Multi-source perception fusion based dynamic optimization control method for air conditioning system of railway station

Abstract

The application discloses a kind of multi-source perception fusion's railway station air conditioning system dynamic optimization control method.The method collects static building parameters, real-time equipment operation and passenger flow perception data, constructs unified space-time index multi-source fusion basic data set;Establish partition heat and humidity balance model, use state estimation algorithm to online inversion equivalent sensible heat and latent heat load of physical partition, carry out short-time rolling prediction;Based on real-time passenger flow thermal map and load characteristics, spatial dynamic clustering is carried out, and virtual partition across physical boundary is generated, and optimization model is constructed to dynamically allocate VAV terminal air volume;Combined with train operation plan, the linkage passenger flow migration and load surge between platform and waiting room are predicted, and the joint control trajectory containing precooling, preheating and ventilation strategy is generated.The application solves the problem of fixed partition and dynamic passenger flow mismatch and control lag through virtual partition reconstruction and load feedforward control, and improves the thermal comfort and operation energy efficiency of passenger station.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of HVAC intelligent control, and particularly relates to a multi-source perception fusion railway station air conditioning system dynamic optimization control method. BACKGROUND

[0002] With the rapid development of high-speed railways, large railway stations as urban transportation hubs have large building sizes and extremely high passenger flow throughput, resulting in a large proportion of air conditioning system energy consumption in operating costs. Such large spaces have the characteristics of strong thermal inertia and uneven temporal and spatial distribution of load. Under the premise of ensuring passenger thermal comfort, energy-saving control of the air conditioning system helps to build a green and low-carbon hub.

[0003] Currently, the air conditioning control of railway stations relies on building automation systems (BAS), and usually adopts start-stop control based on a time table or single-loop PID feedback control based on return air temperature. Air conditioning zones are fixedly divided according to the civil partition or fire partition in the design stage. In the running process, AHU units and VAV terminal devices mainly act according to the preset logic or fixed indoor temperature set value to maintain the average temperature and humidity level in each physical partition.

[0004] The existing technology mainly has the problems of mismatch between static physical boundaries and dynamic passenger flow distribution, and lagging feedback control and lack of foresight. Therefore, further research and innovation are needed to solve the above problems existing in the prior art. SUMMARY

[0005] The present application provides a multi-source perception fusion railway station air conditioning system dynamic optimization control method to solve the above problems of the prior art.

[0006] Technical scheme: According to one aspect of the present application, a multi-source perception fusion railway station air conditioning system dynamic optimization control method comprises:

[0007] Calling a multi-source fusion basic data set, including passenger flow spatial distribution characteristics, train operation plan, device spatial layout information, static building parameters, real-time environment and device operation data, operation plan and historical data, and passenger flow micro-behavior characteristics;

[0008] Based on the multi-source fusion basic data set, a simplified thermal and humidity balance model of the partition is constructed, the sensible heat and latent heat exchange amounts of each physical partition are inversely calculated online, and a partition equivalent thermal and humidity load state set is generated;

[0009] Extracting passenger flow spatial distribution characteristics, combining the partition equivalent thermal and humidity load state set to perform spatial dynamic clustering, and identifying and generating a virtual partition set;

[0010] According to the virtual partition set and the train operation plan, the linkage load demand of the platform and the waiting hall is predicted, the VAV terminal is optimized and solved, and the air conditioning system control trajectory set is generated.

[0011] Beneficial effects: the application solves the problems of fixed partition and dynamic passenger flow mismatch and control lag by virtual partition reconstruction and load feedforward control, and improves the thermal comfort and operation energy efficiency of the passenger station. Related technical effects will be described in detail below in combination with specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0012] Figure 1 A flowchart of a multi-source perception fusion dynamic optimization control method of a railway passenger station air conditioning system is provided for the embodiments of the application.

[0013] Figure 2 A flowchart of identifying and generating a virtual partition set is provided for the embodiments of the application.

[0014] Figure 3 A flowchart of generating an air conditioning system control trajectory set is provided for the embodiments of the application.

[0015] Figure 4 A flowchart of predicting the linkage load demand of the platform and the waiting hall is provided for the embodiments of the application.

[0016] Figure 5 Another flowchart of generating an air conditioning system control trajectory set is provided for the embodiments of the application. DETAILED DESCRIPTION

[0017] In order to enable persons skilled in the art to better understand the application scheme, the technical solutions in the embodiments of the application will be described clearly and completely below in combination with the drawings in the embodiments of the application. Obviously, the described embodiments are only a part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by persons skilled in the art without creative labor should belong to the protection scope of the application.

