A method for minimizing energy consumption of a ventilation network for construction of a group of underground caverns

Abstract

The application discloses a kind of underground cavern group construction ventilation network energy consumption minimization method, it is related to ventilation network optimization and energy consumption control technical field, this method includes: constructing dynamic ventilation network perception and reconstruction model during construction, and the topological structure is dynamically updated by multi-source sensor and construction progress log;Based on Hardy Cross method, the theoretical air flow distribution satisfying physical conservation is solved;Fusion theoretical air flow and real-time data, construct Hardy Cross-GB hybrid prediction model, train gradient boosting regression to correct prediction bias, output air flow and energy consumption prediction value;Based on the model, the system total power minimization objective function is constructed, the frequency conversion strategy of fan is optimized under the constraint of air demand and the control instruction is output.The application realizes dynamic ventilation network accurate simulation and real-time energy efficiency optimization, significantly reduces the energy consumption of ventilation system, guarantees construction safety and control stability.

Description

Technical Field

[0001] This invention relates to the field of ventilation network optimization and energy consumption control technology, and in particular to a method for minimizing energy consumption in the construction ventilation network of underground cavern groups. Background Technology

[0002] As a core infrastructure for ensuring construction safety, removing harmful gases, and regulating environmental temperature and humidity, the ventilation system of underground engineering is mainly achieved through three types of network calculation and optimization techniques.

[0003] The first category is deterministic mathematical modeling methods, represented by the Hardy Cross iterative method. Based on Kirchhoff's laws, this method iteratively corrects the loop airflow until the pressure closure difference converges. Due to its high computational efficiency and clear physical meaning, it is widely used in the design of closed-loop ventilation networks with known physical parameters and fixed structures. The second category includes stochastic and numerical simulation methods encompassing Monte Carlo simulation and computational fluid dynamics. These methods focus on quantifying system uncertainties or performing three-dimensional flow field simulations, and are often used for probabilistic analysis and refined studies of complex flow fields. The third category is pure data-driven machine learning methods utilizing algorithms such as neural networks. These methods train models using historical data to predict ventilation parameters and have the ability to handle nonlinear mapping relationships, making them a hot research area in intelligent ventilation in recent years. These three categories constitute the main technical system in the current field of underground engineering ventilation, playing their respective roles in different applicable scenarios.

[0004] The main drawback of the first type of deterministic mathematical model is that it is essentially a steady-state model, making it difficult to adapt to the highly dynamic environment during construction. The Hardy Cross method relies on preset fixed resistance coefficients and a static network topology. However, during the construction of underground cavern complexes, the network topology changes daily with the tunneling progress, and the random movement of construction machinery causes local resistance to exhibit high nonlinearity and time-varying characteristics. Traditional deterministic models cannot adaptively represent the fluctuations in resistance parameters caused by dynamic environmental changes, resulting in significant deviations between theoretical calculations and actual field measurements, and failing to meet the requirements for real-time control.

[0005] For the second type of stochastic and numerical simulation methods, the core limitation lies in the excessive computational load and lack of deterministic guidance. Monte Carlo methods typically require thousands of iterations to converge, and this enormous computational demand makes them unsuitable for real-time control systems in construction sites that require millisecond or second-level responses. Furthermore, this method focuses on probabilistic statistical analysis of uncertainties rather than directly providing deterministic airflow allocation optimization commands, limiting its practicality in energy consumption optimization scenarios.

[0006] The main shortcomings of the third type of purely data-driven method lie in its poor model interpretability and lack of physical constraints. Pure machine learning models typically exist as black boxes, heavily relying on massive amounts of historical labeled data for training. However, underground engineering is significantly unique and non-replicable due to the influence of geological environment and design layout, making it difficult to obtain sufficient historical samples covering all construction conditions for model training. More importantly, purely data-driven models lack physical constraint mechanisms, failing to guarantee that the output strictly follows the laws of conservation of mass and energy, which may lead to the generated control strategies being physically infeasible. Summary of the Invention

[0007] The main objective of this invention is to provide a method for minimizing energy consumption in the construction ventilation network of underground cavern complexes.

[0008] Another objective of this invention is to provide a device for minimizing energy consumption in the construction ventilation network of underground cavern complexes.

[0009] The third objective of this invention is to provide a computer device.

[0010] A fourth objective of this invention is to provide a non-transitory computer-readable storage medium.

