A method and system for intelligent dispersion control of tunnel contamination blobs based on edge prediction
By dividing ventilation responsibility zones in the tunnel and using an edge prediction-driven approach, accurate identification and dynamic tracking of contamination blobs were achieved, solving the problems of time lag and energy waste in existing technologies, and improving the efficiency and lifespan of tunnel ventilation systems.
Patent Information
- Application Number
- CN202611327735.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-31
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2046-08-31
AI Technical Summary
Existing tunnel pollution cloud control technologies suffer from time lag, energy waste, and equipment wear, making it impossible to achieve accurate identification, dynamic tracking, and proactive dispersion of pollution clouds.
An edge prediction-driven approach is adopted to divide the tunnel into multiple ventilation responsibility zones. Each zone is equipped with an edge computing unit and a variable frequency fan unit. A lightweight time-series prediction network is used to identify pollution blobs in real time and dynamically adjust the fan frequency to achieve accurate identification and advance dispersion of pollution blobs.
It enables accurate identification and dynamic tracking of pollution blobs, reduces ineffective fan operation in clean zones, lowers energy consumption and equipment wear, and improves ventilation efficiency.
Smart Images

Figure CN122815936B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of ventilation control technology for highway tunnels, specifically to an intelligent dispersion control method and system for tunnel pollution clumps based on edge prediction. Background Technology
[0002] As a crucial node in the highway transportation network, the operational safety and air quality control of highway tunnels are directly related to the safety of people's lives and property. With the increasing density of mountain highway networks, the ventilation problem of long tunnels is becoming increasingly prominent.
[0003] Highway tunnels are semi-enclosed tubular spaces where pollutants such as carbon monoxide (CO), nitrogen oxides (NOx), inhalable particulate matter (PM10), and carbon dioxide (CO2) emitted by vehicles accumulate rapidly, forming "pollution plumes" with significant spatial distribution characteristics and longitudinal movement trends. If these pollutants cannot be discharged from the tunnel in a timely and accurate manner, the air quality inside the tunnel will deteriorate, seriously threatening the health of drivers and passengers and driving safety.
[0004] Currently, pollution plume control technology for highway tunnels has undergone the following three generations of development: The first generation, based on the "Design Specifications for Highway Tunnels" (JTG 3370.1-2018), involves deploying CO sensors, visibility detectors (VI), and wind speed detectors within the tunnel. When the detected values exceed a fixed safety threshold, the jet fans are activated or deactivated according to a preset tiered strategy. This scheme is simple in structure and low in implementation cost, and is currently the control method used by the vast majority of operating tunnels in China. However, fixed thresholds cannot adapt to changes in dynamic factors such as traffic flow density, vehicle speed, and ambient temperature and humidity. In congested conditions, they may be too lenient, threatening safety, while in free-flowing conditions, they may be too conservative, leading to unnecessary ventilation energy consumption.
[0005] The second generation, based on fixed threshold control, introduces feedforward signals from traffic volume detectors, combined with feedback signals from CO / VI detection values, to adjust the number of operating fans and the frequency converter frequency via PID controllers or fuzzy controllers. This scheme achieves on-demand ventilation to some extent, but because the control logic is still based on the measured values at the current moment, it is essentially still a passive response control. Tunnel ventilation systems exhibit significant time lag characteristics. From the change in pollutant concentration to sensor detection, then to controller response, and finally to fan operation changing the airflow field, the entire closed-loop response time often lasts from tens of seconds to several minutes. During this period, the pollution plume may have migrated hundreds of meters, causing a serious misalignment between control commands and the actual location of the pollution plume.
[0006] In-depth analysis revealed the following systemic flaws in the aforementioned technical solution: First, existing technologies all use the concentration at a single point or several discrete points as the basis for control triggering. However, as mentioned earlier, the spatial distribution of pollutants in tunnels is not uniform, but rather forms pollution blobs with clear boundaries, centers of gravity, and movement speeds. When the pollution blobs only cover a local section of the tunnel, traditional solutions still trigger the synchronous operation of fans along the entire line or a large section, resulting in the ineffective idling of fans in clean sections and extremely serious energy waste.
[0007] Second, existing pollution plume control systems typically divide tunnels into fixed physical sections (e.g., control zones every 500 m or 1000 m) for control responsibility. However, pollution plumes are dynamic entities that continuously move with traffic flow and natural wind. When a pollution plume migrates from one physical section to the next, upstream fans shut down due to a drop in concentration, while downstream fans start up due to an increase in concentration, resulting in frequent "start-stop-start" oscillations. This not only increases equipment wear and shortens fan lifespan but also leads to turbulent airflow and reduced pollutant removal efficiency.
