Subway station air conditioner load adjusting system and method

Abstract

The invention discloses a subway station air conditioning load adjusting system and method, and the system comprises a load simulation module which is used for simulating the time sequence change of cold loads and heat loads of different types of stations under different meteorological conditions and operation conditions, and generating a system training data set; the passenger travel behavior modeling module is used for establishing a time-period passenger distribution model by using subway historical passenger flow data and urban public transport survey data; outputting passenger flow densities of different areas by using a time period passenger flow distribution model; the machine learning prediction model is used for extracting multi-dimensional features as input features; a system training data set is utilized to learn the mapping relation between the input features and the target cold / heat load, and a load preset value is output; the intelligent load adjusting control module dynamically adjusts the output capacity of the air conditioning system through the building automation system according to the load preset value, and closed-loop control is achieved. According to the invention, TRNSYS simulation, behavior modeling and intelligent control are integrated, and accurate load prediction and energy-saving regulation and control are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of air-conditioning load regulation, and more particularly to a system and method for regulating the air-conditioning load of a subway station. Background Art

[0002] Currently, subway station air conditioning systems generally adopt a design approach with independent air supply for the upper and lower floors, meaning that separate air supply systems are installed for the concourse and platform floors. While this approach facilitates layered control during the initial design phase, it presents numerous drawbacks during operation. First, this design significantly increases the amount of material used for air conditioning ducts, which not only raises initial construction costs but also increases the difficulty of system maintenance. Second, because the concourse air supply ducts are typically located in the ceiling area, the cooling or heat they transport can easily escape upward through the ceiling structure into the soil layer. This heat loss is particularly significant at night or during periods of low passenger flow, resulting in energy waste and inefficient operation.

[0003] In addition, current subway air conditioning load calculations generally use a static algorithm based on the daily average cooling and heating loads. This algorithm calculates the daily average load value based on historical operating data or standard operating conditions and uses this as the basis for system operation. However, this method ignores the significant dynamic changes in passenger flow density within subway stations. For example, the passenger flow density during the morning and evening peak hours is much higher than the daily average, while it is relatively sparse during the lunch and late night hours. This fluctuation in passenger flow at different times directly leads to a large deviation between the actual cooling and heating loads and the static load prediction values, which leads to a series of problems:

[0004] 1) During off-peak hours, the system still operates at high load, resulting in excessive air conditioning output, causing the station temperature to be too low, making passengers feel cold and uncomfortable, and increasing unnecessary energy consumption;

[0005] 2) During peak hours, due to insufficient load forecasting, the air conditioning system cannot respond to actual demand in a timely manner, resulting in stuffy air and poor ventilation in the station, seriously affecting the passenger comfort experience;

[0006] 3) Control lag: Traditional air conditioning systems typically use fixed strategies or control methods based on fixed schedules. They lack the ability to dynamically sense and respond to real-time environmental parameters (such as temperature, humidity, and occupancy density). This results in system control lag and makes it difficult to achieve on-demand energy supply.

[0007] 4) Lack of long-term simulation and prediction capabilities: Existing methods are mostly based on short-term data or empirical formulas. They lack the ability to simulate and model long-term cooling and heating loads under different climate conditions throughout the year. This makes it impossible to accurately predict future load trends, limiting the level of intelligent system operation.

[0008] 5) Lack of multi-factor coupling analysis: The cooling and heating loads at subway stations are not only related to passenger flow density but also influenced by a variety of factors, including outdoor meteorological conditions (such as temperature, humidity, and solar radiation), the operating status of station equipment (such as lighting, elevators, and escalators), and the operating mode of the ventilation system. Existing methods often ignore the coupling relationships between these variables, resulting in significant deviations in prediction results, impacting system efficiency and energy conservation.

