formula is as follows: ; where represents the integration variable, and t0 represents the starting time of carbon emission monitoring;

[0029] B1-4. Construct the tunnel carbon emission model as: ; where i represents the lighting equipment index, n represents the number of several lighting equipment, represents the implicit carbon emission allocation coefficient of the i-th lighting equipment, which is calculated by the formula and represents the real-time carbon emission amount in the tunnel, represents the installation time.

[0030] Construct a comprehensive model including real-time operation energy consumption and implicit carbon emissions. Among them, the real-time power and equivalent operation time can accurately reflect the actual energy consumption of the equipment, and through the correlation between the rated parameters and the real-time working conditions, the refined calculation of the carbon emissions in the whole life cycle of tunnel lighting is realized, providing a basis for carbon emission analysis and optimization.

[0031] Furthermore, the use of Monte Carlo simulation to quantify the uncertainty of the tunnel carbon emission model includes:

[0032] B2-1. Define the input parameters and parameter distributions as follows: the traffic flow V(t) follows a Poisson distribution, the visibility C(t) follows a normal distribution, and the power grid carbon emission factor λ(t) follows a uniform distribution;

[0033] B2-2. Set the number of iterations of the Monte Carlo model, and in each iteration, obtain the simulated traffic flow value, simulated visibility value, and simulated power grid carbon emission factor value through sampling to obtain the sampling simulation value;

[0034] B2-3. Substitute the sampling simulation value of each iteration into the tunnel carbon emission model to obtain the simulated carbon emission value, and store it in an array to obtain the carbon emission simulation data group;

[0035] B2-4. Obtain the simulated mean AE and simulated standard deviation σ E , and use the formula to calculate E CO2 ∈[AE - 1.96σ E , AE + 1.96σ E to obtain the carbon emission confidence interval E CO2 . Among them, AE + 1.96σ E is the upper limit of the carbon emission confidence interval.

[0036] By defining the parameter distributions of traffic flow, visibility, and grid factors, and conducting multiple Monte Carlo simulations to generate carbon emission confidence intervals, the impact of input parameter fluctuations on carbon emissions is quantified, providing a risk assessment range for decision-makers. By determining the confidence interval through the simulation mean and standard deviation, the problem that traditional models cannot reflect uncertainty is solved, enhancing the robustness of decision-making.

[0037] Furthermore, the optimization of device control parameters using the reinforcement learning framework includes:

[0038] C1, setting the state space of the reinforcement learning framework as: traffic flow gradient , predicted visibility , grid carbon emission factor λ(t), and remaining carbon budget E re (t); where the calculation formula for the remaining carbon budget E re (t) is: E re (t) = E quota - E total (t), E quota represents the preset carbon emission limit, t represents the preset time interval.

[0039] C2, setting the action space of the reinforcement learning framework as: dimming level l of lighting equipment i , fan speed f of ventilation equipment j , and equipment maintenance instruction m, obtaining the action vector a(t) = [l i ∈ {l1, l2, …, l p}, f j ∈ {f1, f2, …, f q}, m ∈ {0, 1}]; where p represents the number of parameters of the dimming level, q represents the number of parameters of the fan speed, the equipment maintenance instruction is used to send a signal indicating whether the lighting equipment needs to be replaced, and when m = 1, it means sending a signal to replace the lighting equipment, and when m = 0, it means maintaining the current state of the lighting equipment;

[0040] C3, setting the reward function of the reinforcement learning framework as: ; where R1 represents the carbon emission reward function, R2 represents the comfort reward function, R3 represents the safety penalty function, w j represents the weight coefficient, determined based on historical experience;

[0041] C4, collecting real-time state space data to obtain a state vector, inputting the state vector into the reinforcement learning framework for environmental interaction and experience storage learning, and obtaining device optimization parameters;

[0042] Among them, the device optimization parameters, which are the output of the reinforcement learning framework, include the dimming level of the lighting device, the fan speed of the ventilation device, and the device maintenance instruction.

[0043] By collecting the traffic flow gradient, predicted visibility, grid carbon emission factor, and remaining carbon budget data in real time, the reinforcement learning framework can dynamically perceive the real-time changes in the internal and external environments of the tunnel. When the sudden increase in PM 2.5 value leads to a decrease in visibility, the system can quickly adjust the lighting brightness and ventilation device parameters through the updated policy network to ensure safety while optimizing carbon emissions.

