Tunnel ventilation system and method of controlling the same

Abstract

This invention discloses a tunnel ventilation system and its control method, relating to the field of tunnel ventilation energy-saving control technology. The method first preprocesses and divides the tunnel's traffic data, environmental data, and equipment operation data into datasets. It then constructs an LSTM-based traffic prediction model and a fully connected environmental prediction model, trained using root mean square error and mean absolute percentage error, respectively. The models sequentially predict traffic and environmental data. With energy minimization as the objective and environmental compliance as the constraint, a sequential quadratic programming algorithm is used to solve for the optimal combination of ventilation equipment and adjust its operation. Its advantages lie in the two-level prediction model's ability to achieve proactive regulation, preventing pollutant exceedances and ensuring air quality. Combined with dynamic training, multi-parameter optimization, and fault tolerance mechanisms, it accurately balances energy saving and environmental control while improving system reliability and intelligence, reducing manual intervention.

Description

Technical Field

[0001] This invention relates to the field of tunnel ventilation energy-saving control technology, and in particular to a tunnel ventilation system and its control method. Background Technology

[0002] The demand for tunnel construction in urban areas of my country is booming, with the length and complexity of tunnels increasing year by year. Because tunnels are semi-enclosed, it is difficult to achieve sufficient air circulation and exchange with the outside air. The gradual accumulation of vehicle exhaust pollutants within the tunnel forms a pollution layer about 1.5 meters above the ground, causing a decrease in visibility and air quality. This seriously affects the physical and mental health of drivers and passengers inside the tunnel, as well as residents near tunnel entrances and exhaust vents. Therefore, a reasonable ventilation system is needed to meet the pollutant control requirements inside and outside the tunnel. Furthermore, existing tunnel projects are designed with ventilation equipment based on long-term peak hour conditions, with equipment capacity far exceeding the ventilation needs under most actual traffic conditions. The control system has become a key factor affecting the efficient and energy-saving operation of the ventilation system.

[0003] Chinese patent application CN201711394301.7 discloses a method for predicting CO concentration based on a BP neural network, but it only targets a single indicator and does not integrate multi-source data such as traffic and equipment, thus having limited ability to fit nonlinear and strongly time-series data such as traffic flow. Summary of the Invention

[0004] In view of this, the present invention proposes a tunnel ventilation system and its control method. By constructing an intelligent prediction-optimization control system that integrates the temporal characteristics of traffic flow with the coupling relationship of environmental parameters, the synergistic goals of "accurate prediction, dynamic regulation, and energy saving compliance" of the tunnel ventilation system are achieved.

[0005] This invention provides a tunnel ventilation system and its control method, the steps of which are as follows:

[0006] S1 acquires and preprocesses historical data inside the tunnel, including traffic monitoring data, environmental monitoring data, and equipment activation data; the preprocessing includes data cleaning, outlier removal, normalization, and time-series sliding window processing.

[0007] S2 constructs an initial traffic prediction model, using historical traffic monitoring data with time features, and combines it with an LSTM directional recurrent neural network for model training, using root mean square error as the loss function.

[0008] S3 constructs an initial environment prediction model, inputs all historical data into a fully connected neural network for training, and uses the mean absolute percentage error as the loss function.

[0009] S4 inputs the traffic monitoring data of the current moment into the trained traffic prediction model to obtain the traffic prediction data of the next moment; inputs the traffic prediction data of the next moment, the environmental monitoring data of the current moment, and the equipment activation data of the current moment into the trained environmental prediction model to obtain the environmental prediction data of the next moment.

[0010] S5 constructs a mixed integer nonlinear programming problem with the objective of minimizing equipment energy consumption and the constraint of meeting environmental parameters, based on environmental prediction data and preset environmental parameter thresholds. The optimization problem is solved by a sequential quadratic programming algorithm to determine the optimal combination of ventilation equipment to be turned on and adjust the equipment operating status.

[0011] Furthermore, the traffic monitoring data includes traffic volume, vehicle speed, and large vehicle mixing rate; the environmental monitoring data includes PM concentration, CO concentration, NO2 concentration, cross-sectional wind speed, and temperature; and the equipment operation data includes the on / off status of jet fans, the on / off status of distributed air purification units, the operating frequency of exhaust fans, and the operating frequency of supply fans.

[0012] Furthermore, the traffic prediction model includes an input layer, a hidden layer, and an output layer, wherein:

[0013] The input layer consists of traffic monitoring data and temporal characteristics at the current moment;

[0014] The hidden layer consists of two LSTM layers, each followed by a Dropout layer and a BatchNormalization layer;

[0015] The output layer is a fully connected layer that outputs traffic prediction data for the next time step.

[0016] Furthermore, the environmental prediction model includes an input layer, a hidden layer, and an output layer, wherein:

[0017] The input layer consists of the traffic forecast data for the next time step predicted in step S2, the environmental monitoring data for the current time step, and the equipment activation data.

[0018] The hidden layer consists of two fully connected layers, each followed by a ReLU activation function and a BatchNormalization layer;

[0019] The output layer is a fully connected layer, which outputs environmental prediction data for the next 15-60 minutes.