[0011] To achieve the above objectives, a first aspect of the present invention proposes a method for minimizing energy consumption in the construction ventilation network of underground cavern groups, comprising:

[0012] A dynamic ventilation network perception and reconstruction model is constructed during the construction period. Multi-source sensors are used to collect real-time environmental parameters and equipment status. Combined with construction progress logs, the topology of the ventilation system is dynamically updated. The physical field baseline is calculated based on the Hardy Cross method. According to the reconstructed topology and the initial frictional resistance of each branch, the theoretical wind volume distribution that satisfies the physical conservation law is output using the Hardy Cross iterative algorithm. Building Hardy Cross The GB physical information-enhanced hybrid prediction model takes the theoretical air volume distribution as the core feature, integrates real-time monitoring data to construct a dataset, trains a gradient boosting regression model, uses the gradient boosting algorithm to learn the nonlinear residual between the physical theoretical value and the actual value, corrects the prediction bias, and obtains the predicted values ​​of air volume and energy consumption. Global energy consumption optimization and regulation are performed based on the hybrid prediction model. The trained hybrid prediction model is used as the prediction core to construct the objective function of minimizing the total power of the system. Under the premise of meeting the air volume requirement constraint, the optimization algorithm is used to search for the optimal frequency conversion strategy of the fan and output control commands to the fan frequency conversion cabinet to adjust the fan speed.

[0013] In one embodiment of the present invention, the construction of the dynamic ventilation network perception and reconstruction model during the construction period includes: Sensors are deployed at pre-defined nodes including intersections, working faces, and fan outlets to collect data on wind speed, wind pressure, temperature, humidity, and gas concentration, as well as fan operation data and construction equipment status data. The collected data is cleaned and spatiotemporally aligned, unifying multi-source data onto the same time axis and mapping it to the corresponding branch of the ventilation network topology; Establish a dynamic topology network update triggering mechanism. When any of the conditions of periodic triggering, event triggering, or threshold triggering are met, the node and branch connection relationships of the ventilation network diagram are automatically updated. For newly added or status-changed branches, their geometric attributes are calculated based on the design cross-section parameters to generate the latest adjacency matrix and association matrix.

[0014] In one embodiment of the present invention, the physical field benchmark solution based on the Hardy Cross method includes: Based on Darcy The Weisbach equations are used to calculate the initial frictional drag of each branch; Using the minimum spanning tree method or inheriting the steady-state airflow results from the previous calculation cycle, the airflow of each branch is initialized under the premise of satisfying the node flow balance. The loop pressure closure difference is calculated, and the correction formula is used for iteration until the maximum pressure closure difference is less than the preset threshold. The theoretical airflow and theoretical resistance distribution of each branch are then output.

[0015] In one embodiment of the present invention, the construction of Hardy Cross GB physical information enhancement hybrid prediction model includes: The input feature vector of the hybrid model is constructed, which includes real-time monitoring data and theoretical airflow and theoretical resistance distributions calculated by the Hardy Cross method; A prediction model is constructed using the gradient boosting decision tree algorithm, with the target variables being actual air volume and actual energy consumption. The key hyperparameters of the model are optimized by combining grid search with cross-validation, and the optimal parameter combination is determined with the goal of minimizing the root mean square error on the validation set.

[0016] In one embodiment of the present invention, the global energy consumption optimization and regulation based on a hybrid prediction model includes: Construct an objective function to minimize the total power of the system, where the shaft power of each wind turbine is predicted by the gradient boosting regression model; Set safety constraints, including that the air volume of each working face branch must be greater than the minimum required air volume and the operating frequency of the fan must be limited to the allowable range; The objective function is solved using the particle swarm optimization algorithm, with the wind turbine frequency as the decision variable, and the optimal solution vector is searched within the constraints. The optimal frequency obtained from the solution is converted into a control signal and sent to the fan inverter cabinet to adjust the fan speed.

[0017] In one embodiment of the present invention, it further includes: Establish a closed-loop feedback and adaptive update mechanism for the system, collect system response data after regulation in real time, calculate the deviation between predicted and measured values, and trigger online correction or retraining of model parameters.

[0018] In one embodiment of the present invention, the establishment of the system closed-loop feedback and model adaptive update mechanism includes: After the control command is executed, the system's new steady-state air volume and energy consumption are collected in real time, and the relative error between the predicted value and the measured value is calculated. If the relative error of a set number of consecutive sampling periods exceeds a set threshold, the latest measured data is added to the training set, the oldest historical data is removed using a sliding window mechanism, and the parameters of the gradient boosting regression model are updated using incremental learning or full retraining.