[0044] Further, inputting the state vector into the reinforcement learning framework for environmental interaction and experience storage learning includes:

[0045] C41, initializing the reinforcement learning framework using a multi-objective reinforcement learning algorithm, including initializing the parameters of the policy network, value network, and target network of the reinforcement learning framework, and setting hyperparameters such as the learning rate, discount factor, and soft update coefficient;

[0046] C42, creating an experience replay buffer for storing the experiences of the agent interacting with the environment; where the agent represents the system that controls the lighting device and ventilation device in the tunnel, generates an action vector through the decision network according to the state vector, and observes the feedback of the environment;

[0047] C43, inputting the state vector s(t) into the reinforcement learning framework and generating an action vector a(t) according to the policy network;

[0048] C44, executing the action a(t) to obtain the next state vector s(t + 1), the calculation result r(t) of the reward function, and the flag d(t + 1) indicating whether to terminate learning;

[0049] C45, storing the experience tuple (s(t), a(t), r(t), s(t + 1), d(t + 1)) into the experience replay buffer;

[0050] C46, repeating C43 to C45 until the number of experience tuples in the experience replay buffer reaches the batch size threshold, randomly sampling a batch of experience tuples from the experience replay buffer to obtain an experience sample;

[0051] C47, updating the reinforcement learning framework using the experience sample and calculating the average reward value of the most recent n rounds and the average reward value of the previous n rounds , and judging the difference between and

[0052] By collecting real-time status data and interacting with the environment, reinforcement learning can accumulate experience from actual operations and gradually optimize control strategies. In equipment maintenance decision-making, the system can learn the optimal timing for replacing lighting equipment by analyzing historical data, balancing equipment lifespan and energy consumption costs.

[0053] Furthermore, the predicted visibility is obtained by predicting through a visibility prediction model constructed based on machine learning algorithms;

[0054] Among them, the construction method of the visibility prediction model is as follows:

[0055] D1, collect historical meteorological data, traffic flow, and visibility, and arrange them in chronological order to obtain time series data; among them, the meteorological data includes air humidity, temperature, wind speed, and air pressure, and the visibility C(t) is obtained through the PM 2.5 value PM(t), and the calculation formula is: ;

[0056] D2, construct a time series model based on machine learning algorithms, and define the loss function and training parameters of the model;

[0057] D3, preprocess the time series data, and divide the preprocessed data into data sets according to a preset ratio to obtain a training set, a validation set, and a test set;

[0058] D4, input the training set and the validation set into the model for iterative training and parameter update, and use the test set to evaluate whether the accuracy of the model after parameter update is greater than a preset accuracy threshold; if yes, save the model after parameter update and output it as a visibility prediction model for predicting visibility; if not, adjust the loss function and training parameters of the model, and repeat D4;

[0059] Among them, the input of the visibility prediction model includes meteorological data and traffic flow, and the output is the predicted visibility.

[0060] Furthermore, the reward function of the reinforcement learning framework includes:

[0061] The carbon emission reward function R1 is: ;

[0062] The comfort reward function R2 is: ;

[0063] The safety penalty function is R3: ; among them, represents the indicative function.

[0064] Compared with the prior art, the beneficial effects of the present invention are:

[0065] The present invention realizes the accurate calculation of the required illuminance by collecting data such as traffic flow, PM 2.5 value, and natural illuminance in real time, combining neural networks to update dynamic learning parameters online, and optimizes device parameters through a reinforcement learning framework. While ensuring driving comfort and safety, it minimizes energy consumption and carbon emissions. For example, when the predicted visibility decreases, the system automatically increases the lighting brightness and adjusts the ventilation equipment, and at the same time balances the energy-saving and safety requirements through a multi-objective reward function;

[0066] The present invention integrates real-time operating energy consumption and implicit carbon emissions of equipment, and quantifies the uncertainty of input parameters through Monte Carlo simulation to generate a carbon emission confidence interval, providing a basis for risk assessment for decision-makers, enabling the system to anticipate carbon over-standard risks in advance and adjust strategies; In addition, the reinforcement learning framework continuously optimizes the control strategy through an online learning mechanism to achieve autonomous update of device parameters. For example, it can adjust the dimming level according to the real-time change of the carbon emission factor of the power grid to ensure long-term optimality and improve the green and intelligent level of tunnel operation. Description of the Drawings

[0067] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0068] Figure 1 It is a schematic diagram of the technical process of an intelligent carbon emission analysis system for highway tunnel lighting provided by the present invention;

[0069] Figure 2 It is a schematic diagram of the working process of the demand analysis module in an intelligent carbon emission analysis system for highway tunnel lighting provided by the present invention. Detailed Embodiments