[0009] In summary, the current subway station air-conditioning system has prominent problems in its design and operation, such as high energy consumption, delayed control, inaccurate prediction, and poor comfort. There is an urgent need for a new air-conditioning load regulation method that can combine dynamic passenger flow prediction, multi-source data fusion, intelligent control, and year-round load simulation to achieve efficient operation and sustainable development of subway station air-conditioning systems. Summary of the Invention

[0010] The purpose of the present invention is to provide a system and method for regulating the air-conditioning load of subway stations. By integrating TRNSYS load simulation, passenger travel behavior modeling, random forest regression prediction and intelligent load regulation and control, accurate prediction and dynamic energy-saving regulation of the cooling / heating load of subway stations can be achieved.

[0011] To achieve the above object, the present invention provides the following technical solutions:

[0012] A subway station air conditioning load regulation system includes a load simulation module for simulating the temporal changes in cooling and heating loads at different types of stations under different meteorological conditions and operating conditions, and generating a system training data set;

[0013] The passenger travel behavior modeling module uses historical subway passenger flow data and urban public transportation survey data to establish a time-based passenger distribution model that can reflect the travel habits of urban residents. This time-based passenger flow distribution model describes the dynamic changes in passenger flow at an hourly granularity and maps it to the population density distribution in different areas of the station, outputting the passenger flow density in different areas.

[0014] The machine learning prediction module, with a built-in random forest regression algorithm, is used to extract multi-dimensional features of passenger flow density, external meteorological data, and historical load data as input features to establish a machine learning prediction model. The machine learning prediction model is trained using the system training data set to learn the mapping relationship between input features and target cooling / heating loads, and output preset load values ​​at different time points.

[0015] The intelligent load regulation control module dynamically adjusts the output capacity of the air-conditioning system through the building automation system according to the preset load values ​​at different time nodes to achieve closed-loop control. The dynamic adjustment of the output capacity of the air-conditioning system includes: start and stop control of the cold / heat source equipment, air volume control and supply air temperature adjustment.

[0016] Furthermore, the machine learning prediction module also includes: a data input unit, a data preprocessing unit, and a feature engineering unit;

[0017] The data input unit is used to input passenger flow density, external weather and historical load data to obtain an original data set;

[0018] The data preprocessing unit is used to preprocess the original data set to eliminate the dimensional influence between different feature dimensions; the preprocessing process includes: cleaning and standardization;

[0019] The feature engineering unit is used to extract the attributes of multi-dimensional features from the pre-processed data and construct a feature matrix to achieve on-demand diversion of multi-dimensional features in the machine learning prediction model; the multi-dimensional features include: multi-dimensional time features, spatial features and passenger behavior features

[0020] Furthermore, the data sources of the system training dataset are specifically: air conditioning load data for 8,760 hours a year from TRNSYS simulation; timestamp and origin and destination passenger flow data recorded by the subway AFC system; temperature, humidity, and CO2 concentration information provided by meteorological monitoring stations; and statistical data on passenger travel purposes and transfer rates from urban traffic surveys.

[0021] Furthermore, the processing flow of the data pre-processing unit is specifically as follows:

[0022] The original data were smoothed using the sliding window method and aligned to a time granularity of 15 minutes. The STL method was then used for outlier detection. For outliers with residuals greater than three times the standard deviation, they were replaced with the mean of the adjacent time periods. Finally, all features were normalized by Z-Score to eliminate the dimensional effects between different feature dimensions.

[0023] Furthermore, the multi-dimensional time features include hour periods, week types and holiday signs; the spatial features cover the station hall heat map and platform passenger flow density distribution; the passenger behavior features include: weighted processing of passenger detention time and transfer paths. Different types of features are used as needed in subsequent modeling.

[0024] Furthermore, the training process of the machine learning prediction model is as follows: LightGBM is used as the basic model for initialization, and core parameters such as the number of leaf nodes are set to 127; then, the Bayesian optimization algorithm is used to search for hyperparameters, and the search space is mainly concentrated in the two dimensions of min_data_in_leaf∈[50,200] and lambda_l1∈[0,5]; the machine learning prediction model training adopts a multi-scale input method, which simultaneously accesses short-term features and long-term features.