[0020] Furthermore, the objective of solving the optimization problem S5 is to bring the control index to a target range, where the control index includes k1 and k2, wherein:

[0021] k1 = Predicted PM concentration;

[0022] k2 = max(predicted CO concentration, predicted NO2 concentration, predicted wind speed, predicted temperature);

[0023] When k1 > k2, PM pollution is the primary issue. If k1 > 1, meaning the predicted PM concentration exceeds the standard, prioritize increasing the number of distributed air purification units in operation, then increase the operating frequency of the supply fans, then increase the operating frequency of the exhaust fans, and finally increase the on / off status of the jet fans. Repeat step S5 after each adjustment until k1 and k2 are both no greater than 1. If 1 ≥ k1 ≥ 0.9, keep the current equipment operation status unchanged. If k1 < 0.9, meaning the predicted PM concentration is too low, prioritize decreasing the on / off status of the jet fans, then decrease the operating frequency of the exhaust fans, then decrease the operating frequency of the supply fans, and finally decrease the number of distributed air purification units in operation. Repeat step S5 after each adjustment until k1 and k2 are both no less than 0.9.

[0024] When k1≤k2, the main issues are CO / NO2 / wind speed / temperature. If k2>1, other parameters are out of control. In this case, increase the operating frequency of the supply fan first, then increase the operating frequency of the exhaust fan, and finally increase the on / off state of the jet fan. Repeat step S5 after each adjustment until k1 and k2 are both no greater than 1. If 1≥k2≥0.9, keep the current equipment on. If k2<0.9, over-ventilation, decrease the on / off state of the jet fan first, then decrease the operating frequency of the exhaust fan, and finally decrease the operating frequency of the supply fan. Repeat step S5 after each adjustment until k1 and k2 are both no less than 0.9.

[0025] Furthermore, the model training process also includes model evaluation. During evaluation, mean absolute error, root mean square error, and coefficient of determination are used as core indicators. When the mean absolute error is ≤5%, the root mean square error is ≤8%, and the coefficient of determination is ≥0.9, the model is determined to have reached the preset accuracy threshold. The training process also includes: using cross-validation and early stopping strategies to prevent overfitting, and adjusting the network parameters of the environment prediction model until the preset accuracy threshold is met.

[0026] Furthermore, in the model training steps S2 and S3:

[0027] The Adam optimizer was used, with an initial learning rate of 0.001 and a learning rate decay factor of 0.95.

[0028] Set the batch size to 32-64, the number of training epochs to 100-200, and use an early stopping strategy to prevent overfitting.

[0029] Every 5 training epochs, the model performance is evaluated using the validation set, and the model weights with the smallest validation error are saved.

[0030] Furthermore, in the optimization problem S5:

[0031] The objective function is to minimize the total energy consumption of the tunnel ventilation system, where the total energy consumption is the sum of the energy consumption of each device. Device energy consumption is related to device operating power and operating time, as shown in the formula:

[0032]

[0033] Where P i Let t be the power of device i. i Runtime;

[0034] The constraints include:

[0035] Environmental parameter constraints, i.e., predicted concentration ≤ environmental standard threshold;

[0036] Equipment operation constraints, namely frequency constraints and number of devices activated;

[0037] Logical constraints, namely matching the number of jet fans in operation with the wind speed conditions, to avoid airflow short-circuiting.

[0038] Furthermore, it also includes model optimization and control strategy iteration:

[0039] Online learning mechanism: Every T hours, new monitoring data is collected, the dataset is updated, and the traffic prediction model and environmental prediction model are retrained;

[0040] Adaptive weight adjustment: The weight coefficients of control indicators k1 and k2 are dynamically adjusted according to the importance of environmental parameters in different seasons and time periods;

[0041] Fault diagnosis and fault tolerance: Monitor equipment status in real time, and automatically switch to backup equipment when equipment failure is detected, and re-optimize the control strategy.

[0042] Furthermore, the present invention also includes a tunnel ventilation system, the system comprising: traffic monitoring equipment, tunnel environmental detector, temperature sensor, wind direction and speed meter, jet fan control cabinet, distributed air purification unit control cabinet, exhaust fan control cabinet and supply fan control cabinet, and a tunnel ventilation digital control platform connected to the aforementioned equipment via industrial Ethernet or Modbus protocol;

[0043] The tunnel ventilation digital control platform is embedded with an environmental prediction model and a traffic prediction model trained based on a historical database.

[0044] Preferably, the tunnel-type environmental detector is a CO / VI / NO2 detector.

[0045] The present invention has the following advantages over the prior art:

[0046] The core advantage of this solution lies in its innovative "prediction-optimization-control" process, which enables precise, energy-efficient, and adaptive operation of the tunnel ventilation system.

[0047] On the one hand, by using a two-level prediction model built with LSTM and a fully connected network, the system first accurately captures the temporal fluctuations of traffic flow, and then uses this as a basis to predict future changes in environmental parameters such as PM, CO, and NO2. This breaks through the lag of traditional control, enables proactive regulation of pollutants before they exceed standards, and comprehensively protects the air quality and health of drivers and passengers in the tunnel.

[0048] On the other hand, by using a multi-objective optimization strategy and a sequential quadratic programming algorithm, the system can significantly reduce energy consumption by dynamically adjusting the equipment combination based on the priority of PM and other pollutants, while ensuring that environmental parameters meet the standards. At the same time, the online learning mechanism, adaptive weight adjustment, and fault-tolerant design enable the system to flexibly adapt to traffic patterns, seasonal changes, and equipment status fluctuations, reducing the cost of manual intervention. It is especially suitable for the dynamic operation needs of long-distance, highly complex urban tunnels. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a diagram illustrating the components of a tunnel ventilation system according to an embodiment of the present invention.

[0051] Figure 2 This is a flowchart illustrating the digital control process for tunnel ventilation according to an embodiment of the present invention.