[0070] The following will clearly and completely describe the technical solutions of the present invention in combination with the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0071] Please refer to Figure 1 - Figure 2 , the first aspect embodiment of the present invention provides an intelligent carbon emission analysis system for highway tunnel lighting, including:

[0072] Demand analysis module: used to calculate the required illuminance according to the traffic flow, natural illuminance outside the tunnel, and PM inside the tunnel collected in real time 2.5Value, analyze and control the illumination brightness of several lighting devices in the tunnel;

[0073] Carbon efficiency analysis module: used to construct a tunnel carbon emission model, quantify the uncertainty of the tunnel carbon emission model using Monte Carlo simulation, and obtain a carbon emission confidence interval;

[0074] Environmental optimization module: used to optimize the device control parameters using a reinforcement learning framework when the upper limit of the carbon emission confidence interval is greater than a preset threshold, obtain optimized device parameters, and control the devices using the optimized device parameters; among them, the devices include lighting devices and ventilation devices.

[0075] It should be noted that the demand analysis module, carbon efficiency analysis module, and environmental optimization module of the present invention are in communication connection.

[0076] Based on the above technical modules, the present invention realizes the intelligent and low-carbon management of the highway tunnel lighting system. Among them, the demand analysis module collects data in real time and calculates the required illuminance, providing a control basis for the lighting devices. The carbon efficiency analysis module quantifies the carbon emissions, providing a decision-making reference for the environmental optimization module. The environmental optimization module uses a reinforcement learning framework to optimize the device control parameters according to the carbon emission situation and other state information, achieving energy conservation and emission reduction and multi-objective balance. The visibility prediction model provides predictive information for the entire system, assisting the reinforcement learning framework to make more accurate decisions.

[0077] The carbon emissions of highway tunnels are not only related to vehicle exhaust, but also include the carbon emissions generated by the energy consumption of the lighting system, such as the energy consumption of lighting devices during production and use and the carbon emissions caused by related power grid transmission losses. Therefore, in order to effectively reduce carbon emissions, in the demand analysis module, by collecting key data such as traffic flow, natural light illuminance outside the tunnel, and haze concentration inside the tunnel in real time, a dynamic demand lighting model is constructed to accurately analyze the required illuminance of several lighting devices in the tunnel under different environmental and traffic conditions. To achieve the rational use of energy, avoid energy waste caused by over-illumination, and reduce the carbon emissions of the lighting system from the source.

[0078] Specifically, in the demand analysis module, since the natural light intensity obtained at different positions inside the tunnel is different, the tunnel is first divided into several segments according to a preset distance, and then the traffic flow V(t), PM 2.5 value PM(t), and natural light illuminance L ext (t) of several segments are collected in real time to reflect the real-time changes in the internal and external environments of the tunnel;

[0079] Then, the visibility C(t) is calculated according to the haze concentration PM 2.5 value PM(t) inside the tunnel, and the calculation formula is , considering the impact of haze on visibility, provides key parameters for accurately calculating lighting requirements in the subsequent stage;

[0080] Then, based on the traffic flow V(t), visibility C(t), and natural illumination L outside the tunnel ext (t), construct a dynamic lighting model as: , and calculate the basic lighting brightness I of several segments in real time base (t); where V max represents the maximum traffic capacity of the tunnel and is obtained from the engineering design documents; L base (x) represents the sectional reference illumination, which is determined by the "Detailed Rules for Highway Tunnel Design"; α, β, γ represent dynamic learning parameters and are obtained through online update by a neural network model;

[0081] Next, set the sectional reference illumination L base (x) and the sectional correction coefficient S(x), where x represents the sectional index; then calculate the required illumination I'(t) of several segments according to the formula I'(t) = I base (t) × S(x);

[0082] Meanwhile, it is also necessary to determine whether the required illumination I'(t) of several segments is greater than or equal to the sectional reference illumination L base (x); if so, control the lighting equipment with the required illumination I'(t); if not, control the lighting equipment with the sectional reference illumination L base (x).