[0025] Furthermore, the validation process of the machine learning prediction model is as follows: in the offline stage, the performance evaluation is performed through 5-fold cross validation to ensure that the determination coefficient R 2 Greater than 0.9; during the deployment phase, an incremental learning framework is built, and the full model is set to be updated every 72 hours. At the same time, the model is fine-tuned with real-time data every 15 minutes to ensure the timeliness and accuracy of the prediction results.

[0026] Furthermore, the intelligent load regulation control module controls the output capacity of the air conditioning system as follows: the air conditioning load result predicted by the machine learning prediction model is connected to the interface of the building automation system; specifically, the air volume setting is adjusted according to the predicted value P_pred: Q = α × P_pred + β; and the target set temperature T_set = f(COP, P_pred) of the air conditioning system is determined based on the energy efficiency curve and the load prediction value;

[0027] Where Q represents the ventilation volume required in the air conditioning system; α represents the linear proportional factor between the air volume and the predicted load; p represents the load value of the building or equipment; and β represents the reference offset in the air volume setting.

[0028] The present invention also provides a method for regulating an air conditioning load in a subway station, comprising the following steps:

[0029] S1. Based on the TRNSYS simulation platform, simulate the temporal changes of cooling load and heating load of different types of stations under different meteorological conditions and operating conditions to generate system training data sets;

[0030] S2. Utilizing historical subway passenger flow data and urban public transportation survey data, establish a time-based passenger distribution model that reflects the travel habits of urban residents. This time-based passenger flow distribution model describes the dynamic changes in passenger flow at an hourly granularity and maps this to the population density distribution in different areas of the station, outputting the passenger flow density in each area.

[0031] S3. Extract multi-dimensional features of passenger flow density, external weather, and historical load data as input features for the random forest regression algorithm and establish a machine learning prediction model. Utilize the system training dataset and the machine learning prediction model to learn the mapping relationship between the input features and the target cooling / heating load, and output preset load values ​​at different time points.

[0032] S4. Based on the preset load values ​​at different time points, the output capacity of the air-conditioning system is dynamically adjusted through the building automation system to achieve closed-loop control. The dynamic adjustment of the output capacity of the air-conditioning system includes: start and stop control of the cold / heat source equipment, air volume control, and supply air temperature adjustment.

[0033] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0034] The present invention constructs a load simulation module through the TRNSYS simulation platform, combines subway passenger flow data with meteorological conditions, and establishes an accurate cold / heat load time series prediction model, which significantly improves the prediction accuracy and data quality. The introduction of the random forest regression algorithm enhances the generalization ability of the model and can adapt to complex and changing operating environments. At the same time, the passenger travel behavior modeling module refines the passenger flow distribution to the hourly granularity and maps it to the population density in the station area as a key input variable, so that the load forecast is closer to actual needs. The intelligent load regulation control module dynamically optimizes the operation of the air-conditioning system based on the prediction results, realizes the start and stop of the cold / heat source equipment, and accurately adjusts the air volume and supply air temperature, improves energy efficiency and reduces energy consumption. The overall system has closed-loop control capabilities and can be seamlessly connected with the building automation system to achieve efficient, energy-saving and comfortable operation goals, providing advanced and reliable technical support for the intelligent management of urban rail transit stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0036] The following further describes the system and method for regulating the air conditioning load of a subway station according to the present invention with reference to the accompanying drawings.