[0018] It should be noted that the terms first, second, etc. in the specification of the application and in the above-described drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms include and have and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily limit to the clearly listed steps or units, but can include other steps or units that are not clearly listed or inherent to the process, method, product or device.

[0019] To solve the above problems, the applicant has conducted in-depth retrieval and analysis, and found that:

[0020] Specifically, the traditional control relies on fixed civil partitions and cannot adapt to the tidal passenger flow that flows quickly with train ticket checking in the waiting hall, resulting in a mismatch between supply and demand of cold energy in high-density aggregation areas (such as ticket checking gates) and idle areas (such as passages).

[0021] Next, the feedback logic relying on temperature deviation cannot distinguish between sensible heat and latent heat demand, and ignores the influence of building envelope radiation and infiltration wind, making it difficult to handle the thermal hysteresis of large spaces; on this basis, the control systems of the waiting hall and the platform are independent of each other, and the train timetable is not included in the control closed loop, resulting in the system being unable to cope with the explosive passenger flow surge and thermal shock induced by the instant arrival and departure of trains, and may be adjusted passively after the environment has deteriorated.

[0022] To solve these problems, in combination with Figures 1 to 5 The present application is specifically illustrated by the following embodiments.

[0023] Embodiment 1, a multi-source perception fusion railway station air conditioning system dynamic optimization control method is provided. An exemplary scheme of fusing multi-source heterogeneous data, online inverting partition load and performing dynamic optimization control is described. This method can run on the building automation system BAS of the station, the edge computing server or the private cloud platform, and realize data interaction and instruction issuing through the communication interface with various subsystems. Specifically, the method can include the following steps:

[0024] Step S101, a multi-source fusion basic data set is constructed, mainly including passenger flow spatial distribution characteristics, train operation plan, device spatial layout information, static building parameters, real-time environment and device operation data, operation plan and historical data, and passenger flow micro-behavior characteristics.

[0025] In some embodiments, static building parameters, real-time device operation data, real-time passenger flow perception data and train operation plan are collected, spatio-temporal alignment and multi-source data fusion are performed, and a multi-source fusion basic data set is generated.

[0026] Since various systems in the station are usually constructed independently, for example, building information model BIM data is static, while building automation system BAS, video passenger flow analysis system and train dispatching system operation data are dynamic, and the time stamps and spatial identifiers of each are incompatible. Therefore, a data base with a unified space-time reference needs to be constructed.

[0027] Specifically, the station static building parameters are mainly extracted from design drawings, BIM models and equipment account books, including the heat transfer coefficient of the envelope structure, the window-wall ratio, the three-dimensional coordinates of the supply and return air outlets, and the air-water pipe network topology structure of the air conditioning system. Real-time equipment operation data are collected through communication protocols such as BACnet or Modbus of the building automation system, including the supply air temperature, supply air volume, valve opening, fan frequency, return air temperature and humidity, and carbon dioxide concentration of the air conditioning unit. Real-time passenger flow perception data are derived from the passenger flow heat map of the video analysis camera, the in-out station count of the automatic ticket gate, and the positioning data of the wireless probe, which are used to depict the distribution density and movement trajectory of passengers in the station, forming the micro-behavior characteristics and spatial distribution characteristics of passenger flow. The train operation plan is read from the railway dispatching interface, including the train number, planned arrival and departure time, length of the train, and ticket sales rate information fed back by the ticketing system.

[0028] Further, the system performs a space-time alignment operation. In the time dimension, the system sets a uniform global time step, for example, 1 minute. For data with a sampling frequency higher than the step, such as 5-second sensor readings, the system uses a sliding window average method for downsampling; for data with a sampling frequency lower than the step, such as train timetables, the system uses numerical retention or linear interpolation method for filling. In the spatial dimension, the system establishes a unified three-dimensional coordinate system covering the entire station, and maps all sensors, equipment components and passenger flow observation points to specific spatial grids in the coordinate system.

[0029] After the above processing, the system outputs a structured multi-source fusion basic data set, which is indexed by a unified timestamp and spatial grid ID, and integrates building parameters, environmental status, equipment operation status and passenger flow characteristics, providing a standardized input interface for subsequent load inversion and control decision-making. The problem of standardization and unification of multi-source heterogeneous data is solved.

[0030] Step S102, based on the multi-source fusion basic data set, a simplified thermal-hygroscopic balance model is constructed, and the sensible heat and latent heat exchange amounts of each physical partition are inversed online to generate a partition equivalent thermal-hygroscopic load status set.