[0083] In one implementation, the dynamic learning parameters in the dynamic lighting model are obtained through online update by a neural network model, and the specific operation steps may include the following:

[0084] A31, according to the data acquisition period, use the sliding window method to obtain the time series data within the most recent preset time period to obtain the input features;

[0085] Among them, the time series data includes traffic flow V, visibility C, natural illumination L outside the tunnel ext , traffic flow gradient , and the traffic flow gradient is obtained through the formula and represents the preset time interval;

[0086] A32, perform standardized preprocessing on the input features to obtain the preprocessed data;

[0087] A33, construct a recurrent neural network based on the deep learning algorithm, and set the dynamic loss function, online incremental training parameters, and initial dynamic parameters α0, β0, γ0 of the recurrent neural network to obtain an online update model; among them, the online incremental training parameters include an optimizer, the number of iterations, and the gradient norm threshold;

[0088] A34. Input the preprocessed data into the online update model for parameter update according to the update period to obtain the dynamic parameter increment and and ;

[0089] A35. Obtain the dynamic learning parameter according to the parameter update formula , where represents the learning rate of parameter update, which is determined according to historical experience. Usually, the learning rate is 0.01

[0090] The dynamic learning parameter obtained through online update can enable the dynamic demand lighting model to continuously adapt to the changes in the environment and traffic conditions, continuously optimize the calculation of lighting brightness, so as to minimize the energy consumption and carbon emissions of the lighting system while meeting the tunnel lighting requirements

[0091] Suppose there is a double - hole tunnel with a length of 1000 meters, which is divided into the following four sections, and the collected data for each section is shown in Table 1

[0092] Table 1 is an example table for collecting sectional data

[0093]

[0094] Since the middle section cannot receive natural light, in practical applications, the lighting equipment in the middle section can be directly set according to the reference illuminance value. For the entrance section and the exit section, the optimal illuminance is calculated by using the dynamic lighting model and the sectional correction coefficient

[0095] At a certain moment, the traffic flow data collected is: V(t) = 800 vehicles / h; the maximum traffic capacity of the tunnel V max = 1200 vehicles / h; the natural light illuminance outside the tunnel is L ext (t) = 5000 lux; the initial values of the dynamic parameters are: α = 1.2, β = 0.8, γ = 0.5

[0096] (1) Calculation process of the entrance section

[0097] Visibility: C(t)=1 / (1 + 0.02×30)≈0.625

[0098] Basic lighting brightness: I base (t)=1.2×(800 / 1200)^0.8×e^[0.5×(1 - 0.625)]×(5000 / 80)≈65.8 lux

[0099] Required illuminance: I’(t)=65.81×[max(1.0,5000 / 1500)]≈219 lux

[0100] (2) Calculation process of the export section

[0101] Visibility: C(t)=1 / (1+0.02×40)≈0.556;

[0102] Basic lighting brightness: I base (t)=1.2×(800 / 1200)^0.8×e^[0.5×(1-0.556)]×(5000 / 90)≈58.7lux;

[0103] Required illumination: I'(t)=58.7×[min(1.5,5000 / 1000)]=88.1lux<90lux, with 90lux as the segmented lighting brightness I(t) for control.

[0104] In this embodiment, by collecting real-time traffic flow, PM 2.5 Based on data such as concentration and external light intensity, a dynamic lighting model is constructed and combined with segmented correction coefficients to precisely adjust the illumination in each tunnel zone, effectively addressing the impact of complex environments such as smog and alternating light and dark on visibility. A dynamic correction coefficient at the entrance and a natural light transition mechanism at the exit mitigate the "black hole" and "white hole" effects experienced by drivers entering and exiting the tunnel, ensuring a smooth visual adaptation process. Furthermore, a baseline illumination threshold is set, forcing the use of a fixed baseline value when the dynamically calculated illumination falls below a safety threshold, striking a balance between energy conservation and safety. A traffic flow-sensitive formula links lighting brightness with traffic density, automatically increasing illumination in high-traffic scenarios to enhance responsiveness. Furthermore, a neural network is introduced to update dynamic parameters online, continuously optimizing the model's adaptability to environmental changes. Finally, through multi-parameter fusion and full-tunnel segmented collaborative design, this solution addresses the entire safety process, from strong light adaptation at the entrance to stable lighting in the middle section and then to the natural light transition at the exit, forming a multi-dimensional collaborative protection system. While reducing carbon emissions, this solution significantly improves tunnel driving safety through five core mechanisms: dynamic perception, precise control, safety thresholds, traffic linkage, and adaptive learning.

[0105] In order to grasp the carbon emissions of highway tunnel lighting in real time and provide a scientific basis for tunnel operation and management, the carbon efficiency analysis module collects data such as the rated power, rated illumination, unit production carbon emissions, design life, and grid carbon emission factors of lighting equipment to construct a tunnel carbon emission model that includes real-time operating energy consumption and implicit carbon emissions. At the same time, Monte Carlo simulation is used to perform multiple sampling calculations to obtain carbon emission placement confidence intervals to identify potential carbon exceeding risks in advance.