[0019] To achieve the above objectives, a second aspect of the present invention provides a device for minimizing energy consumption in the construction ventilation network of underground cavern groups, comprising: The dynamic sensing and reconstruction module is used to build a dynamic ventilation network sensing and reconstruction model during the construction period. It uses multi-source sensors to collect real-time environmental parameters and equipment status, and combines them with construction progress logs to dynamically update the topology of the ventilation system. The physical field benchmark solution module is used to perform physical field benchmark solution based on the Hardy Cross method. Based on the reconstructed topology and the initial frictional resistance of each branch, the module uses the Hardy Cross iterative algorithm to output the theoretical air volume distribution that satisfies the physical conservation law. Hybrid prediction model building module, used to build Hardy Cross The GB physical information-enhanced hybrid prediction model takes the theoretical air volume distribution as the core feature, integrates real-time monitoring data to construct a dataset, trains a gradient boosting regression model, uses the gradient boosting algorithm to learn the nonlinear residual between the physical theoretical value and the actual value, corrects the prediction bias, and obtains the predicted values ​​of air volume and energy consumption. The energy consumption optimization and control module is used to perform global energy consumption optimization and control based on the hybrid prediction model. Taking the trained hybrid prediction model as the prediction core, it constructs the objective function of minimizing the total power of the system. Under the premise of meeting the air volume requirement constraint, it uses the optimization algorithm to search for the optimal frequency conversion strategy of the fan and outputs control commands to the fan frequency conversion cabinet to adjust the fan speed.

[0020] To achieve the above objectives, a third aspect of this application provides a computer device comprising a processor and a memory; wherein the processor runs a program corresponding to the executable program code by reading executable program code stored in the memory, for implementing a method for minimizing energy consumption of a construction ventilation network for underground cavern groups as described in the first aspect embodiment.

[0021] To achieve the above objectives, the fourth aspect of this application proposes a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a method for minimizing energy consumption of a construction ventilation network for an underground cavern complex as described in the first aspect embodiment.

[0022] The embodiments of the present invention have the following beneficial effects: This invention constructs a hybrid prediction architecture with enhanced physical information by integrating the Hardy Cross deterministic physics model and the gradient boosting data-driven model. This effectively solves the bottleneck problem of traditional methods struggling to balance physical conservation constraints with the nonlinear characteristics of complex on-site environments. Simultaneously, this invention constructs a dynamic ventilation network perception and reconstruction model during construction, combining closed-loop feedback and adaptive update mechanisms to achieve accurate simulation and real-time energy efficiency optimization of the dynamic ventilation network. This significantly reduces the power consumption of the construction ventilation system and ensures optimal control performance throughout the entire construction cycle. Attached Figure Description

[0023] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 A flowchart of a method for minimizing energy consumption in the construction ventilation network of an underground cavern complex, provided as an embodiment of the present invention; Figure 2 A flowchart illustrating the construction and training of the Hardy Cross-GB hybrid prediction model provided in this embodiment of the invention; Figure 3 The global energy consumption optimization and closed-loop adaptive control logic diagram based on the hybrid model provided in this embodiment of the invention; Figure 4 This is a structural diagram of a device for minimizing energy consumption in the construction ventilation network of an underground cavern complex, provided as an embodiment of the present invention. Detailed Implementation

[0024] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0025] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0026] Traditional ventilation technologies for underground cavern construction have several significant shortcomings, including: deterministic mathematical models cannot adapt to nonlinear resistance fluctuations when dealing with dynamic topological changes, resulting in large prediction deviations and making it difficult to meet real-time control requirements; stochastic simulation methods have extremely high computational loads and cannot achieve second-level response, making them unsuitable for real-time on-site control; and purely data-driven machine learning models suffer from poor interpretability and weak generalization ability due to a lack of physical constraints and a high dependence on massive historical samples.

[0027] To address the problems of poor model dynamic adaptability, insufficient real-time computation, and lack of physical consistency in existing technologies, this invention proposes a hybrid intelligent ventilation system integrating the Hardy Cross physical model and the Gradient Boosting (GB) machine learning algorithm. Specifically, this invention first uses a dynamically reconstructed Hardy Cross model to calculate the baseline physical field, ensuring that airflow distribution satisfies the basic conservation laws of the fluid network. Then, the physical calculation results are used as prior features input into the GB model, leveraging its powerful nonlinear fitting capabilities to correct residuals caused by equipment movement and interference from local components, thereby achieving high-precision prediction of complex dynamic conditions. This "physically guided data" hybrid architecture not only compensates for the inability of traditional deterministic models to handle dynamic nonlinear resistance but also effectively reduces the dependence of machine learning models on the amount of training samples.

[0028] In terms of energy consumption optimization, traditional methods are usually designed based on static worst-case operating conditions, resulting in wind turbines operating in a state of ineffective high energy consumption for extended periods. This invention constructs a global energy consumption minimization objective function based on the prediction results of a hybrid model. It uses an intelligent optimization algorithm to calculate the optimal operating frequency of the wind turbine in real time, achieving on-demand air supply and significantly improving energy utilization efficiency. Furthermore, this system integrates an environmental perception and feedback correction mechanism, using real-time monitoring data to perform closed-loop correction of the model, ensuring the robustness and effectiveness of the control strategy throughout the entire construction cycle.