[0008] Third, existing control schemes all use the exceeding of pollutant concentration as the action signal, which is a typical passive response mode. Due to the time lag characteristic of tunnel ventilation systems, by the time the sensor detects that the concentration exceeds the standard, the pollution cloud has often already spread to a wider area or migrated to downstream areas. At this time, starting the fan can only "chase after the pollution" rather than "disperse it before it occurs." This lagging control mode not only weakens the ventilation effect, but also causes a lot of ineffective energy consumption because the fan continues to run after the pollution peak. Summary of the Invention
[0009] The technical problem to be solved by the present invention is to provide a method and system for intelligent dispersion control of tunnel contaminant clumps based on edge prediction, so as to achieve accurate identification, dynamic tracking and advanced dispersion of contaminant clumps.
[0010] The technical solution adopted by this invention to solve its technical problem is an intelligent dispersion control method for tunnel contamination clumps based on edge prediction, comprising the following steps: S1: Divide the entire tunnel length longitudinally into... Each ventilation responsibility zone is equipped with an edge computing unit and a variable frequency fan unit. Adjacent zones have a physical overlap section, with the overlap section length being... Occupying the nominal length of a single interval of Number of intervals
[0011] in, The total length of the tunnel, Indicates rounding up; S2: The edge computing unit collects multi-source sensor data within this area and extracts environmental feature factors. And based on environmental characteristic factors and smoothed prediction error feedback correction terms. Real-time calculation of dynamic security threshold ; S3: Deploy a lightweight time-series prediction network in the edge computing unit, based on the current time. Four-dimensional feature vectors The input is S(t), where S(t) is the current concentration, V(t) is the average vehicle speed, C(t) is the traffic flow, and ΔH(t) is the humidity deviation. ; The hidden state is obtained after temporal feature extraction of the hidden layer. The output layer calculates the future preset time window respectively. Predicted pollutant concentrations and prediction confidence ,in:
[0012]
[0013] in Output for hidden layer 3D hidden state vector; and These are the weight vector and bias of the concentration prediction output, respectively; and These are the weight vector and bias of the confidence score output, respectively; It is the Sigmoid activation function. ; S4: The edge computing unit performs pollution cluster identification based on the predicted data of its own interval and adjacent intervals, marking continuous mileage intervals where the predicted concentration exceeds the dynamic safety threshold as pollution clusters, and estimating... The coordinates of the center of gravity of the pollution cloud at any moment and The migration speed of the pollution clump ; S5: When the center of gravity of the pollution plume enters the overlapping section of the adjacent interval, the dynamic migration and relay control of the responsibility area is triggered, the upstream interval smoothly reduces the frequency, the downstream interval increases the frequency in advance, and the even downstream interval is warmed up and ready for standby.
[0014] Furthermore, the dynamic security threshold mentioned in step S2 Calculate using the following formula:
[0015] in, The current moment; These are the basic safe concentration limits specified by national standards; This represents the total number of environmental characteristic factors. The serial number of the environmental characteristic factor; For the first Individual environmental characteristic factors The normalized value at time , ; For the first The weight coefficients of each environmental characteristic factor are signed real numbers that satisfy the L1 norm constraint. ; This is the error penalty coefficient, with a range of values. ; for The prediction error feedback correction term after time-smoothing; calculated Satisfy physical constraints If the value exceeds the range, it will be truncated to the boundary value.
[0016] Furthermore, the smoothed prediction error feedback correction term From the original prediction error After first-order hysteresis filtering, the following is obtained:
[0017]
[0018] in, The current moment; This refers to the measurement range of the concentration sensor. The total number of prediction periods involved in error calculation; This refers to the sequence number of the prediction period; For the most recent In the prediction period, the _th The time corresponding to each cycle; To predict the network in The predicted pollutant concentration output at any given time; for Measured values of pollutant concentration collected by the time sensor; These are the filter coefficients. ; for The original prediction error at time; This represents the prediction error after smoothing from the previous moment.
[0019] Furthermore, in step S3, a tiered response is executed based on the predicted confidence level: when Execute the complete relay control logic at the time; when At that time, a prediction-measurement weighted fusion regulation was adopted, and the fusion value was... ,like If so, execute the full relay control; otherwise, remain in standby mode. Predictive control is disabled and downgraded to traditional threshold control based on the current measured value; whereby... For high confidence threshold, The low confidence threshold, and .