[0106] Specifically, in the carbon efficiency analysis module, building a tunnel carbon emission model may include the following steps:

[0107] First, collect the key parameters of the lighting equipment, including the rated power P rated , the rated illuminance I rated , the carbon emissions per unit production M, and the design life L, and obtain the grid carbon emission factor λ(t) through the real-time data interface of the State Grid;

[0108] According to the collected data, through the formulas and calculate the real-time power P(t) and the equivalent operating time T(t) of the lighting equipment respectively; to reflect the actual energy consumption of the equipment at different times;

[0109] Based on the above calculation results, construct the tunnel carbon emission model as: ; where i represents the lighting equipment index, n represents the number of several lighting equipment, [[ID=IS=18]] represents the implicit carbon emission allocation coefficient of the i-th lighting equipment, and is calculated through the formula obtained, represents the real-time carbon emissions in the tunnel, represents the installation time. This model comprehensively considers the real-time operating energy consumption and implicit carbon emissions of the equipment, and comprehensively shows the carbon emission status of the tunnel lighting system throughout its life cycle;

[0110] In this embodiment, in order to cope with the influence of input parameter fluctuations on the carbon emission calculation results, the Monte Carlo simulation method is adopted;

[0111] First, define the input parameters and their distributions, so that the traffic flow V(t) follows a Poisson distribution, the visibility C(t) follows a normal distribution, and the grid carbon emission factor λ(t) follows a uniform distribution;

[0112] Then set the number of iterations of the Monte Carlo model, such as setting it to 1000 times; in each iteration, obtain the simulated traffic flow value, the simulated visibility value, and the simulated grid carbon emission factor value through sampling, substitute these sampled simulation values into the demand illuminance calculation formula and the tunnel carbon emission model, obtain the simulated carbon emission value, and store it in an array to form a carbon emission simulation data set;

[0113] Finally, calculate the simulated mean AE and the simulated standard deviation σ E , and use the formula E CO2 ∈[AE - 1.96σ E , AE + 1.96σ E to calculate the carbon emission confidence interval E CO2 , where AE + 1.96σ E is the upper limit of the carbon emission confidence interval;

[0114] The carbon emission confidence interval obtained through Monte Carlo simulation effectively quantifies the impact of input parameter fluctuations on carbon emissions and provides a clear risk assessment range for tunnel operation managers. When the upper limit of the confidence interval approaches or exceeds the preset threshold, managers can promptly detect potential carbon over - standard risks and take corresponding measures in advance, such as adjusting the control parameters of lighting equipment, optimizing the operation mode of ventilation equipment, etc., to avoid carbon emission over - standard and ensure the green environmental protection of tunnel operation.

[0115] Therefore, when the upper limit of the carbon emission confidence interval is greater than the preset threshold, the environmental optimization module will be activated to optimize the device control parameters using the reinforcement learning framework, obtain the optimal device control parameters, and achieve effective control of the lighting and ventilation equipment in the tunnel, reducing carbon emissions and ensuring driving comfort and safety.

[0116] In one implementation, the workflow of the environmental optimization module may include the following operation steps:

[0117] Set the state space: Take the traffic flow gradient , predicted visibility , the power grid carbon emission factor λ(t), and the remaining carbon budget E re (t) as the state space of the reinforcement learning framework; among them, the remaining carbon budget E re (t) is calculated by the formula E re (t)=E quota -E total (t), where E quota represents the preset carbon emission limit;

[0118] Set the action space: The dimming level l of the lighting equipment i , the fan speed f of the ventilation equipment j , and the equipment maintenance instruction m are obtained to get the action vector a(t)=[l i ∈{l1,l2,…,l p},f j ∈{f1,f2,…,f q},m∈{0,1}]; where p represents the number of parameters of the dimming level, q represents the number of parameters of the fan speed, and l i can be the illuminance value or the percentage of the lighting brightness value output by the demand analysis module, f j can be the fan speed value or the percentage of the initial fan speed value. The equipment maintenance instruction is used to send a signal indicating whether the lighting equipment needs to be replaced, and when m = 1, it means sending a signal to replace the lighting equipment, and when m = 0, it means maintaining the current state of the lighting equipment;

[0119] Set the reward function: Construct the reward function , where R jrepresents each reward function, w j represents the weight coefficient determined based on historical experience;

[0120] Specifically, the carbon emission reward function , which is used to encourage the reduction of carbon emissions; the comfort reward function , which aims to ensure driving comfort; the safety penalty function , where is an indicative function that imposes a penalty when the visibility C(t) < 50 to ensure driving safety;