[0029] The following describes a method for minimizing energy consumption in the construction ventilation network of an underground cavern complex, according to an embodiment of the present invention, with reference to the accompanying drawings.

[0030] Example 1 This embodiment provides a method for minimizing energy consumption in the ventilation network during the construction of an underground cavern complex. The main powerhouse has a designed excavation cross-sectional area of ​​A = 400 m², and the auxiliary guide tunnel has a cross-sectional area of ​​A = 45 m². The main ventilation system uses two 110kW axial flow variable frequency fans. Figure 1 As shown, the method includes the following steps: S1. Construct a dynamic ventilation network perception and reconstruction model during the construction period. Use multi-source sensors to collect real-time environmental parameters and equipment status, and combine them with construction progress logs to dynamically update the topology of the ventilation system.

[0031] Specifically, S1 includes the following steps: S11: Sensors are deployed at pre-defined nodes including intersections, working faces, and fan outlets to collect data on wind speed, wind pressure, temperature, humidity, and gas concentration, as well as fan operation data and construction equipment status data.

[0032] Specifically, the first step is to deploy and collect data from multiple sources. Wind speed, wind pressure, temperature, humidity, and gas concentration sensors are placed at key nodes in the cavern complex (intersections, working faces, and fan outlets) to ensure monitoring coverage. The number of sensors is determined based on the cavern size, and the recorded parameters include wind speed (…). ), static pressure ( ), dry bulb temperature ( ), wet-bulb temperature ) and timestamp. Simultaneously collect wind turbine operating data (voltage) U Current I ,frequency f ) and the status of construction equipment (type, quantity, location).

[0033] S12 cleans and aligns the collected data in time and space, unifying multi-source data onto the same time axis and mapping it to the corresponding branch of the ventilation network topology.

[0034] Specifically, data cleaning and spatiotemporal alignment are performed. The raw data is preprocessed to remove outliers caused by sensor malfunctions or signal interference (using the 3σ criterion). The clocks of various sensors are synchronized using the Network Time Protocol (NTP) to unify environmental data, fan data, and construction progress data sampled at different frequencies onto the same time axis, and the monitoring data is mapped to the corresponding branches of the ventilation network topology based on spatial coordinates.

[0035] S13. Establish a dynamic topology network update triggering mechanism. When any of the conditions of periodic triggering, event triggering, or threshold triggering are met, the node and branch connection relationships of the ventilation network diagram are automatically updated. For newly added or status-changed branches, their geometric attributes are calculated according to the design section parameters to generate the latest adjacency matrix and association matrix.

[0036] Specifically, dynamic topology network reconstruction is implemented. A multi-dimensional triggering mechanism for dynamic topology network updates is established. When any of the following conditions are met—periodic triggering (based on excavation progress data in daily construction logs), event-based triggering (such as key construction events like tunnel breakthrough, significant extension or bending of ducts, and fan relocation), or threshold-based triggering (when abnormal fluctuations in wind speed or pressure in a branch exceed a preset threshold, indicating a sudden change in local resistance)—the system automatically updates the node and branch connections in the ventilation network diagram. For newly added excavation branches or branches with changed status, their initial geometric properties (length L, area A, perimeter P) are calculated based on the design cross-sectional parameters, generating the latest adjacency and correlation matrices. This accurately reflects the high-frequency dynamic changes in the physical boundaries on site, providing an accurate geometric basis for subsequent physical field calculations using the Hardy Cross method.

[0037] In this embodiment, the sensor collects the wind speed at a key node. = 2.8 m / s, static pressure = 1250 Pa, dry bulb temperature = 28.5℃. Based on the construction log of the day, the tunnel underwent periodic advances, with an additional excavation length L = 4.5 m. The system reconstructed the topology accordingly and calculated the initial geometric properties of the newly added branch (perimeter P = 24.5 m, area A = 45 m²).

[0038] S2, based on the Hardy Cross method, performs physical field benchmark calculations. According to the reconstructed topology and the initial frictional resistance of each branch, the Hardy Cross iterative algorithm is used to output the theoretical air volume distribution that satisfies the physical conservation law.

[0039] Specifically, S2 includes the following steps: S21, based on Darcy The Weisbach equation is used to calculate the initial frictional drag of each branch.

[0040] Specifically, the initial frictional drag of each branch is calculated based on the Darcy-Weisbach Equation. The calculation formula is:

[0041] in, The coefficient of friction of the tunnel. For branch length, For the perimeter, This represents the cross-sectional area.

[0042] S22 uses the minimum spanning tree method or inherits the steady-state airflow results from the previous calculation cycle to initialize the airflow of each branch under the premise of satisfying the node flow balance, calculates the loop pressure closure difference, and uses the correction formula to iterate until the maximum pressure closure difference is less than the preset threshold, and outputs the theoretical airflow and theoretical resistance distribution of each branch.