[0020] Furthermore, the pollution cluster identification in step S4 includes: predicting the concentration of each interval. Dynamic safety threshold corresponding to the interval Compare and label the set of predicted out-of-range intervals. ,in For interval index, set A continuous interval of medium mileage constitutes a pollution plume; the centroid coordinates of the pollution plume are estimated by the following formula:
[0021] in The interval number, To predict the set of intervals exceeding the standard, For the first The center mileage coordinates of each interval; the migration speed is estimated by the following formula:
[0022] in, for The coordinates of the center of gravity of the pollution cluster at all times. For the calculation period; combined with The symbols and traffic flow direction determine the migration trend of the pollution plume and the downstream target area it will enter.
[0023] Furthermore, the triggering condition for the dynamic migration of the responsibility area in step S5 is:
[0024] in, Mileage coordinates of the physical boundary points between adjacent intervals; duration of relay control link maintenance.
[0025] in, To maintain the duration of the relay link, The vertical length of the pollution plume is used as the condition for the relay to be terminated when the center of gravity of the pollution plume completely leaves the current main control zone, and the predicted concentration of each zone is lower than the dynamic safety threshold of the corresponding zone in several consecutive prediction periods.
[0026] An edge prediction-driven intelligent dispersion control system for tunnel contamination blobs, used to execute an edge prediction-driven intelligent dispersion control method for tunnel contamination blobs, characterized in that it includes: The physical sensing layer consists of environmental sensor arrays and traffic flow detection equipment deployed longitudinally along the tunnel to collect data on pollutant concentration, wind speed, temperature, humidity, and traffic flow parameters. The edge computing layer divides the entire tunnel length longitudinally into... Each ventilation responsibility zone is equipped with an edge computing unit, a safety control unit, and a variable frequency fan unit. Adjacent zones have physical overlap sections. Each zone adopts a heterogeneous architecture with dual computing and safety channels. The main computing channel is an edge computing unit, and the safety channel is an industrial-grade programmable logic controller or embedded real-time processor independent of the main computing unit. Each zone deploys two sets of edge computing units, one main and one backup, and supports... Automatic fault switching; In the communication network layer, adjacent edge computing units are interconnected through an industrial Ethernet ring network, and the edge computing units are connected to the cloud platform.
[0027] The beneficial effects of this invention are as follows: By employing edge prediction and pollution plume identification technologies, this invention upgrades the target of ventilation control from traditional single-point concentration thresholds to pollution plumes with spatial centroids and migration velocities. The system only implements targeted dispersion for the local area where the pollution plume is located and its adjacent areas, eliminating the need for fans in clean areas to idle. This invention sets up physical overlap sections between adjacent areas and designs dynamic migration and relay control logic for the responsibility area. When the center of gravity of the pollution plume enters the overlap section, the upstream section smoothly reduces the frequency, the downstream section increases the frequency in advance, and the even downstream section preheats and goes into standby mode, forming a relay dispersion chain. This invention deploys a lightweight time-series prediction network on the edge side, which can output predicted pollutant concentration values for future periods in advance, and combined with a prediction confidence-level response mechanism, initiates proactive control before the pollution plume arrives or exceeds the standard. Attached Figure Description
[0028] Figure 1 This is a flowchart of the present invention. Detailed Implementation
[0029] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.
[0030] like Figure 1 As shown, the present invention provides an intelligent dispersion control method for tunnel contamination clumps based on edge prediction, comprising the following steps: S1: Delineation of ventilation responsibility zones and deployment of edge nodes.
[0031] S101: Delineation of ventilation responsibility zones.
[0032] The tunnel's entire length is divided longitudinally into N ventilation responsibility zones. The core basis for this zone division is that the net coverage length of a single zone is determined comprehensively based on the tunnel's design wind speed, the effective operating distance of the jet fans, and traffic flow characteristics, typically ranging from 400m to 800m, with 500m to 650m being the preferred range.
[0033] Adjacent intervals are physically overlapped, and the overlap length is... Occupying the nominal length of a single interval The frequency of the wind turbines in the upstream section gradually decreases, while the frequency of the wind turbines in the downstream section gradually increases. The coordinated operation of the wind turbines in the overlapping section can eliminate the low-speed airflow zone (wind speed <1.5m / s) at the junction, ensuring the uniformity and continuity of the longitudinal wind speed distribution.
[0034] Number of intervals Calculate using the following formula:
[0035] in This represents the total length of the tunnel. In actual deployment, the smallest integer not less than the calculated value should be used, and adjustments should be made based on the actual situation. Value back substitution adjustment or This ensures full coverage and eliminates redundancy at the end points.
[0036] S102: Deployment of a dual-channel heterogeneous architecture for secure computing.