[0121] Environmental interaction and experience storage learning:

[0122] Initializing the reinforcement learning framework: Using a multi-objective reinforcement learning algorithm to initialize the parameters of the policy network, value network, and target network of the reinforcement learning framework, and setting hyperparameters such as the learning rate, discount factor, and soft update coefficient;

[0123] Creating an experience replay buffer: Used to store the experiences of the agent (the system that controls the lighting and ventilation equipment in the tunnel) interacting with the environment. The agent generates an action vector through the decision network based on the state vector and observes the feedback from the environment;

[0124] Generating an action vector: Input the state vector s(t) into the reinforcement learning framework, and generate an action vector a(t) according to the policy network.

[0125] Executing the action and obtaining feedback: Execute the action a(t) to obtain the next state vector s(t + 1), the calculation result r(t) of the reward function, and the flag d(t + 1) indicating whether to terminate learning.

[0126] Storing the experience tuple: Store the experience tuple (s(t), a(t), r(t), s(t + 1), d(t + 1)) in the experience replay buffer;

[0127] Experience sampling and framework update: Repeat the above steps until the number of experience tuples in the experience replay buffer reaches the batch size threshold. Randomly sample a batch of experience tuples from the experience replay buffer to obtain the experience sampling. Use the experience sampling to update the reinforcement learning framework and calculate the average reward for the most recent n rounds and the average reward value for the previous n rounds ; and determine whether the difference between the two is less than the preset change threshold. If so, terminate learning and obtain the output device optimization parameters; otherwise, repeat the above process;

[0128] It should be noted that the device optimization parameters include the dimming level of the lighting equipment, the fan speed of the ventilation equipment, and the equipment maintenance instructions.

[0129] In one implementation, predicting visibility It is predicted by a visibility prediction model constructed based on a machine learning algorithm, and the construction method of the visibility prediction model may include the following steps:

[0130] D1. Collect historical meteorological data, traffic flow, and visibility, and arrange them in chronological order to obtain time series data; among them, the meteorological data includes air humidity, temperature, wind speed, and air pressure, and the visibility C(t) is obtained through the PM 2.5 value PM(t), and the calculation formula is: ;

[0131] D2. Construct a time series model based on a machine learning algorithm, and define the loss function and training parameters of the model;

[0132] D3. Preprocess the time series data, and divide the preprocessed data into data sets according to a preset ratio to obtain a training set, a validation set, and a test set;

[0133] D4. Input the training set and the validation set into the model for iterative training and parameter update, and use the test set to evaluate whether the accuracy of the model after parameter update is greater than a preset accuracy threshold; if yes, save the model after parameter update and output it as a visibility prediction model for predicting visibility; if not, adjust the loss function and training parameters of the model, and repeat D4;

[0134] Among them, the input of the visibility prediction model includes meteorological data and traffic flow, and the output is the predicted visibility.

[0135] In the reinforcement learning framework, by collecting data such as traffic flow gradient, predicted visibility, grid carbon emission factor, and remaining carbon budget in real time, a dynamic response is made to the changes in the internal and external environments of the tunnel; at the same time, through continuous interaction with the environment and storing experience for autonomous learning, experience is accumulated from actual operation, and the control strategy is gradually optimized to accurately determine the best time for replacing lighting equipment and balance the equipment life and energy consumption cost. In addition, with the help of a multi-objective reward function including carbon emission, comfort, and safety, while reducing energy consumption and carbon emission, the driving comfort and safety can be guaranteed, and the effective balance of multiple objectives can be achieved, thereby improving the comprehensive benefit of tunnel operation.

[0136] Some data in the above formula are calculated by removing the dimension and taking their numerical values. The formula is obtained by software simulation of a large amount of collected data to get a formula closest to the actual situation; the preset parameters and preset thresholds in the formula are set by those skilled in the art according to the actual situation or obtained through simulation of a large amount of data.

[0137] The working principle of the present invention:

[0138] The present invention collects traffic flow, illuminance, and PM in real time through a demand analysis module2.5 Data such as values are calculated and adjusted to obtain the required illuminance for each segment, realizing lighting adaptive control; the carbon efficiency analysis module is used to collect data on lighting equipment and the power grid, construct a carbon emission model, and use Monte Carlo simulation to obtain the confidence interval of carbon emissions, quantifying the uncertainty of carbon emissions; finally, when the upper limit of the carbon emission confidence interval exceeds the threshold, the environmental optimization module is activated. By setting the state, action space, and reward function through the reinforcement learning framework, and through data collection and interactive learning, the control parameters of the equipment are optimized, realizing precise control of the equipment and optimization of maintenance decisions, and improving the green and intelligent level of tunnel operation.