[0043] Specifically, nodal flow balance equations and loop pressure balance equations are constructed based on Kirchhoff's laws. The nodal flow balance equation is as follows:

[0044] in, , These are the volumetric airflow at the inflow and outflow nodes, respectively, in m³ / s, with the positive direction being the direction of flow towards the node.

[0045] The circuit pressure balance equation is:

[0046] in, It is the algebraic sum of the pressure losses of each branch in a single loop of the ventilation network, i.e., the loop pressure closure difference.

[0047] Furthermore, using the minimum spanning tree method or directly inheriting the steady-state airflow results from the previous calculation cycle, the airflow of each branch is initialized under the premise of strictly satisfying the above-mentioned node flow balance. Calculate the pressure closure difference of the loop. (and (The meaning is the same), and iteratively using the corrected formula, the expression is:

[0048]

[0049] in, For branch air volume, This is the air volume correction amount. This is the new branch airflow output in this iteration after airflow correction. This refers to the branch air volume calculated in the previous round before the start of this iteration.

[0050] Iterate until the maximum pressure closure difference is less than a preset threshold (this threshold is set according to the accuracy requirements of the on-site engineering, usually the absolute value of the maximum pressure closure difference ≤ 0.1~1.0 Pa, or the relative correction rate of a single air volume ≤ 0.1%), and output the theoretical air volume of each branch. and theoretical resistance distributed.

[0051] In this embodiment, the friction coefficient of the excavation tunnel in the granite rock mass is taken as... = 0.012Ns² / m 4 Substituting the values ​​into Darcy's formula, the initial wind resistance of the newly added branch is calculated. = 0.0000145 N·s² / m 8 In the Hardy Cross iteration, the maximum pressure closure difference convergence threshold is set to ≤0.5Pa, and the branch airflow is initialized. = 15.0 m³ / s, after iterative calculation, the theoretical air volume of this branch is output. =15.6m³ / s.

[0052] S3, construct Hardy Cross The GB physical information-enhanced hybrid prediction model takes the theoretical air volume distribution as the core feature, integrates real-time monitoring data to construct a dataset, trains a gradient boosting regression model, and uses the gradient boosting algorithm to learn the nonlinear residual between the physical theoretical value and the actual value, corrects the prediction bias, and obtains the predicted values ​​of air volume and energy consumption.

[0053] like Figure 2 As shown, this is Hardy Cross. The GB physical information-enhanced hybrid prediction model is the overall process of construction and training.

[0054] Specifically, S3 includes the following steps: S31, construct the input feature vector of the hybrid model. The input feature vector contains real-time monitoring data as well as theoretical airflow and theoretical resistance distributions calculated by the Hardy Cross method.

[0055] Specifically, physical information enhancement feature engineering. This involves constructing the input feature vector for the hybrid model. X It includes not only real-time monitoring data (ambient temperature) T Quantity of equipment N eq Fan frequency f It also explicitly includes the theoretical air volume calculated by S2. and theoretical resistance This design embeds prior knowledge from the physical model into the machine learning model, allowing it to focus on learning the nonlinear residuals that the physical model cannot explain.

[0056] S32 uses a gradient boosting decision tree algorithm to construct a prediction model, with the target variables being actual air volume and actual energy consumption.

[0057] Specifically, a gradient boosting (GB) regression model is constructed. The gradient boosting decision tree (GBDT) algorithm is used to build the prediction model, with the actual air volume as the target variable. and actual energy consumption The model combines multiple base learners (decision trees) in an additive model manner. The model in the m-th iteration is represented as:

[0058] in, For learning rate, To fit the basis decision tree of the current residual, This represents the predicted output of the gradient boosting model after the m-th iteration. For the mth The predicted output of the gradient boosting model after one iteration.

[0059] S33 utilizes grid search combined with cross-validation to optimize the key hyperparameters of the model, aiming to determine the optimal parameter combination by minimizing the root mean square error on the validation set.

[0060] Specifically, hyperparameter adaptive optimization is employed. Grid search combined with K-fold cross-validation (K=5) is used to optimize the model's key hyperparameters. Optimized parameters include: the number of trees (n_estimators), the maximum depth (max_depth), the learning rate (learning_rate), and the minimum number of leaf node samples (min_samples_leaf). The optimal parameter combination is determined with the goal of minimizing the root mean square error (RMSE) on the validation set, preventing overfitting and improving generalization ability.

[0061] In this embodiment, the GB model integrates the theoretical air volume and the sensor-measured air volume (14.2 m³ / s) to specifically learn the nonlinear residual of 1.4 m³ / s. The specific values ​​of the hyperparameters of the grid search optimized model are: number of trees n_estimators = 150, maximum depth max_depth = 5, and learning rate learning_rate = 0.05.