[0037] Each ventilation responsibility zone is configured with an independent edge node, employing a compute-safe dual-channel heterogeneous architecture: The main computing channel utilizes an edge processor with AI inference acceleration capabilities (such as an edge AI chip based on an ARM Cortex-A series multi-core processor or GPU acceleration) to run core algorithms such as lightweight time-series prediction networks, dynamic threshold calculation, contaminant cluster identification, and relay decision-making. It supports INT8 / INT16 model quantization inference with a single inference time of ≤15ms. It is equipped with local solid-state storage (≥128GB) for runtime logs and data caching during failures.
[0038] The safety channel employs an industrial-grade programmable logic controller (PLC) or an embedded real-time processor (such as a Cortex-M7 architecture microcontroller), operating independently of the main computing channel. It is responsible for hardware-level watchdog monitoring, safety interlock logic, and relay direct-drive control in fail-safe mode. It interacts with the edge computing unit via an industrial fieldbus with millisecond-level heartbeats (heartbeat cycle ≤ 50ms). If no heartbeat is received from the edge computing unit for three consecutive heartbeat cycles, a main computing channel failure is determined, and the system automatically takes over control of the underlying wind turbine within 50ms.
[0039] S103: Primary and backup redundancy deployment.
[0040] Each ventilation responsibility area is equipped with two edge computing units, a primary and a backup, connected via an industrial Ethernet ring network that supports a fast failover protocol (failover time <50ms). During normal operation, the primary edge computing unit executes control logic, while the backup edge computing unit is in hot standby mode, synchronizing in real-time with the primary unit's operating status, prediction model parameters, and current relay link information. When the primary computing unit fails, the safety control unit triggers a hardware switching signal, transferring control to the backup computing unit. The backup computing unit completes state synchronization and inherits all current operating contexts within 500ms.
[0041] S104: Communication network deployment.
[0042] Adjacent edge computing units are interconnected via an industrial Ethernet ring network for relay control command broadcasting, pollution plume migration information sharing, and status synchronization, with a communication latency of <5ms. Cloud-edge communication connects to the cloud platform via a wired network or 4G / 5G wireless network, employing a lightweight message communication protocol (such as MQTT, QoS1 level) and supporting breakpoint resumption.
[0043] S2: Multi-source environmental perception and dynamic safety threshold calculation.
[0044] S201 Multi-source data acquisition and preprocessing.
[0045] The edge computing unit collects data from all sensors in the local area at a preset acquisition cycle and performs the following preprocessing operations: Outlier removal: The 3σ criterion (three standard deviations) is used to detect outliers in the most recent M periods (e.g., M=12, i.e., 60 seconds). Outliers exceeding the standard deviation are removed from the data. Data within the range is marked as invalid and replaced with the most recent valid value.
[0046] Moving average filtering: Apply a moving average filter with a window length of K (e.g., K=6, i.e., 30 seconds) to the effective data to eliminate high-frequency noise.
[0047] Feature Calculation: Extract the following environmental feature factors from the filtered data: Vehicle speed risk factor The lower the speed, the higher the risk. V The maximum design speed for a tunnel is determined by the tunnel design standards (such as 80 km / h or 60 km / h) and is a fixed constant.
[0048] Traffic density factor The greater the traffic volume, the higher the risk. The maximum designed traffic flow in this section is determined based on the tunnel's designed capacity (such as the maximum hourly flow under the designed service level) and is a fixed constant.
[0049] Humidity deviation factor The further the humidity deviates from the baseline, the worse the diffusion conditions. The reference relative humidity is a fixed constant, taken as the humidity value suitable for pollutant diffusion (e.g., 60%RH).
[0050] Natural wind headwind factor Only the backwind component increases the ventilation burden. For headwind weight, when (When facing headwinds) , indicating the intensity of the headwind; when (When there is a tailwind or calm wind) Taking 0 indicates no headwind burden; The maximum headwind speed is taken as the maximum longitudinal wind speed considered in the tunnel design (e.g., 10 m / s), which is a fixed constant.
[0051] S202 Dynamic Security Threshold Calculation.
[0052] The edge computing unit calculates the dynamic security threshold in real time using the following formula. :
[0053] in, The basic safe concentration limits specified by national standards (e.g., CO is 150 ppm, visibility index VI is 0.0075 m) are the limits for concentrations that meet the national standards. -1 ), which is the benchmark value for dynamic safety threshold adjustment; Let be the normalized value of the i-th environmental characteristic factor at time t. Normalization allows characteristic factors of different dimensions to be calculated in the same formula with weights. Let be the weight coefficient of the i-th feature factor, and be a signed real number. When When the risk factor is higher, the threshold is lower (i.e., the safety requirements are more stringent); when At this point, the larger the factor, the higher the threshold (i.e., relatively relaxed). The weighting coefficients satisfy the L1 norm constraint. To prevent extreme weights from causing threshold distortion; This is the error penalty coefficient, ranging from 0.1 to 0.5, used to control the strength of the correction of the threshold by the prediction error feedback; This is the smoothed prediction error feedback correction term. When the prediction accuracy is high, this value approaches 0, and the threshold is mainly determined by environmental characteristics. When the prediction error increases, the threshold automatically tightens.