[0139] The above embodiments are only used to illustrate the technical method of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An intelligent carbon emission analysis system for highway tunnel lighting, characterized in that, including: Requirement analysis module: used to analyze and control the illumination brightness of several lighting devices in the tunnel according to the real-time collected traffic flow, natural light intensity outside the tunnel, and PM value inside the tunnel; 2.5 value, analyze and control the illumination brightness of several lighting devices in the tunnel; A carbon efficiency analysis module: used to construct a tunnel carbon emission model, quantify the uncertainty of the tunnel carbon emission model using Monte Carlo simulation, and obtain a carbon emission confidence interval; An environmental optimization module: used to optimize the device control parameters using a reinforcement learning framework when the upper limit of the carbon emission confidence interval is greater than a preset threshold, obtain device optimization parameters, and control the devices using the device optimization parameters; wherein, the devices include lighting devices and ventilation devices; The analysis and control of the lighting brightness of several lighting devices in the tunnel includes: A1. Divide the tunnel into several segments according to a preset distance, and collect the traffic flow V(t), the PM value PM(t) inside the tunnel, and the natural illuminance L 2.5 (t) outside the tunnel in real time; where t represents the current time; ext (t); A2. Calculate the visibility C(t) based on the PM value PM(t) in the tunnel. The calculation formula is as follows: 2.5 ;​ A3. Based on the traffic flow V(t), visibility C(t), and the natural light illumination L outside the tunnel ext (t), several segmented basic lighting brightnesses I base (t) are calculated. The calculation formula is: ; where α, β, and γ represent dynamic learning parameters, which are obtained through online update by a neural network model. V max represents the maximum traffic capacity of the tunnel, and L base (x) represents the segmented reference illumination, and x represents the segmented index; A4. Set the sectional correction coefficient S(x) for several sections, calculate the sectional lighting brightness according to the sectional correction coefficient, and control several lighting devices according to the sectional lighting brightness; Among them, the dynamic learning parameters are obtained through online update by a neural network model, including: A31. According to the data acquisition period, use the sliding window method to obtain time series data within a recent preset time period to obtain input features; Among them, the time series data includes traffic volume V, visibility C, and natural light intensity L outside the tunnel ext , traffic volume gradient ; A32. Perform standardized preprocessing on the input features to obtain preprocessed data; A33. Build a recurrent neural network based on a deep learning algorithm, and set the dynamic loss function, online incremental training parameters, and initial dynamic parameters α0, β0, γ0 of the recurrent neural network to obtain an online update model; wherein, the online incremental training parameters include an optimizer, the number of iterations, and a gradient norm threshold; A34. Input the preprocessed data into the online update model for parameter update according to the update period to obtain the dynamic parameter increment , , ; A35. Obtain dynamic learning parameters according to the parameter update formula , where represents the learning rate of parameter update.

2. The intelligent carbon emission analysis system for highway tunnel lighting according to claim 1, characterized in that, The traffic flow gradient is calculated as follows: ; where represents a preset time interval.

3. The intelligent carbon emission analysis system for highway tunnel lighting according to claim 2, wherein, The calculation of the sectional lighting brightness according to the sectional correction coefficient includes: A41, the required illuminance I'(t) in several segments is calculated according to the formula I'(t) = I base (t) × S(x); A42, determine whether the required illuminance I'(t) of several segments is greater than or equal to the segment reference illuminance L base (x); if yes, mark the required illuminance I'(t) as the segment illumination brightness I(t); if not, mark the segment reference illuminance L base (x) as the segment illumination brightness I(t).

4. An intelligent carbon emission analysis system for highway tunnel lighting according to claim 1, characterized in that, The quantification of the uncertainty of the tunnel carbon emission model using Monte Carlo simulation includes: B2-1. Define the input parameters and parameter distributions as follows: the traffic flow V(t) follows a Poisson distribution, the visibility C(t) follows a normal distribution, and the power grid carbon emission factor λ(t) follows a uniform distribution; B2-2. Set the number of iterations of the Monte Carlo model, and in each iteration, obtain simulated traffic flow values, simulated visibility values, and simulated power grid carbon emission factor values through sampling to obtain sampling simulation values; B2-3. Substitute the sampling simulation values of each iteration into the tunnel carbon emission model to obtain simulated carbon emission values, and store them in an array to obtain a carbon emission simulation data group; B2-4, obtaining the simulated mean value AE and the simulated standard deviation σ from the carbon emission simulation data set E , calculating E using the formula CO2 ∈[AE - 1.96σ E , AE + 1.96σ E to obtain the carbon emission confidence interval E CO2 ; among them, AE + 1.96σ E is the upper limit of the carbon emission confidence interval.