[0052] Figure 3 This is a diagram of the neural network structure of the traffic prediction model in an embodiment of the present invention.

[0053] Figure 4 This is a typical traffic prediction result diagram based on an embodiment of the present invention;

[0054] Figure 5 This is a diagram of the neural network structure of the environmental prediction model in an embodiment of the present invention.

[0055] Figure 6 This diagram illustrates the calculation method for control indices k1 and k2 in an embodiment of the present invention. Detailed Implementation

[0056] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0057] This invention provides a novel tunnel ventilation system and its control method, namely, a tunnel ventilation system combining jet fans and distributed air purification units, and its control method. The tunnel ventilation system is designed to meet the tunnel's ventilation and air exchange requirements. Several jet fans and distributed air purification units are arranged longitudinally within the tunnel. Ventilation shafts are installed near the tunnel entrance, and several exhaust fans and supply fans are arranged within these shafts, based on environmental impact assessments and ventilation requirements. Furthermore, to achieve energy-saving control of the aforementioned jet fans, distributed air purification units, exhaust fans, and supply fans, this invention provides a corresponding energy-saving control method. This method utilizes a neural network model to reasonably predict traffic volume. Based on the traffic prediction results, the neural network model predicts future environmental parameters under different ventilation equipment operation modes, and ultimately determines the ventilation equipment operation mode that balances energy efficiency and environmental quality. This tunnel ventilation system and its control method can meet the air quality requirements during long-term peak hours and dynamically adjust the ventilation equipment operation status according to changes in traffic conditions, avoiding energy waste.

[0058] like Figure 1As shown, the system is installed inside tunnel 1. Tunnel 1 has a tunnel entrance 101 and a tunnel exit 102. Motor vehicles enter from the tunnel entrance 101 and exit from the tunnel exit 102. To meet the long-term peak-hour vehicle pollutant control requirements and tunnel ventilation needs, several jet fans 2 and distributed air purification units 3 are installed inside tunnel 1. Considering specific civil engineering conditions, environmental impact assessment requirements, and the need for segmented longitudinal ventilation, the tunnel generally has one or more ventilation shafts 103. Several exhaust fans 4 and supply fans 5 are installed in the ventilation shafts 103. To achieve energy-saving control of the jet fans 2, distributed air purification units 3, exhaust fans 4, and supply fans 5, the tunnel... Unit 1 is equipped with multiple traffic monitoring devices 6, CO / VI / NO2 detectors 7, temperature sensors 8, and wind direction and speed meters 9. Jet fans 2 are equipped with jet fan control cabinets 10, distributed air purification units 3 are equipped with distributed air purification unit control cabinets 11, exhaust fans 4 are equipped with exhaust fan control cabinets 12, and supply fans 5 are equipped with supply fan control cabinets 13. The traffic monitoring devices 6, CO / VI / NO2 detectors 7, temperature sensors 8, wind direction and speed meters 9, jet fan control cabinets 10, distributed air purification unit control cabinets 11, exhaust fan control cabinets 12, and supply fan control cabinets 13 are connected to the tunnel ventilation digital control platform 15 via communication cables 14. A tunnel ventilation system and its control method, in order to achieve energy-saving control, includes a traffic monitoring device 6, a CO / VI / NO2 detector 7, a temperature sensor 8, a wind direction and speed meter 9, a jet fan control cabinet 10, a distributed air purification unit control cabinet 11, an exhaust fan control cabinet 12, and a supply fan control cabinet 13, which can collect traffic monitoring data, environmental monitoring data, and equipment operation data in real time, and transmit them to a tunnel ventilation digital control platform 15 via a communication cable 14. The tunnel ventilation digital control platform 15 can perform data cleaning and data storage on the traffic monitoring data, environmental monitoring data, and equipment operation data. The tunnel ventilation digital control platform 15 is embedded with a system based on... The environmental prediction model and traffic prediction model trained from the historical database can be used to predict traffic data for the next moment based on real-time traffic monitoring data. The environmental prediction model can obtain environmental prediction data for the next moment under different equipment operation modes based on traffic prediction data and current environmental monitoring data. Through iterative calculation, the ventilation equipment operation mode that takes into account both energy-saving benefits and environmental quality can be finally determined. The tunnel ventilation digital control platform 15 outputs the calculation results to the jet fan control cabinet 10, the distributed air purification unit control cabinet 11, the exhaust fan control cabinet 12, and the supply fan control cabinet 13, ultimately realizing energy-saving control of the ventilation equipment.

[0059] like Figure 2 As shown, the present invention provides a tunnel ventilation system and its control method, the steps of which are as follows:

[0060] S1 acquires and preprocesses historical data from inside the tunnel, including traffic monitoring data, environmental monitoring data, and equipment activation data; preprocessing includes data cleaning, outlier removal, normalization, and time-series sliding window processing.