[0037] Figure 1 This is a schematic diagram of the connections between the modules in the subway station air conditioning load regulation system of the present invention;

[0038] Figure 2 This is a schematic diagram of a subway station load simulation of a subway station air conditioning load regulation system according to the present invention;

[0039] Figure 3 This is a schematic diagram of the energy consumption simulation of the subway station refrigeration unit of the subway station air conditioning load regulation system of the present invention;

[0040] Figure 4 Schematic diagrams of passenger flow simulation in the subway station air conditioning load regulation system of the present invention; (a) is a schematic diagram simulating a situation with no passenger flow, (b) is a schematic diagram simulating a situation with a small passenger flow, and (c) is a schematic diagram simulating a situation with a large passenger flow;

[0041] Figure 5 It is a flow chart of the method for adjusting the air-conditioning load of a subway station according to the present invention. DETAILED DESCRIPTION

[0042] The following embodiments of the present invention are described in further detail with reference to the accompanying drawings and examples. The following examples are used to illustrate the present invention but are not intended to limit the scope of the present invention.

[0043] In order to better understand the purpose, structure and function of the present invention, the present invention is further described in detail below with reference to the accompanying drawings.

[0044] Example 1

[0045] like Figure 1 As shown, the present invention provides a subway station air conditioning load adjustment system, comprising:

[0046] The load simulation module is used to simulate the time series changes of cooling load and heating load of different types of stations under different meteorological conditions and operating conditions, and generate system training data sets, such as Figure 2 and Figure 3 As shown;

[0047] The passenger travel behavior modeling module is used to use the historical subway passenger flow data and urban public transportation survey data to establish a time-based passenger distribution model that can reflect the travel habits of urban residents; the time-based passenger flow distribution model describes the dynamic changes of passenger flow at an hourly granularity and maps it to the population density distribution in different areas of the station, outputting the passenger flow density in different areas, such as Figure 4 As shown;

[0048] The machine learning prediction module has a built-in random forest regression algorithm, which is used to extract multi-dimensional features of passenger flow density, external meteorological and historical load data as input features to establish a machine learning prediction model. The machine learning prediction model is trained using the system training data set to learn the mapping relationship between input features and target cooling / heating loads, and output preset load values ​​at different time nodes. The random forest regression algorithm has the advantages of strong nonlinear fitting ability, insensitivity to feature dimensions, and high prediction stability, making it suitable for load prediction problems in complex subway environments.

[0049] In this embodiment, the selection of the machine learning prediction model is shown in Table 1. The training process of the machine learning prediction model is specifically as follows: LightGBM is initialized as the basic model, and core parameters such as the number of leaf nodes are set to 127; then, the Bayesian optimization algorithm is used to search for hyperparameters, and the search space is mainly concentrated in the two dimensions of min_data_in_leaf∈[50,200] and lambda_l1∈[0,5]; the machine learning prediction model training adopts a multi-scale input method, and simultaneously accesses short-term features (4-hour sliding window) and long-term features (7-day periodic features).

[0050] Table 1 Model selection

[0051]

[0052] The verification process of the machine learning prediction model is as follows: in the offline stage, the performance is evaluated by 5-fold cross validation to ensure the determination coefficient R 2 Greater than 0.9; during the deployment phase, an incremental learning framework is built, and the full model is set to be updated every 72 hours. At the same time, the model is fine-tuned with real-time data every 15 minutes to ensure the timeliness and accuracy of the prediction results.

[0053] In this embodiment, the verification formula of the machine learning prediction model is specifically:

[0054] Average error at each time of the day As evaluation criteria for model calibration:

[0055]

[0056] In the above formula, n is the number of samples; y i is the true value of the i-th sample; is the output value of the RF model for the i-th sample.

[0057] The intelligent load adjustment control module dynamically adjusts the output capacity of the air conditioning system through the building automation system according to the target cooling / heating load to achieve closed-loop control. The dynamic adjustment of the output capacity of the air conditioning system includes: start / stop control of the cooling / heating source equipment, air volume control and supply air temperature adjustment.

[0058] In this embodiment, the intelligent load regulation control module controls the output capacity of the air conditioning system as follows: the air conditioning load result predicted by the machine learning prediction model is connected to the interface of the building automation system; specifically, the air volume setting is adjusted according to the predicted value P_pred: Q = α × P_pred + β; and the target set temperature T_set = f(COP, P_pred) of the air conditioning system is determined based on the energy efficiency curve and the load prediction value.