[0054] S203 Prediction Error Feedback Correction Term.
[0055] Original prediction error Defined as the normalized root mean square error (NRMSE) between predicted and measured values over the most recent N prediction periods:
[0056] in The range of the concentration sensor is used to normalize the error to a dimensionless value; To predict the network in The predicted pollutant concentration output at any given time; for The measured value of pollutant concentration collected by the time sensor.
[0057] To suppress closed-loop feedback oscillation, Apply first-order hysteresis filter:
[0058] Filter coefficients The smaller value (0.1 to 0.3) makes the response of the error correction term smoother, avoiding drastic fluctuations in the threshold with instantaneous prediction error.
[0059] S204 Physical constraints are applied.
[0060] Calculated Physical constraints must be met If the value exceeds the specified range, it will be truncated to the boundary value. This constraint ensures that the dynamic safety threshold does not deviate too far from the national standard limit, preventing threshold distortion under extreme operating conditions from causing safety risks or excessive energy consumption.
[0061] S3: Lightweight timing prediction on the edge side.
[0062] S301: Lightweight temporal prediction network architecture.
[0063] Edge computing units deploy lightweight time-series prediction networks to predict current time series data. Four-dimensional feature vectors The input is defined as follows: S(t) is the pollutant concentration measured by the sensor at the current time, V(t) is the average speed of vehicles in this interval, C(t) is the number of vehicles passing through this interval per unit time, and ΔH(t) is the deviation between the current humidity and the reference humidity. The hidden layer consists of 2-3 temporal feature extraction layers, which can employ gated recurrent units (GRU), one-dimensional temporal convolutional networks (TCN), or lightweight attention mechanisms. The output of the hidden layer is... 3D hidden state vector It is a compressed representation of the input feature vector after time-series feature extraction, which includes the current and historical working condition information.
[0064] The output layer calculates two outputs respectively: Predicted concentration values:
[0065] in and These are the weight vector and bias of the concentration prediction output, respectively; The ReLU activation function ensures that the concentration prediction value is always non-negative, which is physically reasonable.
[0066] Prediction confidence level:
[0067] in and These are the weight vector and bias of the confidence score output, respectively; The Sigmoid activation function maps the output to... Interval. The closer it is to 1, the more reliable the network is for the current prediction result—that is, the current input features are highly similar to the samples in the training data that make accurate predictions; The closer it is to 0, the more complex the current working conditions, the greater the difference between the data characteristics and historical patterns, or the stronger the uncertainty of the prediction.
[0068] The model has ≤50K parameters, supports INT8 static quantization deployment, and a single inference time of ≤15ms.
[0069] S302: Ranked response mechanism for prediction confidence.
[0070] Based on prediction confidence Depending on the level, the system adopts a three-level hierarchical response strategy: High confidence level ( (e.g., 0.85): Execute the complete relay control logic. At this point, the network has a high degree of confidence in the prediction results and can completely rely on the predicted values for proactive control.
[0071] Medium confidence level ( (e.g., 0.6–0.85): Predictive-measured weighted fusion regulation is employed. Fusion value ,in This represents the current sensor-measured concentration value. The fusion strategy strikes a balance between prediction and measurement; the higher the confidence level, the greater the weight of the predicted value; the lower the confidence level, the greater the weight of the measured value. If... If it does, then execute the full relay control; otherwise, remain in standby mode.
[0072] Low confidence level ( (e.g., 0.6): Predictive control is disabled, and the system is downgraded to traditional threshold control based on current measured values. In this case, the network lacks confidence in the prediction results and adopts a conservative measured control strategy to avoid false triggering.
[0073] S4: Pollution clump identification and migration trend analysis.
[0074] S401: Pollution clump identification.
[0075] The edge computing unit executes contaminant plume identification logic based on the predicted data of its own interval and adjacent intervals: Predicted concentrations for each interval Dynamic safety threshold corresponding to the interval Compare and label the set of predicted out-of-range intervals. ,in This is the interval index.
[0076] To prevent false triggering due to single instantaneous noise, a persistence criterion is set: with the prediction period (5 seconds) as the sampling interval, at least two consecutive inference results must meet the exceeding condition before it can be officially marked as a valid exceeding interval. A single instantaneous exceeding is considered noise and will not be responded to.