[0061] (1) Traffic monitoring data includes traffic volume, vehicle speed, and large vehicle mixing rate; environmental monitoring data includes PM concentration, CO concentration, NO2 concentration, cross-sectional wind speed, and temperature; equipment operation data includes the on / off status of jet fans, the on / off status of distributed air purification units, the operating frequency of exhaust fans, and the operating frequency of supply fans, as detailed below:

[0062] 1) Traffic monitoring data is collected by traffic monitoring equipment group i (6), and mainly includes:

[0063] Traffic volume q v1 ,...,q vi

[0064] Vehicle speed v1,...,vi

[0065] Large vehicle mixing rate r1,...,ri

[0066] 2) Environmental monitoring data was collected by CO / VI / NO2 detector 7, temperature sensor 8, and wind direction and speed meter 9, mainly including:

[0067] PM concentration (pm1,...,pm) j

[0068] CO concentration (co1,...,co) j

[0069] NO2 concentrations: no1,...,no j

[0070] Cross-sectional wind speeds fs1,...,fs j

[0071] Temperature t1,...,t j

[0072] 3) Equipment start-up data is collected from m jet fan control cabinets 10, n distributed air purification unit control cabinets 11, k exhaust fan control cabinets 12, and p supply fan control cabinets 13, and mainly includes:

[0073] Jet fan switch status n s1 ,...,n sm (1 indicates on, 0 indicates off)

[0074] Distributed air purification unit on / off status n jh1 ,...,n jhn (1 indicates on, 0 indicates off)

[0075] Exhaust fan operating frequency n pf1 ,...,n pfk

[0076] Blower operating frequency n sf1 ,...,n sfp

[0077] (2) Data processing of traffic monitoring data, environmental monitoring data, and equipment activation data:

[0078] 1) Processing traffic monitoring data, including:

[0079] Traffic volume is taken as the average.

[0080] Take the average speed.

[0081] The average rate of large vehicle mixing is taken.

[0082] 2) Processing environmental monitoring data, including:

[0083] PM concentration is taken as the maximum value and is dimensionless.

[0084] The CO concentration is taken as the maximum value and is dimensionless.

[0085] The NO2 concentration is taken as the maximum value and is dimensionless.

[0086] The cross-sectional wind speed is taken as the minimum value and is dimensionless.

[0087] Temperature is taken as the maximum value and is dimensionless.

[0088] pm set ,co set no set fs set t set The corresponding limits for PM concentration, CO concentration, NO2 concentration, cross-sectional wind speed, and temperature are as follows:

[0089] symbol <![CDATA[pm set ]]> <![CDATA[co set ]]> <![CDATA[no set ]]> <![CDATA[fs set ]]> <![CDATA[t set ]]> Standard limits <![CDATA[0.005m -1 ]]> 70ppm 1ppm 1.5m / s 40℃

[0090] 3) Process the device startup data, including:

[0091] Calculation of the sum of the number of jet fans in operation.

[0092] Calculation of the sum of the number of distributed air purification units in operation.

[0093] Calculation of exhaust fan operating frequency.

[0094] Calculation of the sum of the operating frequencies of the blower.

[0095] (3) Store historical data and build a database.

[0096] Preprocessed traffic monitoring data (q) is stored at regular time intervals (e.g., 5 minutes). v Environmental monitoring data (PM, CO, NO, Fs, T) and equipment operation data (n) s n jh n pf n sf This forms a historical database.

[0097] S2 constructs an initial traffic prediction model, using historical traffic monitoring data with time features, and combines it with an LSTM directional recurrent neural network for model training, using root mean square error as the loss function.

[0098] according to Figure 3 Based on the structure, construct a neural network model for traffic prediction:

[0099] Input layer: Traffic monitoring data at the current time (q) v (v, r), time feature T;

[0100] Hidden layers: two LSTM layers with 64-32 neurons each, each followed by a Dropout layer and a BatchNormalization layer;

[0101] Output layer: The output layer is a fully connected layer that outputs the traffic prediction data for the next time step (q). v _pre, v_pre, r_pre);

[0102] Traffic monitoring data is divided into training and testing sets. A traffic prediction model is trained using an LSTM (Low-Speed ​​Targeted Recurrent Neural Network). The LSTM network can deeply explore the time-series relationships in prediction problems, effectively fitting non-linear data. Furthermore, the LSTM model can store information from time series data for long periods, making it effective even with data that has long time intervals and high latency, and it boasts high long-term prediction accuracy. The model also exhibits good robustness and flexibility, making it applicable to traffic flow prediction in a vehicle-to-everything (V2X) environment.

[0103] Prediction results are as follows Figure 4As shown, the traffic volume is taken as the average value q. v and q v The true trajectory of _pre.

[0104] S3 constructs an initial environment prediction model, inputs all historical data into a fully connected neural network for training, and uses the mean absolute percentage error as the loss function.

[0105] according to Figure 5 Based on the structure, construct a neural network model for environmental prediction:

[0106] Input layer: Traffic monitoring data at the current time (q) v (v, r), environmental monitoring data from the previous moment (pm) last ,co last no last fs last t last ) and device activation data at the current moment (n s n jh n pf n sf )

[0107] Hidden layers: Two hidden layers of neurons, ranging from 64 to 32.

[0108] Output layer: Environmental monitoring data at the current moment (PM, CO, NO, Fs, T)

[0109] Historical data for pm, co, no, and fs were divided into training and testing sets to train an environment prediction model, which then performed independent predictions for each set. The neural network model used for environment prediction was a Dense model, i.e., a fully connected neural network.

[0110] The model training parameters for the traffic prediction model and the environmental prediction model are set as follows:

[0111] The Adam optimizer was used, with an initial learning rate of 0.001 and a learning rate decay factor of 0.95.

[0112] Set the batch size to 32-64, the number of training epochs to 100-200, and use an early stopping strategy to prevent overfitting.

[0113] Every 5 training epochs, the model performance is evaluated using the validation set, and the model weights with the smallest validation error are saved.