[0059] Where Q represents the ventilation volume required in the air conditioning system; α represents the linear proportional factor between the air volume and the predicted load; p represents the load value of the building or equipment; and β represents the reference offset in the air volume setting.

[0060] In this embodiment, Q: air volume set value; its physical meaning is: refers to the ventilation volume that needs to be set in the air conditioning system (BAS, building automation system), the unit is usually m 3 / h (cubic meters per hour) or m 3 / s (cubic meters per second); Function: Dynamically adjust the air volume according to the load forecast results to maintain the stability of indoor temperature, humidity, air quality and other parameters, while optimizing energy consumption.

[0061] α: Proportional coefficient; Physical meaning: The linear proportional factor between air volume and predicted load, dimensionless (or determined by unit conversion); Implications: Reflects the sensitivity of load changes to air volume adjustment. For example, if α = 0.5, for every 1 unit increase in load, the air volume increases by 0.5 units. Determination: Typically determined through historical operating data fitting, system characteristics testing, or engineering experience, and must meet system safety and energy efficiency requirements.

[0062] P_pred: Load prediction value; Physical meaning: The building or equipment load value predicted by a model (such as LightGBM), usually expressed in kW (representing heat load or electrical load); Background: P_pred is derived from the model training process described above and integrates multi-dimensional inputs such as TRNSYS simulation data, passenger flow data, and meteorological parameters. It is used to represent energy demand (such as air conditioning cooling load and ventilation load) in future time periods.

[0063] β: offset coefficient

[0064] Physical meaning: The reference offset in air volume setting, the unit is consistent with Q (such as m 3 / h).

[0065] Meaning: Represents the minimum air volume setting value of the system under zero load or basic operating conditions, used to meet basic ventilation needs (such as the amount of fresh air required for people to breathe) and avoid zero air volume.

[0066] Function: The β value ensures that the system can maintain necessary air circulation at low load, thus ensuring indoor air quality and equipment safety.

[0067] COP: Coefficient Of Performance

[0068] Chinese name: performance parameters

[0069] Calculation formula: COP = air conditioning cooling capacity / compressor input power; dimensionless

[0070] f function: represents the relationship between T_set, COP and P_pred. Due to too many exogenous influencing factors in air conditioning and refrigeration, the f function cannot give a specific relationship, but what we can know is that when T_set is constant, the higher the P_pred, the higher the COP of the air conditioning system is required.

[0071] Intelligent load adjustment control module: Dynamically adjusts the output capacity of the air conditioning system based on the predicted load value, including start / stop control of cooling / heating equipment, air volume control, and air supply temperature adjustment. This module can be connected to the BAS (Building Automation System) system to achieve full-process closed-loop control.

[0072] This implementation breaks away from the traditional static calculation method of the "daily average load" approach, enabling the station air conditioning system to dynamically respond to real-time passenger flow fluctuations. Compared to existing systems, this solution improves passenger comfort scores by 15% during peak hours and reduces air conditioning system energy consumption by approximately 22% during off-peak hours. By combining intelligent prediction with adaptive control, subway air conditioning operations have shifted from "experience-based" to "data-driven," significantly improving the intelligence and overall energy efficiency of subway air conditioning systems.

[0073] The intelligent load regulation control module includes: a data input unit, a data preprocessing unit, and a feature engineering unit;

[0074] The data input unit is used to input passenger flow density, external weather and historical load data to obtain an original data set;

[0075] The data preprocessing unit is used to preprocess the original data set to eliminate the dimensional influence between different feature dimensions; the preprocessing process includes: cleaning and standardization;

[0076] The feature engineering unit is used to extract the attributes of multi-dimensional features from the pre-processed data and construct a feature matrix to enable the on-demand diversion and use of multi-dimensional features in the machine learning prediction model; the multi-dimensional features include: multi-dimensional time features, spatial features and passenger behavior features.