[0077] gather A continuous section within a medium mileage constitutes a pollution cluster. If there are multiple discontinuous sets of sections exceeding the standard, they are identified as multiple independent pollution clusters. For example, there may be two pollution accumulation areas within a tunnel caused by different traffic flow segments.
[0078] S402: Spatial center of gravity estimation.
[0079] For the identified contamination clumps, the edge computing unit estimates their value using the following formula. Coordinates of the center of gravity of the pollution plume at any given time:
[0080] in For the first The center mileage coordinates of each interval are used. A weighted average is then calculated based on the predicted concentration values of each exceeding interval to obtain the centroid position of the pollution plume along the tunnel's longitudinal direction. Intervals with higher predicted concentrations contribute more to the centroid position.
[0081] S403: Migration speed estimation.
[0082] The migration speed of the pollution plume can be estimated using the following formula:
[0083] in To calculate the period (e.g., 10 seconds), calculate the displacement of the center of gravity of the pollution plume per unit time. A positive value indicates that the pollution plume is moving downstream, while a negative value indicates that it is moving upstream (in cases of headwinds or reversed traffic flow).
[0084] Combination The symbols and traffic flow direction are used to determine the migration trend of the pollution plume and the downstream target area it will enter. For example, in a one-way tunnel, if... If the direction is consistent with the traffic flow, then the pollution plume is migrating downstream, and the next target area is... .
[0085] S5 responsibility area dynamic migration and relay control.
[0086] S501: Triggering conditions for dynamic migration of responsibility area.
[0087] When the centroid coordinates of the contamination plume are identified When entering the overlapping segment of adjacent intervals, dynamic migration of the responsibility area is triggered. The triggering condition is:
[0088] in These are the mileage coordinates of the physical boundary point between the upstream and downstream sections. The physical meaning of this condition is that the center of gravity of the pollution plume has entered the overlapping area at the boundary between the two sections (centered on the boundary point, with coordinates on both sides). (within the specified range), at this point, it is necessary to start the relay of wind turbines in the upstream and downstream sections to ensure a smooth transition.
[0089] S502: Relay control execution logic.
[0090] Transfer of control: Upstream section (original main control): during the transition time Within 30 to 60 seconds, smoothly reduce the fan frequency from the target value to the standby frequency (e.g., 20Hz) and gradually withdraw from main ventilation. Downstream section (new main control): Start the fans to the target frequency in advance, take the lead in dispersing the pollution cloud, and assume the main responsibility for ventilation; Further downstream (preheating zone): Start up to standby frequency in advance to prepare for the next relay and shorten response time.
[0091] Relay link maintenance: Duration of relay control link Determined by the migration speed and longitudinal length of the contaminant plume:
[0092] in The longitudinal length of the pollution plume (i.e., the set of intervals exceeding the standard) (Covered mileage range). Relay control needs to continue until the contamination plume has completely passed through the overlapping section and entered the core area of the downstream section. The duration of maintenance is equal to the longitudinal length of the contamination plume plus the length of the overlapping section divided by the migration speed. Simulation results show that the typical maintenance duration is 90–150 seconds.
[0093] Relay terminated: The relay control link will be terminated when all of the following conditions are met: The center of gravity of the pollution plume has completely moved away from the current main control area (i.e. (It has already crossed the downstream boundary of the current main control area). In subsequent consecutive prediction periods (e.g., M=6, i.e., 30 seconds), the predicted concentration in each interval was lower than the dynamic safety threshold of the corresponding interval. The deviations between the measured and predicted concentrations in each interval are within a reasonable range (verifying the accuracy of the prediction).
[0094] An edge prediction-driven intelligent dispersion control system for tunnel contamination blobs, used to execute an edge prediction-driven intelligent dispersion control method for tunnel contamination blobs, characterized in that it includes: The physical sensing layer consists of environmental sensor arrays and traffic flow detection equipment deployed longitudinally along the tunnel to collect data on pollutant concentration, wind speed, temperature and humidity, and traffic flow parameters, including vehicle speed and traffic density.
[0095] The edge computing layer divides the entire tunnel length longitudinally into N ventilation responsibility zones. Each zone is equipped with an edge computing unit, a safety control unit, and a variable frequency fan unit. Adjacent zones have physical overlap sections. Each zone adopts a heterogeneous architecture with dual computing and safety channels. The main computing channel is the edge computing unit, and the safety channel is an industrial-grade programmable logic controller or embedded real-time processor independent of the main computing unit. Each zone deploys two sets of edge computing units, one main and one backup, and supports automatic fault switching in less than 50ms. In the communication network layer, adjacent edge computing units are interconnected through an industrial Ethernet ring network, and the edge computing units are connected to the cloud platform.