[0114] When training the traffic prediction model, an LSTM directed recurrent neural network is used, with two LSTM hidden layers containing 64-32 neurons each. A dropout layer with a dropout rate of 0.2-0.3 is added after each LSTM layer. The output layer is a fully connected layer with the same number of target variables as the target variables. The input is a preprocessed traffic monitoring data sequence containing time features. The Adam optimizer is used, with an initial learning rate of 0.001 and a decay factor of 0.95 every 5 rounds. The batch size is 32-64, and the number of training rounds is 100-200. The root mean square error is used as the loss function. The performance is evaluated using a validation set every 5 rounds. Early stopping is triggered when the validation set RMSE does not decrease for 10 consecutive rounds, and the model weights with the smallest validation error are saved.

[0115] When training the environment prediction model, a fully connected neural network is used, with 2-3 hidden layers containing 64-32 neurons each. A BatchNormalization layer is added after each hidden layer. The output layer is a fully connected layer with the same number of predicted environment parameters. The inputs are traffic prediction data, historical environmental monitoring data, equipment activation status data, and time features. The Adam optimizer is also used, with an initial learning rate of 0.001 and a decay factor of 0.95 every 5 rounds. The batch size is 32-64, and the number of training rounds is 100-200. The mean absolute percentage error (MAPE) is used as the loss function. The model is validated every 5 rounds. The model is stopped early when the MAPE on the validation set does not decrease for 10 consecutive rounds, and the optimal model weights are saved.

[0116] The accuracy of the environmental prediction model is evaluated using a test set of environmental monitoring data. During the evaluation, mean absolute error, root mean square error, and coefficient of determination are used as core indicators. When the mean absolute error is ≤5%, the root mean square error is ≤8%, and the coefficient of determination is ≥0.9, the model is considered to have reached the preset accuracy threshold. The training process also includes using cross-validation and early stopping strategies to prevent overfitting, and adjusting the network parameters of the environmental prediction model until the preset accuracy threshold is met.

[0117] Both models use a data generator to load training data in batches and perform real-time data augmentation. After training, the final performance is evaluated using a test set, and the optimal weights are loaded for actual prediction.

[0118] S4 inputs the traffic monitoring data of the current moment into the trained traffic prediction model to obtain the traffic prediction data of the next moment; inputs the traffic prediction data of the next moment, the environmental monitoring data of the current moment, and the equipment activation data of the current moment into the trained environmental prediction model to obtain the environmental prediction data of the next moment.

[0119] Using the traffic prediction model trained in step S2, input the traffic monitoring data (q) at the current time. vThe traffic prediction data for the next time step (q) is calculated from the values ​​of v and r. v _pre, v_pre, r_pre). Using the environmental prediction model trained in step S2, input the traffic prediction data (q) for the next time step. v _pre, v_pre, r_pre), current environmental monitoring data (pm, co, no, fs, t), current device activation data (n s n jh n pf n sf The environmental prediction data (PM) for the next time step is calculated. pre ,co pre no pre fs pre t pre ).

[0120] S5 constructs a mixed integer nonlinear programming problem with the objective of minimizing equipment energy consumption and the constraint of meeting environmental parameters, based on environmental prediction data and preset environmental parameter thresholds. The optimization problem is solved by a sequential quadratic programming algorithm to determine the optimal combination of ventilation equipment to be turned on and adjust the equipment operating status.

[0121] The objective of solving the optimization problem is to bring the control indicators to a target range, where the control indicators include k1 and j2, wherein:

[0122] j1 = Predicted PM concentration;

[0123] j2 = max(predicted CO concentration, predicted NO2 concentration, predicted wind speed, predicted temperature);

[0124] When k1 > k2, PM pollution is the primary issue. If j1 > 1, meaning the predicted PM concentration exceeds the standard, prioritize increasing the number of distributed air purification units in operation, then increase the operating frequency of the supply fans, then increase the operating frequency of the exhaust fans, and finally increase the on / off status of the jet fans. Repeat step S5 after each adjustment until k1 and k2 are both no greater than 1. If 1 ≥ k1 ≥ 0.9, keep the current equipment operation status unchanged. If k1 < 0.9, meaning the predicted PM concentration is too low, prioritize decreasing the on / off status of the jet fans, then decrease the operating frequency of the exhaust fans, then decrease the operating frequency of the supply fans, and finally decrease the number of distributed air purification units in operation. Repeat step S5 after each adjustment until k1 and k2 are both no less than 0.9.

[0125] When k1≤k2, the main issues are CO / NO2 / wind speed / temperature. If j2>1, other parameters are out of control. In this case, increase the operating frequency of the supply fan first, then increase the operating frequency of the exhaust fan, and finally increase the on / off state of the jet fan. Repeat step S5 after each adjustment until k1 and k2 are both no greater than 1. If 1≥k2≥0.9, keep the current equipment on. If j2<0.9, over-ventilation, decrease the on / off state of the jet fan first, then decrease the operating frequency of the exhaust fan, and finally decrease the operating frequency of the supply fan. Repeat step S5 after each adjustment until k1 and j2 are both no less than 0.9.

[0126] The specific equipment control table is shown below:

[0127]

[0128] like Figure 6 As shown, how to output the final j1 and j2 based on environmental prediction data? It is known that jet fan 2, exhaust fan 4, and supply fan 5 act on air, and their operating status affects the ventilation volume inside the tunnel, thus affecting PM2.5 levels. pre ,co pre no pre fs pre t pre Impact; the distributed air purification unit 3 targets particulate matter, and its operation only affects PM2.5. pre Therefore, ventilation equipment should be adjusted according to the two scenarios. The goal of the adjustment is to ensure that the highest environmental parameter index is controlled within a reasonable range (0.9 to 1.0), and to ensure that all environmental parameters meet the requirements of the specifications, while reducing the number of ventilation equipment that need to be turned on and avoiding energy waste.