[0062] S4. Based on the hybrid prediction model, global energy consumption optimization and regulation are performed. The trained hybrid prediction model is used as the prediction core to construct the objective function of minimizing the total power of the system. Under the premise of meeting the air volume requirement constraint, the optimization algorithm is used to search for the optimal frequency conversion strategy of the fan and output control commands to the fan frequency conversion cabinet to adjust the fan speed.

[0063] like Figure 3 As shown, this is the global energy consumption optimization and closed-loop adaptive control logic based on a hybrid model.

[0064] Specifically, S4 includes the following steps: S41, construct the objective function to minimize the total power of the system, where the shaft power of each wind turbine is predicted by the gradient boosting regression model.

[0065] Specifically, we construct an objective function to minimize energy consumption. We define the total system power as the optimization objective, and its expression is:

[0066] Where J is the objective function for total system power, and M is the number of wind turbines. For the first i Typhoon generators at frequency The shaft power can be predicted using the GB model.

[0067] S42, set safety constraints, including that the air volume of each working face branch must be greater than the minimum required air volume and the fan operating frequency is limited to the allowable range.

[0068] Specifically, airflow constraints are required: airflow at each work area branch. Requires a volume greater than the minimum required air volume (The required air volume to remove fumes and harmful gases); Wind turbine operating constraints: The wind turbine operating point must be located within the stable operating region of the characteristic curve to avoid surge, and the frequency... Limited to the permitted range Inside, among which The minimum frequency required for stable operation of the wind turbine. The maximum frequency for safe operation of the wind turbine.

[0069] S43 uses the particle swarm optimization algorithm to solve the objective function, taking the wind turbine frequency as the decision variable, and searches for the optimal solution vector within the constraints.

[0070] Specifically, a global optimization solution is used. The particle swarm optimization algorithm is employed to solve the objective function. The operating parameters of the particle swarm optimization algorithm are set: the particle population size is initialized (e.g., 30-50 particles) and the maximum number of iterations is set (e.g., 100-200 iterations). The wind turbine frequency is then... As a decision variable, search for the optimal solution vector within the constraints:

[0071] in, This represents the optimal operating frequency vector for the wind turbine. This is the optimal frequency combination array for all wind turbines. The optimal operating frequency for fan #1 The optimal operating frequency for fan #2 The optimal operating frequency for fan M is given by M, where M is the total number of fans in the system.

[0072] S44 converts the optimal frequency obtained from the solution into a control signal and sends it to the fan inverter cabinet to adjust the fan speed.

[0073] Specifically, control commands are issued. The optimal frequency obtained from the solution is then... The signal is converted into a control signal for the frequency converter and sent to each fan frequency converter cabinet through the PLC system to adjust the fan speed and achieve energy-saving operation of the system.

[0074] In this embodiment, the minimum required air volume constraint for this branch is set based on the number of 3 diesel generators and operators on site. =12.5 m³ / s; the safe frequency range for the fan inverter is set as follows: = 30 Hz to = 50 Hz. The particle swarm optimization algorithm is set with 40 particles and a maximum number of iterations of 150. After optimization, the optimal frequency control command that meets the required air volume and minimizes the total energy consumption is output: 38.5Hz for fan #1 and 36.0Hz for fan #2.

[0075] In this embodiment, a closed-loop feedback and model adaptive update mechanism is also established to collect system response data after regulation in real time, calculate the deviation between the predicted value and the measured value, and trigger online correction or retraining of model parameters.

[0076] Specifically, this step includes: (1) After the control command is executed, the new steady-state air volume and energy consumption of the system are collected in real time, and the relative error between the predicted value and the measured value is calculated.

[0077] Specifically, this involves implementing effect monitoring and deviation calculation. After the control command is executed, the system's new steady-state airflow is collected in real time. and energy consumption Calculate the relative error between the predicted and measured values:

[0078] in, The relative error between the predicted and measured values. To predict air volume for the model, The measured steady-state air volume after adjustment.

[0079] (2) If the relative error of a number of consecutive sampling periods exceeds the set threshold, the latest measured data will be added to the training set, the oldest historical data will be removed by the sliding window mechanism, and the parameters of the gradient boosting regression model will be updated by incremental learning or full retraining.