[0096] Example
[0097] A two-way highway tunnel (3,600m in length) is divided into 6 ventilation responsibility zones according to step S101 (each zone is about 600m long, with an overlap of about 210m and an overlap rate of 35%). Each zone is equipped with an edge computing unit, a safety control unit, primary and backup redundancy, and a communication network according to steps S102 to S104.
[0098] Execution process: t=0s: The edge computing unit of interval 3 collects and preprocesses sensor data according to step S201, and detects that the current CO concentration is 132ppm, which is lower than the dynamic safety threshold calculated according to step S202. ppm, the fan is in standby mode (20Hz).
[0099] t=5s: The edge computing unit of interval 3 executes the lightweight temporal prediction network inference according to step S301. Using the feature vector at the current time... As input, after temporal feature extraction in the hidden layer, the output is the predicted value for the next 60 seconds. ppm, exceeding the dynamic safety threshold ppm, Following step S302, the system determines the confidence level to be high and executes the complete relay control logic. The system marks interval 3 as the predicted overshoot interval.
[0100] t=10s: Interval 4 also showed predicted overshoot ( ppm> (ppm), the predicted value in interval 2 is normal. Follow step S401 to determine the set of intervals exceeding the standard. This forms a pollution plume covering an area of approximately 1,200 meters.
[0101] t=15s: Estimate the centroid of the contamination plume using the spatial centroid formula in step S402. m (located slightly downstream of the center of interval 3); estimated according to the migration speed formula in step S403. The pollution plume is moving downstream in direction 4 (approximately 58 km / h, in the same direction as the traffic flow) at a speed of m / s.
[0102] t=20s: The center of gravity of the pollution plume enters the overlapping segment of interval 3 and interval 4. m, m, If m < 105m, the trigger condition is met according to step S501, and the dynamic migration of the responsibility area is initiated.
[0103] t=20~65s (transition period 45s): Perform the master control transfer according to step S502. The fan in section 3 smoothly drops from 38Hz to 22Hz; the fan in section 4 rises from standby (20Hz) to 36Hz; the fan in section 5 starts up in advance to 22Hz for preheating.
[0104] t=65s: The relay link is established, and interval 4 becomes the new master control interval. Calculate the relay maintenance duration according to step S502. It actually lasted for about 110 seconds.
[0105] t=65~175s: In interval 4, the pollution cloud is dispersed, while in interval 3, low-frequency operation is maintained to assist the transition of airflow.
[0106] t=175s: The center of gravity of the pollution plume has completely entered the central region of interval 4 ( m > m), and the predicted concentration in each interval is lower than the dynamic safety threshold of the corresponding interval within the next 6 consecutive prediction cycles (30 seconds). According to step S502, all the conditions for relay release are met, and the relay link is released.
[0107] t=180s: The fan in section 3 stops, while section 4 continues to operate. Since the center of gravity of the pollution cloud (2250m) has not yet entered the trigger zone of the overlapping section (2295m) of section 4 and section 5, the system determines that there is no need for the output of section 5 at present, so it is switched back from the preheating frequency to the standby frequency to reduce ineffective energy consumption.
[0108] The embodiments described herein are preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape, and principle of the present invention should be covered within the scope of protection of the present invention.
Claims
1. A method for intelligent dispersion control of tunnel contamination blobs based on edge prediction, characterized in that, Includes the following steps: S1: Divide the entire tunnel length longitudinally into... Each ventilation responsibility zone is equipped with an edge computing unit and a variable frequency fan unit. Adjacent zones have a physical overlap section, with the overlap section length being... Occupying the nominal length of a single interval of Number of intervals in, The total length of the tunnel, Indicates rounding up; S2: The edge computing unit collects multi-source sensor data within this area and extracts environmental feature factors. And based on environmental characteristic factors and smoothed prediction error feedback correction terms. Real-time calculation of dynamic security threshold The dynamic security threshold Calculate using the following formula: in, The current moment; These are the basic safe concentration limits specified by national standards; This represents the total number of environmental characteristic factors. The serial number of the environmental characteristic factor; For the first Individual environmental characteristic factors The normalized value at time , ; For the first The weight coefficients of each environmental characteristic factor are signed real numbers that satisfy the L1 norm constraint. ; This is the error penalty coefficient, with a range of values. ; for The prediction error feedback correction term after time-smoothing; calculated Satisfy physical constraints If the value exceeds the range, it will be truncated to the boundary value; Smoothed prediction error feedback correction term From the original prediction error After first-order hysteresis filtering, the following is obtained: in, The current moment; This refers to the measurement range of the concentration sensor. The total number of prediction periods involved in error calculation; This refers to the sequence number of the prediction period; For the most recent In the prediction period, the _th The time corresponding to each cycle; To