[0129] Take control index k1 = pm pre ;

[0130] Take the control index k2 = max{co pre no pre fs pre t pre};

[0131] When k1 > k2, the equipment is adjusted according to the following logic:

[0132] ① When k1>1, increase n first. jh (Increase by 1 with each adjustment), then increase by n. sf (Adjust by +5 each time), then increase n. pf (Adjust by +5 each time), finally increase by n. s (Increase by 1 for each adjustment), and repeat step five after each adjustment until both k1 and k2 are no greater than 1 and then stop adjusting.

[0133] ② When 1≥k1≥0.9, keep the current device on state unchanged;

[0134] ③ When k1 < 0.9, finally decrease n. s (Adjust by -1 each time), then decrease n. pf (Adjust by -5 each time), then decrease n. sf (Adjust by -5 each time), finally decrease n. jh (Adjust by -1 each time), repeat step five after each adjustment, until both k1 and k2 are not less than 0.9 and then stop adjusting.

[0135] When k1≤k2, the equipment is adjusted according to the following logic:

[0136] ① When k2>1, increase n first. sf (Adjust by +5 each time), then increase n. pf (Adjust by +5 each time), finally increase by n. s (Increase by 1 for each adjustment), and repeat step five after each adjustment until both k1 and k2 are no greater than 1 and then stop adjusting.

[0137] ② When 1≥k2≥0.9, keep the current device on state unchanged;

[0138] ③ When k2 < 0.9, finally decrease n. s (Adjust by -1 each time), then decrease n. pf (Adjust by -5 each time), finally decrease n. sf (Adjust by -5 each time), repeat step five after each adjustment until both k1 and k2 are not less than 0.9 and then stop adjusting.

[0139] Output device start data (n) when adjustment stops pre_s n pre_jh n pre_pf n pre_sf )

[0140] According to the principle of proximity to exhaust shaft 103, additional (or closed) jet fans and distributed air purification units will be added (or closed), and the number of supply and exhaust fans will be increased (or decreased) according to the principle of unit control. Finally, the tunnel ventilation digital control platform 15 will output the equipment operation status data for the next moment, including:

[0141] m jet fan switch status n pre_s1 ,...,n pre_sm (1 indicates on, 0 indicates off)

[0142] n distributed air purification units on / off status n pre_jh1 ,...,n pre_jhn (1 indicates on, 0 indicates off)

[0143] k exhaust fans operating frequency n pre_pf1 ,...,n pre_pfk

[0144] p units of blower operating frequency n pre_sf1 ,...,n pre_sfp

[0145] After receiving the control signals input from the tunnel ventilation digital control platform 15, the jet fan control cabinet 10, the distributed air purification unit control cabinet 11, the exhaust fan control cabinet 12, and the supply fan control cabinet 13 adjust the on / off (frequency) status of the jet fan 2, the distributed air purification unit 3, the exhaust fan 4, and the supply fan 5.

[0146] The sequential quadratic programming algorithm is an efficient numerical method for solving nonlinear programming problems. Its core idea is to approximate the optimal solution of the original problem by successively constructing and solving quadratic programming subproblems.

[0147] Specifically, for optimization problems where both the objective function and constraints are nonlinear, the core steps of the SQP algorithm include:

[0148] In each iteration, the original nonlinear problem is approximated as a quadratic programming subproblem, that is, the objective function is a quadratic function and the constraints are linear functions. The constraints are linearized by Taylor expansion and the objective function is approximated based on the Lagrange function.

[0149] Solve this quadratic programming subproblem to obtain the next search direction and step size;

[0150] The iteration step size is determined by methods such as line search, the variables are updated and the convergence is checked, and the iteration is repeated until the accuracy requirement is met.

[0151] Compared to other optimization algorithms, the SQP algorithm converges faster and is more accurate when dealing with constrained nonlinear problems. It is suitable for solving complex optimization problems with multiple objectives and strong coupling, and can efficiently solve for the optimal combination of ventilation equipment activation.

[0152] In the optimization problem, the objective function is to minimize the total energy consumption of the tunnel ventilation system, where the total energy consumption is the sum of the energy consumption of each device. Device energy consumption is related to device operating power and operating time, as shown in the formula:

[0153]

[0154] Where P i Let t be the power of device i. i Runtime;

[0155] The constraints include:

[0156] Environmental parameter constraints, i.e., predicted concentration ≤ environmental standard threshold;

[0157] Environmental parameter constraints need to cover multiple types of pollutants and environmental indicators, and take into account differences in tunnel operating conditions, such as tightening thresholds during peak hours.

[0158]

[0159] Among them, C PM2.5,pred C CO,pred Predicted concentrations of PM2.5 and CO; C PM2.5,std C CO,std The corresponding pollutant standard thresholds are: k1 and k2 are dynamic adjustment coefficients; during peak hours, k1 = 0.9 to tighten the threshold, and during off-peak hours, k1 = 1.0; T pred V air,pred To predict temperature and wind speed; T std,max T std,min V represents the upper and lower threshold values ​​of the temperature rate. std,max V std,min These are the upper and lower threshold values ​​for wind speed.