[0080] Specifically, the model employs an online update mechanism. If the error ε over N consecutive sampling periods exceeds a threshold, the model update mechanism is triggered. The latest measured data is added to the training set, and the oldest historical data is removed using a sliding window mechanism. The GB model parameters are updated using incremental learning or full retraining to adapt to the long-term evolution of construction environments such as tunnel connection and duct aging. Through this closed-loop mechanism and hybrid prediction model, this system achieves significant advantages over existing technologies: Compared to the traditional static Hardy Cross method, which often exhibits large deviations when facing nonlinear dynamic resistance (traditional methods typically have airflow prediction deviations of around 15% to 25%), this system, using physical field benchmarks and GB residual correction, can significantly reduce airflow prediction deviations to within 5%. Simultaneously, thanks to high-precision nonlinear fitting, the overall energy consumption prediction error of the system can be stably controlled within ≤5%, truly achieving energy efficiency optimization that balances physical constraints and high-precision dynamic prediction.

[0081] In this embodiment, after the control command is issued and stabilized, the new steady-state air volume of the system is measured. = 13.1 m³ / s, calculate the relative error ε between the predicted value (e.g., 13.0 m³ / s) and the measured value, ε ≈ 0.76%. The system is set to trigger a model update based on the average error threshold over N = 10 consecutive sampling periods (i.e., 10 minutes). >5%. Under this specific operating condition, compared with the traditional Hardy Cross method which typically has an air volume prediction error of up to 18%, this system stably reduces the prediction error to about 3%, and controls the overall energy consumption prediction error to ≤ 4.5%, successfully achieving high-precision on-demand air supply in underground spaces.

[0082] Example 2 This invention also provides a device for minimizing energy consumption in the construction ventilation network of underground cavern groups, such as... Figure 4 As shown, the device 10 includes: The dynamic sensing and reconstruction module 100 is used to build a dynamic ventilation network sensing and reconstruction model during the construction period. It uses multi-source sensors to collect real-time environmental parameters and equipment status, and combines the construction progress log to dynamically update the topology of the ventilation system. The physical field benchmark solution module 200 is used to perform physical field benchmark solution based on the Hardy Cross method. Based on the reconstructed topology and the initial frictional resistance of each branch, the module uses the Hardy Cross iterative algorithm to output the theoretical air volume distribution that satisfies the physical conservation law. Hybrid Prediction Model Building Module 300 is used to build Hardy Cross The GB physical information-enhanced hybrid prediction model takes the theoretical air volume distribution as the core feature, integrates real-time monitoring data to construct a dataset, trains a gradient boosting regression model, uses the gradient boosting algorithm to learn the nonlinear residual between the physical theoretical value and the actual value, corrects the prediction bias, and obtains the predicted values ​​of air volume and energy consumption. The energy consumption optimization and control module 400 is used to perform global energy consumption optimization and control based on the hybrid prediction model. The trained hybrid prediction model is used as the prediction core to construct the objective function of minimizing the total power of the system. Under the premise of meeting the air volume requirement constraint, the optimization algorithm is used to search for the optimal frequency conversion strategy of the fan and output control commands to the fan frequency conversion cabinet to adjust the fan speed.

[0083] Example 3 To implement the methods of the above embodiments, the present invention also provides a computer device, which includes a memory and a processor; wherein the processor runs a program corresponding to the executable program code by reading executable program code stored in the memory, so as to implement the various steps of the methods described above.

[0084] Example 4 To implement the above embodiments, this application also proposes a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method described in the foregoing embodiments.

[0085] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0086] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0087] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. A method for minimizing energy consumption in the construction ventilation network of an underground cavern complex, characterized in that, Includes the following steps: A dynamic ventilation network perception and reconstruction model is constructed during the construction period. Multi-source sensors are used to collect real-time environmental parameters and equipment status. Combined with construction progress logs, the topology of the ventilation system is dynamically updated. The physical field baseline is calculated based on the Hardy Cross method. According to the reconstructed topology and the initial frictional resistance of each branch, the theoretical wind volume distribution that satisfies the physical conservation law is output using the Hardy Cross iterative algorithm. Building Hardy Cross The GB physical information-enhanced hybrid prediction model takes the theoretical air volume distribution as the core feature, integrates real-time monitoring data to construct a dataset, trains a gradient boosting regression model, uses the gradient boosting algorithm to learn the nonlinear residual between the physical theoretical value and the actual value, corrects the prediction bias, and obtains the predicted values ​​of air volume and energy consumption. Global energy consumption optimization and regulation are performed based on the hybrid prediction model. The trained hybrid prediction model is used as the prediction core to construct the objective function of minimizing the total power of the system. Under the premise of meeting the air volume requirement constraint, the optimization algorithm is used to search for the optimal frequency conversion strategy of the fan and output control commands to the fan frequency conversion cabinet to adjust the fan speed.