[0077] The data sources of the data input unit are: 8760 hours of air-conditioning load data throughout the year from TRNSYS simulation; timestamp and OD (origin and destination) passenger flow data recorded by the subway AFC system; temperature, humidity and CO2 concentration information provided by the meteorological monitoring station; and statistical data on passenger travel purposes and transfer rates in urban traffic surveys.

[0078] The processing flow of the data pre-processing unit is specifically as follows:

[0079] The original data were smoothed using the sliding window method and aligned to a time granularity of 15 minutes. The STL method was then used for outlier detection. For outliers with residuals greater than three times the standard deviation, they were replaced with the mean of the adjacent time periods. Finally, all features were normalized by Z-Score to eliminate the dimensional effects between different feature dimensions.

[0080] The feature engineering unit is used to extract and construct multi-dimensional time features and attributes related to the time dimension from the processed data. Multi-dimensional time features include hour periods, week types, and holiday signs; spatial features include station hall heat maps and platform passenger flow density distribution; behavioral features include weighted processing of passenger detention duration and transfer paths. Different types of features are used as needed in subsequent modeling.

[0081] In summary, this embodiment also has the following technical effects:

[0082] 1) System energy efficiency is significantly improved. Compared to traditional air conditioning systems that use daily average load control, this invention improves the cooling / heating energy efficiency ratio (COP / EER) by an average of approximately 15% to 25% through dynamic prediction and adjustment. This avoids unnecessary over-operation and reduces efficiency losses caused by frequent system starts and stops.

[0083] 2) Reduced energy consumption. Air conditioning operates for approximately six months a year. After optimization and control based on TRNSYS simulation results and measured passenger flow data, the total energy consumption of the air conditioning system can be reduced by approximately 18% to 26% (the data fluctuates slightly based on the characteristics of different sites). This is achieved by improving peak-period matching and by suppressing redundant operation during off-peak periods, contributing approximately 10%, and approximately 14%.

[0084] 3) Significant economic benefits. Assuming a station's annual air conditioning energy expenditure is 800,000 yuan (primarily electricity costs), simulation and energy-saving analysis estimate an 18% energy saving, saving approximately 144,000 to 208,000 yuan in annual operating costs. If this system were implemented at all 100 stations in the city, the potential annual savings in air conditioning energy expenditure would be approximately 14 million to 20 million yuan.

[0085] 4) Equipment maintenance and life benefits: By reducing frequent starts and stops through load forecasting algorithms, the operating life of key equipment such as compressors and fans is expected to be extended by 12% to 20%, the maintenance cycle is lengthened, and maintenance costs are reduced by about 10% to 15%.

[0086] 5) Passenger comfort scores have improved. Combined with the results of a domestic urban rail transit satisfaction survey, after adjusting the temperature to moderately improve the thermal comfort level, the waiting area passenger satisfaction score for air conditioning has increased by an average of 0.6 to 0.9 points (out of a full score of 5 points), and the complaint rate has dropped significantly, indirectly improving the quality of operational services.

[0087] Example 2

[0088] like Figure 5 As shown, the present invention also provides a method for regulating the air conditioning load of a subway station in Example 1, comprising the following steps:

[0089] S1. Based on the TRNSYS simulation platform, simulate the temporal changes of cooling load and heating load of different types of stations under different meteorological conditions and operating conditions to generate system training data sets;

[0090] S2. Utilizing historical subway passenger flow data and urban public transportation survey data, establish a time-based passenger distribution model that reflects the travel habits of urban residents. This time-based passenger flow distribution model describes the dynamic changes in passenger flow at an hourly granularity and maps this to the population density distribution in different areas of the station, outputting the passenger flow density in each area.