5. The intelligent carbon emission analysis system for highway tunnel lighting according to claim 3, characterized in that The optimization of the device control parameters using the reinforcement learning framework includes: C1, set the state space of the reinforcement learning framework as: traffic flow gradient , predicted visibility , power grid carbon emission factor λ(t), and remaining carbon budget E re (t); where the calculation formula for the remaining carbon budget E re (t) is: E re (t)=E quota -E total (t), E quota represents the preset carbon emission limit, t represents time, represents the preset time interval, represents the real-time carbon emission in the tunnel; Set the action space of the reinforcement learning framework as: the dimming level l of the lighting device i , the fan speed f of the ventilation device j and the equipment maintenance instruction m, to obtain the action vector a(t) = [l i ∈ {l1, l2, …, l p}, f j ∈ {f1, f2, …, f q}, m ∈ {0, 1}]; where p represents the number of parameters of the dimming level, q represents the number of parameters of the fan speed, the equipment maintenance instruction is used to send a signal indicating whether the lighting device needs to be replaced, and when m = 1, it means sending a signal to replace the lighting device, and when m = 0, it means maintaining the current state of the lighting device; C3, set the reward function of the reinforcement learning framework as follows: ; where, R1 represents the carbon emission reward function, R2 represents the comfort reward function, R3 represents the safety penalty function, and w g represents the weight coefficient, which is determined based on historical experience; C4. Collect real-time state space data to obtain a state vector, input the state vector into the reinforcement learning framework for environmental interaction and experience storage learning to obtain device optimization parameters; Among them, the device optimization parameters include the dimming level of the lighting device, the fan speed of the ventilation device, and device maintenance instructions.

6. The intelligent carbon emission analysis system for highway tunnel lighting according to claim 5, characterized in that, The input of the state vector into the reinforcement learning framework for environmental interaction and experience storage learning includes: C41. Initialize the reinforcement learning framework using a multi-objective reinforcement learning algorithm, including initializing the parameters of the policy network, value network, and target network of the reinforcement learning framework, and setting the learning rate, discount factor, and soft update coefficient; C42. Create an experience replay buffer; C43. Input the state vector s(t) into the reinforcement learning framework, and generate an action vector a(t) according to the policy network; C44. Execute the action a(t) to obtain the next state vector s(t+1), the calculation result r(t) of the reward function, and the flag d(t+1) indicating whether to terminate learning; C45. Store the experience tuple (s(t), a(t), r(t), s(t+1), d(t+1)) in the experience replay buffer; C46. Repeat C43 to C45 until the number of experience tuples in the experience replay buffer reaches the batch size threshold. Then, randomly sample a batch of experience tuples from the experience replay buffer to obtain the experience sampling; C47. Update the reinforcement learning framework using experience sampling and calculate the average reward value for the most recent n rounds and the average reward value for the previous n rounds , and determine whether the difference between is less than a preset change threshold; if so, terminate the learning and obtain the output result; if not, repeat C43 to C46 7. An intelligent carbon emission analysis system for highway tunnel lighting according to claim 6, characterized in that, The predicted visibility is obtained by predicting through a visibility prediction model constructed based on a machine learning algorithm; Among them, the construction method of the visibility prediction model is as follows: D1 collects historical meteorological data, traffic flow, and visibility, and arranges them in chronological order to obtain time-series data; among them, the meteorological data includes air humidity, temperature, wind speed, and air pressure, and the visibility C(t) is obtained through the PM 2.5 value PM(t), and the calculation formula is: ; D2. Construct a time series model based on a machine learning algorithm and define the loss function and training parameters of the model; D3. Preprocess the time series data and divide the preprocessed data into a data set according to a preset ratio to obtain a training set, a validation set, and a test set; D4. Input the training set and the validation set into the model for iterative training and parameter update. Use the test set to evaluate whether the accuracy of the model after parameter update is greater than the preset accuracy threshold. If so, save the model after parameter update and output it as the visibility prediction model for predicting visibility. If not, adjust the loss function and training parameters of the model and repeat D4; Among them, the input of the visibility prediction model includes meteorological data and traffic flow, and the output is the predicted visibility.

Images