2. The method according to claim 1, characterized in that, The construction of the dynamic ventilation network perception and reconstruction model during the construction period includes: Sensors are deployed at pre-defined nodes including intersections, working faces, and fan outlets to collect data on wind speed, wind pressure, temperature, humidity, and gas concentration, as well as fan operation data and construction equipment status data. The collected data is cleaned and spatiotemporally aligned, unifying multi-source data onto the same time axis and mapping it to the corresponding branch of the ventilation network topology; Establish a dynamic topology network update triggering mechanism. When any of the conditions of periodic triggering, event triggering, or threshold triggering are met, the node and branch connection relationships of the ventilation network diagram are automatically updated. For newly added or status-changed branches, their geometric attributes are calculated based on the design cross-section parameters to generate the latest adjacency matrix and association matrix.

3. The method according to claim 1, characterized in that, The physical field benchmark solution based on the Hardy Cross method includes: Based on Darcy The Weisbach equations are used to calculate the initial frictional drag of each branch; Using the minimum spanning tree method or inheriting the steady-state airflow results from the previous calculation cycle, the airflow of each branch is initialized under the premise of satisfying the node flow balance. The loop pressure closure difference is calculated, and the correction formula is used for iteration until the maximum pressure closure difference is less than the preset threshold. The theoretical airflow and theoretical resistance distribution of each branch are then output.

4. The method according to claim 1, characterized in that, The construction of Hardy Cross GB physical information enhancement hybrid prediction model includes: The input feature vector of the hybrid model is constructed, which includes real-time monitoring data and theoretical airflow and theoretical resistance distributions calculated by the HardyCross method; A prediction model is constructed using the gradient boosting decision tree algorithm, with the target variables being actual air volume and actual energy consumption. The key hyperparameters of the model are optimized by combining grid search with cross-validation, and the optimal parameter combination is determined with the goal of minimizing the root mean square error on the validation set.

5. The method according to claim 1, characterized in that, The global energy consumption optimization and control based on the hybrid prediction model includes: Construct an objective function to minimize the total power of the system, where the shaft power of each wind turbine is predicted by the gradient boosting regression model; Set safety constraints, including that the air volume of each working face branch must be greater than the minimum required air volume and the operating frequency of the fan must be limited to the allowable range; The objective function is solved using the particle swarm optimization algorithm, with the wind turbine frequency as the decision variable, and the optimal solution vector is searched within the constraints. The optimal frequency obtained from the solution is converted into a control signal and sent to the fan inverter cabinet to adjust the fan speed.

6. The method according to claim 1, characterized in that, Also includes: Establish a closed-loop feedback and adaptive update mechanism for the system, collect system response data after regulation in real time, calculate the deviation between predicted and measured values, and trigger online correction or retraining of model parameters.

7. The method according to claim 6, characterized in that, The establishment of the system closed-loop feedback and model adaptive update mechanism includes: After the control command is executed, the system's new steady-state air volume and energy consumption are collected in real time, and the relative error between the predicted value and the measured value is calculated. If the relative error of a set number of consecutive sampling periods exceeds a set threshold, the latest measured data is added to the training set, the oldest historical data is removed using a sliding window mechanism, and the parameters of the gradient boosting regression model are updated using incremental learning or full retraining.

8. An integrated Hardy Cross GB's ventilation network energy consumption minimization system is characterized by... The system includes: The dynamic sensing and reconstruction module is used to build a dynamic ventilation network sensing and reconstruction model during the construction period. It uses multi-source sensors to collect real-time environmental parameters and equipment status, and combines them with construction progress logs to dynamically update the topology of the ventilation system. The physical field benchmark solution module is used to perform physical field benchmark solution based on the Hardy Cross method. Based on the reconstructed topology and the initial frictional resistance of each branch, the module uses the Hardy Cross iterative algorithm to output the theoretical air volume distribution that satisfies the physical conservation law. Hybrid prediction model building module, used to build Hardy Cross The GB physical information-enhanced hybrid prediction model takes the theoretical air volume distribution as the core feature, integrates real-time monitoring data to construct a dataset, trains a gradient boosting regression model, uses the gradient boosting algorithm to learn the nonlinear residual between the physical theoretical value and the actual value, corrects the prediction bias, and obtains the predicted values ​​of air volume and energy consumption. The energy consumption optimization and control module is used to perform global energy consumption optimization and control based on the hybrid prediction model. Taking the trained hybrid prediction model as the prediction core, it constructs the objective function of minimizing the total power of the system. Under the premise of meeting the air volume requirement constraint, it uses the optimization algorithm to search for the optimal frequency conversion strategy of the fan and outputs control commands to the fan frequency conversion cabinet to adjust the fan speed.

9. A computer device, characterized in that, Including processor and memory; The processor reads executable program code stored in the memory to run a program corresponding to the executable program code, so as to implement the method for minimizing energy consumption of underground cavern group construction ventilation network as described in any one of claims 1-7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements a method for minimizing energy consumption in the construction ventilation network of an underground cavern complex as described in any one of claims 1-7.

Images