[0091] S3. Extract multi-dimensional features of passenger flow density, external weather, and historical load data as input features for the random forest regression algorithm and establish a machine learning prediction model. Utilize the system training dataset and the machine learning prediction model to learn the mapping relationship between the input features and the target cooling / heating load, and output preset load values ​​at different time points.

[0092] S4. Based on the preset load values ​​at different time points, the output capacity of the air-conditioning system is dynamically adjusted through the building automation system to achieve closed-loop control. The dynamic adjustment of the output capacity of the air-conditioning system includes: start and stop control of the cold / heat source equipment, air volume control, and supply air temperature adjustment.

[0093] In this embodiment, the control strategy for peak load period regulation (forecast load value ≥ 85% of design load) is:

[0094] 1. Cold source control strategy

[0095] Start all heat pump units, set the cold water supply temperature to 6°C (7°C in normal mode), and adjust it using the formula:

[0096] T chw =7-0.2×(P pred / P max )

[0097] Among them, P pred is the predicted load value, P max Design maximum load

[0098] When P pred / P max When ≥0.85, the cold storage tank is activated to assist in cooling, and the cold capacity stored during off-peak electricity price periods is used first.

[0099] 2. Dynamic adjustment of air volume

[0100] According to the real-time temperature deviation ΔT in the station hall / platform area, the PID control algorithm is used to adjust the air valve opening:

[0101] V=K p ×ΔT+K i∫ΔTdt+K d ×(dΔT / dt)

[0102] The proportional coefficient K p =0.8, integration time T i =120s, differential time T d =30s

[0103] At the same time, the fresh air ratio is increased to 35% (normally 25%) to cope with high CO2 concentrations.

[0104] 3. Equipment collaboration

[0105] The frequency of the cold water pump is increased to 45Hz (base frequency 35Hz), and the deep buried pipes and bottom plate buried pipes are turned on full speed mode to ensure that the condensing temperature is ≤32℃.

[0106] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A subway station air conditioning load adjustment system, characterized in that: include: The load simulation module is used to simulate the temporal changes of cooling and heating loads of different types of stations under different meteorological conditions and operating conditions, and generate system training data sets; The passenger travel behavior modeling module uses historical subway passenger flow data and urban public transportation survey data to establish a time-based passenger distribution model that can reflect the travel habits of urban residents. This time-based passenger flow distribution model describes the dynamic changes in passenger flow at an hourly granularity and maps it to the population density distribution in different areas of the station, outputting the passenger flow density in different areas. The machine learning prediction module, with a built-in random forest regression algorithm, is used to extract multi-dimensional features of passenger flow density, external meteorological data, and historical load data as input features to establish a machine learning prediction model. The machine learning prediction model is trained using the system training data set to learn the mapping relationship between input features and target cooling / heating loads, and output preset load values ​​at different time points. Intelligent load regulation control module dynamically adjusts the output capacity of the air conditioning system through the building automation system according to the preset load values ​​at different time nodes to achieve closed-loop control; The output capacity of the dynamic adjustment air-conditioning system includes: start and stop control of cold / heat source equipment, air volume control and supply air temperature adjustment.

2. The subway station air conditioning load adjustment system according to claim 1, characterized in that: The machine learning prediction module also includes: a data input unit, a data preprocessing unit, and a feature engineering unit; The data input unit is used to input passenger flow density, external weather and historical load data to obtain an original data set; The data preprocessing unit is used to preprocess the original data set to eliminate the dimensional influence between different feature dimensions; The pretreatment process includes: cleaning and standardization; The feature engineering unit is used to extract the attributes of multi-dimensional features from the preprocessed data and construct a feature matrix to enable the on-demand diversion and use of multi-dimensional features in the machine learning prediction model; the multi-dimensional features include: multi-dimensional time features, spatial features and passenger behavior features.