[0160] Equipment operation constraints, namely frequency constraints and number of devices activated;

[0161] Equipment operation constraints must limit the physically feasible range to prevent equipment from operating beyond its capacity.

[0162]

[0163] Among them, f i Where N is the operating frequency of device i. i P represents the number of devices i that are enabled, with the subscripts min / max indicating the upper and lower limits of the corresponding parameters. i,op (f i P represents the power of device i. i,rated This refers to the rated power of device i.

[0164] Logical constraints, namely matching the number of jet fans in operation with the wind speed conditions, to avoid airflow short-circuiting.

[0165] Logical constraints focus on the rationality of airflow organization to prevent airflow conflicts between jet fans and main ventilation fans. This applies when the number of jet fans is N. jet At this time, the wind speed coordination condition must be met:

[0166] |V jet,avg -0.6·V main,avg |≤ΔV thresh

[0167] Among them, V jet,avg V is the average outlet velocity of the jet fan. main,avg Average outlet velocity of the main ventilation fan (supply / exhaust fan), ΔVthresh The wind speed difference threshold is 2 m / s; exceeding this threshold may cause airflow vortices or short circuits.

[0168] At the same time, the number of jet fans activated must be matched with the tunnel segments:

[0169]

[0170] Where, N jet L represents the number of devices that can be turned on. jet,seg The optimal segment spacing for the jet fan, for example, is 200 meters, which corresponds to the tunnel length L. tunnel One jet fan is matched for every 200 meters to ensure uniform airflow coverage and avoid sudden changes in local wind speed.

[0171] The technical solution of this invention also includes model optimization and control strategy iteration:

[0172] Online learning mechanism: Every T hours, new monitoring data is collected, the dataset is updated, and the traffic prediction model and environmental prediction model are retrained;

[0173] Adaptive weight adjustment: The weight coefficients of control indicators k1 and k2 are dynamically adjusted according to the importance of environmental parameters in different seasons and time periods;

[0174] Fault diagnosis and fault tolerance: Monitor equipment status in real time, and automatically switch to backup equipment when equipment failure is detected, and re-optimize the control strategy.

[0175] This invention's model optimization and control strategy iteration revolve around online learning, adaptive weight adjustment, fault diagnosis, and fault tolerance: For each configurable interval of T hours (3-6 hours during peak hours and 12-24 hours during off-peak hours), traffic, environmental, and equipment operation data are collected from the tunnel monitoring system. After preprocessing via the S1 process (data cleaning, etc.), the historical training set is merged, and the traffic and environmental prediction model is retrained according to the S2 and S3 processes to adapt to traffic flow and seasonal fluctuations. Based on the statistical patterns of environmental parameter influences from historical data, the seasonal dimension k1 (PM2.5 control weight) is increased in winter. The system adjusts the control weight to 0.6-0.7 during the summer and increases k2 (CO control weight) to 0.6-0.7 during the summer. It also increases k2 during the morning peak hours and k1 during the night hours. The system supports real-time dynamic weighting when parameters exceed the limit continuously, thus optimizing control priority. By collecting operating parameters such as fan current through the equipment IoT module, the system can detect main equipment failures and switch to backup equipment within 10 seconds. Based on the new set of available equipment, the system re-runs the S5 process to reconstruct the control strategy. After the equipment is repaired, the system automatically verifies and switches back, ensuring that the system model is self-evolving, the strategy is self-adapting, and the operation is highly reliable. The system is also integrated with the project equipment table and data report to refine parameters.

[0176] This invention includes a tunnel ventilation system, comprising:

[0177] Traffic monitoring equipment, tunnel environmental detectors, temperature sensors, wind direction and speed meters, jet fan control cabinets, distributed air purification unit control cabinets, exhaust fan control cabinets and supply fan control cabinets, and a tunnel ventilation digital control platform connected to the aforementioned equipment via industrial Ethernet or Modbus protocol; the tunnel ventilation digital control platform is embedded with an environmental prediction model and a traffic prediction model trained based on a historical database.

[0178] Preferably, the tunnel-type environmental detector is a CO / VI / NO2 detector.

[0179] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A tunnel ventilation control method, characterized in that, Includes the following steps: S1 acquires and preprocesses historical data inside the tunnel, including traffic monitoring data, environmental monitoring data, and equipment activation data; the preprocessing includes data cleaning, outlier removal, normalization, and time-series sliding window processing. S2 constructs an initial traffic prediction model, using historical traffic monitoring data with time features, and combines it with an LSTM directional recurrent neural network for model training, using root mean square error as the loss function. S3 constructs an initial environment prediction model, inputs all historical data into a fully connected neural network for training, and uses the mean absolute percentage error as the loss function. The environmental prediction model includes an input layer, a hidden layer, and an output layer, wherein: The input layer consists of the traffic forecast data for the next time step predicted in step S2, the environmental monitoring data for the current time step, and the equipment activation data. The hidden layer consists of two fully connected layers, each followed by a ReLU activation function and a BatchNormalization layer; The output layer is a fully connected layer, which outputs environmental prediction data for the next 15 to 60 minutes. S4 inputs the traffic monitoring data of the current moment into the trained traffic prediction model to obtain the traffic prediction data of the next moment; inputs the traffic prediction data of the next moment, the environmental monitoring data of the current moment, and the equipment activation data of the current moment into the trained environmental prediction model to obtain the environmental prediction data of the next moment. S5 constructs a mixed-integer nonlinear programming problem with the objective of minimizing equipment energy consumption and the constraint of meeting environmental parameter standards, based on environmental prediction data and preset environmental parameter thresholds. The optimization problem is solved by a sequential quadratic programming algorithm to determine the optimal combination of ventilation equipment to be turned on and adjust the equipment operating status. The objective of solving the optimization problem S5 is to bring the control index to a target range. The control index includes k1 and k2, where: k 1=PM浓度预测值； k2 = max(predicted CO concentration, predicted NO2 concentration, predicted wind speed, predicted temperature); When k 1＞ k2 indicates that PM pollution is the primary issue. If k1 > 1, meaning the predicted PM concentration exceeds the standard, then prioritize increasing the number of distributed air purification units in operation, followed by increasing the operating frequency of the supply fans, then increasing the operating frequency of the exhaust fans, and finally adjusting the on / off status of the jet fans. Repeat step S5 after each adjustment until both k1 and k2 are no greater than 1. If 1 ≥ k1 ≥ 0.9, then keep the current equipment operation status unchanged. If k1 < 0.9, meaning the predicted PM concentration is too low, then prioritize decreasing the on / off status of the jet fans, followed by decreasing the operating frequency of the exhaust fans, then decreasing the operating frequency of the supply fans, and finally decreasing the number of distributed air purification units in operation. Repeat step S5 after each adjustment until both k1 and k2 are no less than 0.