predict the network in The predicted pollutant concentration output at any given time; for Measured values of pollutant concentration collected by the time sensor; These are the filter coefficients. ; for The original prediction error at time; This represents the smoothed prediction error from the previous moment. S3: Deploy a lightweight time-series prediction network in the edge computing unit, based on the current time. Four-dimensional feature vectors The input is S(t), where S(t) is the current concentration, V(t) is the average vehicle speed, C(t) is the traffic flow, and ΔH(t) is the humidity deviation; the hidden state is obtained after temporal feature extraction of the hidden layer. The output layer calculates the future preset time window respectively. Predicted pollutant concentrations and prediction confidence ,in: in Output for hidden layer 3D hidden state vector; and These are the weight vector and bias of the concentration prediction output, respectively; and These are the weight vector and bias of the confidence score output, respectively; It is the Sigmoid activation function. ; S4: The edge computing unit performs pollution cluster identification based on the predicted data of its own interval and adjacent intervals, marking continuous mileage intervals where the predicted concentration exceeds the dynamic safety threshold as pollution clusters, and estimating... The coordinates of the center of gravity of the pollution cloud at any moment and The migration speed of the pollution clump ; S5: When the center of gravity of the pollution plume enters the overlapping section of the adjacent interval, the dynamic migration and relay control of the responsibility area is triggered, the upstream interval smoothly reduces the frequency, the downstream interval increases the frequency in advance, and the even downstream interval is warmed up and ready for standby.
2. The intelligent dispersion control method for tunnel contamination clumps based on edge prediction as described in claim 1, characterized in that, In step S3, a tiered response is executed based on the prediction confidence level: when Execute the complete relay control logic at the time; when At that time, a prediction-measurement weighted fusion regulation was adopted, and the fusion value was... ,like If so, execute the full relay control; otherwise, remain in standby mode. Predictive control is disabled and downgraded to traditional threshold control based on the current measured value; whereby... For high confidence threshold, The low confidence threshold, and .
3. The intelligent dispersion control method for tunnel contamination blobs based on edge prediction as described in claim 1, characterized in that, The pollution cluster identification in step S4 includes: predicting the concentration of each interval. Dynamic safety threshold corresponding to the interval Compare and label the set of predicted out-of-range intervals. ,in For interval index, set A continuous interval of medium mileage constitutes a pollution plume; the centroid coordinates of the pollution plume are estimated by the following formula: in The interval number, To predict the set of intervals exceeding the standard, For the first The center mileage coordinates of each interval; the migration speed is estimated by the following formula: in, for The coordinates of the center of gravity of the pollution cluster at all times. For the calculation period; combined with The symbols and traffic flow direction determine the migration trend of the pollution plume and the downstream target area it will enter.
4. The intelligent dispersion control method for tunnel contamination blobs based on edge prediction as described in claim 1, characterized in that, The triggering condition for the dynamic migration of the responsibility area in step S5 is: in, Mileage coordinates of the physical boundary points between adjacent intervals; duration of relay control link maintenance. in, To maintain the duration of the relay link, The vertical length of the pollution plume is used as the condition for the relay to be terminated when the center of gravity of the pollution plume completely leaves the current main control zone, and the predicted concentration of each zone is lower than the dynamic safety threshold of the corresponding zone in several consecutive prediction periods.
5. A tunnel contamination pyre intelligent dispersal control system based on edge prediction, used to execute the tunnel contamination pyre intelligent dispersal control method based on edge prediction as described in any one of claims 1 to 4, characterized in that, include: The physical sensing layer consists of environmental sensor arrays and traffic flow detection equipment deployed longitudinally along the tunnel to collect data on pollutant concentration, wind speed, temperature, humidity, and traffic flow parameters. The edge computing layer divides the entire tunnel length longitudinally into... Each ventilation responsibility zone is equipped with an edge computing unit, a safety control unit, and a variable frequency fan unit. Adjacent zones have physical overlap sections. Each zone adopts a heterogeneous architecture with dual computing and safety channels. The main computing channel is an edge computing unit, and the safety channel is an industrial-grade programmable logic controller or embedded real-time processor independent of the main computing unit. Each zone deploys two sets of edge computing units, one main and one backup, and supports... Automatic fault switching; In the communication network layer, adjacent edge computing units are interconnected through an industrial Ethernet ring network, and the edge computing units are connected to the cloud platform.
Citation Information
Patent Citations
Road traffic tunnel fan motor control method and device
CN118944522A
Tunnel energy-saving optimization dynamic regulation and control method and system
CN120724859A