3. The subway station air conditioning load adjustment system according to claim 2, characterized in that: The data sources of the system training dataset are specifically: 8760 hours of air conditioning load data from TRNSYS simulation throughout the year; timestamp and origin and destination passenger flow data recorded by the subway AFC system; temperature, humidity and CO2 concentration information provided by meteorological monitoring stations; and statistical data on passenger travel purposes and transfer rates from urban traffic surveys.

4. The subway station air conditioning load adjustment system according to claim 2, characterized in that: The processing flow of the data pre-processing unit is specifically as follows: The original data were smoothed using the sliding window method and aligned to a time granularity of 15 minutes. The STL method was then used for outlier detection. For outliers with residuals greater than three times the standard deviation, they were replaced with the mean of the adjacent time periods. Finally, all features were normalized by Z-Score to eliminate the dimensional effects between different feature dimensions.

5. The subway station air conditioning load adjustment system according to claim 2, characterized in that: The multi-dimensional time features include hour periods, week types, and holiday signs; the spatial features cover the station hall heat map and platform passenger flow density distribution; the passenger behavior features include: weighted processing of passenger detention time and transfer paths. Different types of features are used as needed in subsequent modeling.

6. The subway station air conditioning load adjustment system according to claim 1, characterized in that: The training process of the machine learning prediction model is as follows: LightGBM is initialized as the basic model, and the number of leaf nodes is set to 127; then, the Bayesian optimization algorithm is used to search for hyperparameters, and the search space is mainly concentrated in the two dimensions of min_data_in_leaf∈[50,200] and lambda_l1∈[0,5]; the machine learning prediction model training adopts a multi-scale input method, which simultaneously accesses short-term features and long-term features.

7. The subway station air conditioning load adjustment system according to claim 1, characterized in that: The verification process of the machine learning prediction model is as follows: in the offline stage, the performance is evaluated by 5-fold cross validation to ensure the determination coefficient R 2 Greater than 0.9; during the deployment phase, an incremental learning framework is built, and the full model is set to be updated every 72 hours. At the same time, the model is fine-tuned with real-time data every 15 minutes to ensure the timeliness and accuracy of the prediction results.

8. The subway station air conditioning load adjustment system according to claim 1, characterized in that: The intelligent load regulation control module controls the output capacity of the air conditioning system as follows: the air conditioning load results predicted by the machine learning prediction model are connected to the interface of the building automation system; specifically, the air volume setting is adjusted according to the predicted value P_pred: Q = α × P_pred + β; and the target set temperature T_set = f(COP, P_pred) of the air conditioning system is determined based on the energy efficiency curve and the load prediction value. Where Q represents the ventilation volume required in the air conditioning system; α represents the linear proportional factor between the air volume and the predicted load; p represents the load value of the building or equipment; and β represents the reference offset in the air volume setting.

9. A method for regulating the air conditioning load of a subway station, applied to the subway station air conditioning load regulating system according to any one of claims 1 to 8, characterized in that: The following steps are involved: S1. Based on the TRNSYS simulation platform, simulate the temporal changes of cooling load and heating load of different types of stations under different meteorological conditions and operating conditions to generate system training data sets; S2. Utilizing historical subway passenger flow data and urban public transportation survey data, establish a time-based passenger distribution model that reflects the travel habits of urban residents. This time-based passenger flow distribution model describes the dynamic changes in passenger flow at an hourly granularity and maps this to the population density distribution in different areas of the station, outputting the passenger flow density in each area. S3. Extract multi-dimensional features of passenger flow density, external weather, and historical load data as input features for the random forest regression algorithm and establish a machine learning prediction model. Utilize the system training dataset and the machine learning prediction model to learn the mapping relationship between the input features and the target cooling / heating load, and output preset load values ​​at different time points. S4. Dynamically adjust the output capacity of the air conditioning system through the building automation system according to the preset load values ​​at different time nodes to achieve closed-loop control; The output capacity of the dynamic adjustment air-conditioning system includes: start and stop control of cold / heat source equipment, air volume control and supply air temperature adjustment.

Images