9. When k1≤k2, the main issues are CO / NO2 / wind speed / temperature. If k2>1, other parameters are out of control. In this case, increase the operating frequency of the supply fan first, then increase the operating frequency of the exhaust fan, and finally increase the on / off state of the jet fan. Repeat step S5 after each adjustment until k1 and k2 are both no greater than 1. If 1≥k2≥0.9, keep the current equipment on. If k2<0.9, over-ventilation, decrease the on / off state of the jet fan first, then decrease the operating frequency of the exhaust fan, and finally decrease the operating frequency of the supply fan. Repeat step S5 after each adjustment until k1 and k2 are both no less than 0.

9. In the optimization problem S5: The objective function is to minimize the total energy consumption of the tunnel ventilation system, where the total energy consumption is the sum of the energy consumption of each piece of equipment. Equipment energy consumption is related to the equipment's operating power and operating time, as shown in the formula: ; Where P i For the power of device i, t i Runtime; The constraints include: Environmental parameter constraints, i.e., predicted concentration ≤ environmental standard threshold; Equipment operation constraints, namely frequency constraints and number of devices activated; Logical constraints, namely matching the number of jet fans in operation with the wind speed conditions, to avoid airflow short-circuiting.

2. The tunnel ventilation control method as described in claim 1, characterized in that, The traffic monitoring data includes traffic volume, vehicle speed, and large vehicle mixing rate; the environmental monitoring data includes PM concentration, CO concentration, NO2 concentration, cross-sectional wind speed, and temperature; the equipment operation data includes the on / off status of jet fans, the on / off status of distributed air purification units, the operating frequency of exhaust fans, and the operating frequency of supply fans.

3. The tunnel ventilation control method as described in claim 1, characterized in that, The traffic prediction model comprises an input layer, a hidden layer, and an output layer, wherein: The input layer consists of traffic monitoring data and temporal characteristics at the current moment; The hidden layer consists of two LSTM layers, each followed by a Dropout layer and a BatchNormalization layer; The output layer is a fully connected layer that outputs traffic prediction data for the next time step.

4. The tunnel ventilation control method as described in claim 1, characterized in that, The model training process also includes model evaluation. During the evaluation, mean absolute error, root mean square error, and coefficient of determination are used as core indicators. When the mean absolute error is ≤5%, the root mean square error is ≤8%, and the coefficient of determination is ≥0.9, the model is judged to have reached the preset accuracy threshold. The training process also includes: using cross-validation and early stopping strategies to prevent overfitting, and adjusting the network parameters of the environment prediction model until a preset accuracy threshold is met.

5. The tunnel ventilation control method as described in claim 1, characterized in that, In the model training steps S2 and S3: The Adam optimizer was used, with an initial learning rate of 0.001 and a learning rate decay factor of 0.

95. Set the batch size to 32-64, the number of training epochs to 100-200, and use an early stopping strategy to prevent overfitting. Every 5 training epochs, the model performance is evaluated using the validation set, and the model weights with the smallest validation error are saved.

6. The tunnel ventilation control method as described in claim 1, characterized in that, It also includes model optimization and control strategy iteration: Online learning mechanism: Every T hours, new monitoring data is collected, the dataset is updated, and the traffic prediction model and environmental prediction model are retrained; Adaptive weight adjustment: The weight coefficients of control indicators k1 and k2 are dynamically adjusted according to the importance of environmental parameters in different seasons and time periods; Fault diagnosis and fault tolerance: Monitor equipment status in real time, and automatically switch to backup equipment when equipment failure is detected, and re-optimize the control strategy.

7. A tunnel ventilation system, characterized in that, include: Traffic monitoring equipment, tunnel environmental detectors, temperature sensors, wind direction and speed meters, jet fan control cabinets, distributed air purification unit control cabinets, exhaust fan control cabinets and supply fan control cabinets, and a tunnel ventilation digital control platform connected to the aforementioned equipment via industrial Ethernet or Modbus protocol; The tunnel ventilation digital control platform is embedded with an environmental prediction model and a traffic prediction model trained based on a historical database, and is used to execute the tunnel ventilation control method according to any one of claims 1-6.

Images