[0031] The constructed simplified thermal-hygroscopic balance model is a dynamic mathematical model based on the law of conservation of energy and mass. The model regards each physical air conditioning partition as a lumped parameter node, and considers the heat transfer of the envelope structure, air permeation, internal heat source heat dissipation, and the cold or heat brought by the supply air.

[0032] In the process of online inversion, the partitioned instantaneous sensible and latent heat loads are regarded as unknown disturbance terms or state variables in the model which cannot be directly measured. The real-time environmental monitoring data in the multi-source fusion basic data set, such as indoor temperature and humidity rate, air supply parameters, are taken as observation values and substituted into the model. The preset estimation algorithm, such as extended Kalman filter (EKF), is used to calculate the net heat and net moisture required to maintain the current temperature and humidity trend in real time, that is, to invert the current equivalent sensible and latent heat loads.

[0033] Further, this step combines the physical model with data-driven, which can solve the problem that the current partition actual load cannot be accurately obtained in traditional control. It can also eliminate measurement noise and accurately capture the load fluctuations caused by passenger flow aggregation or sudden changes in solar radiation. The generated partition equivalent heat and moisture load state set not only contains the inversion value at the current time, but also contains the prediction value of the load evolution in the future short time domain based on the current state, which provides a quantitative load benchmark for subsequent control.

[0034] Step S103, extract the passenger flow spatial distribution characteristics in the multi-source fusion basic data set, and perform spatial dynamic clustering based on the partition equivalent heat and moisture load state set to identify and generate a virtual partition set.

[0035] Correspondingly, in a large railway station waiting hall, the distribution of passengers often shows strong unevenness and fluidity, for example, a high-density aggregation area is formed near the ticket gate, while the surrounding area is relatively empty.

[0036] Among them, extracting passenger flow spatial distribution characteristics refers to obtaining real-time passenger flow density, average stay time, and moving speed and other indicators in each spatial grid from the data set. Combined with the partition equivalent heat and moisture load state set, the system constructs a comprehensive feature field containing load intensity and passenger flow attributes.

[0037] On this basis, spatial dynamic clustering refers to applying a clustering algorithm to merge grid cells that are physically adjacent and have similar passenger flow characteristics and load demand to form new control units, i.e., virtual partitions. For example, several high-load grids in front of the ticket gate corresponding to a train are merged into a virtual partition, while low-load grids far from the ticket gate are merged into another virtual partition. The boundaries of the virtual partition are dynamically adjusted with the changes in passenger flow lines, and are no longer limited by the original physical partition boundaries.

[0038] Further, the virtual partition set defines the control objects that need to be focused on at the current time, so that the air conditioning system can concentrate on delivering cold energy to high-load areas, while implementing energy-saving operation in low-load areas. That is, this step can break the mode of dividing air conditioning partitions based on the civil partition wall or fixed air outlet, and realize on-demand energy supply.

[0039] Step S104, according to the virtual partition set and the train operation plan in the multi-source fusion basic data set, the linkage load demand of the platform and the waiting hall is predicted, the VAV terminal and the unit operation parameter are optimized and solved, and a set of air conditioning system control trajectories are generated.

[0040] Further, the particularity of the railway station is that there is a large passenger flow interaction between the waiting hall and the platform. The arrival of the train will cause the passenger flow to flow from the waiting hall to the platform, or from the platform to the outbound channel. The short-term passenger flow surge will cause a great load impact on the air conditioning system. According to the train operation plan, the system can predict the arrival and departure time and the passenger capacity of the train in advance, and predict the migration process of the passenger flow between the waiting hall and the platform, that is, the linkage load demand.

[0041] In the optimization solving stage, the system takes the virtual partition as the target object and establishes an optimization model. The model comprehensively considers the comfort index of the waiting hall, the ventilation requirement of the platform and the total energy consumption of the system. The system calculates the air volume distribution ratio required by each variable air volume VAV terminal and the best supply air temperature and fan frequency of the air conditioning unit AHU in order to meet the temperature and humidity demand of each virtual partition in the future period of time.

[0042] For the platform area, the system calculates the start and stop time and the running gear position of the fresh air unit and the exhaust air unit. The generated set of air conditioning system control trajectories is a series of control instruction sequences that change with time, which is directly issued to the bottom execution mechanism, realizing intelligent control from passive response to active prediction and from local adjustment to global coordination. Or, it realizes the linkage of the system across space and the global optimization control.

[0043] In the embodiment, it also includes optimizing and determining the unit operation parameters of the current period according to the output results of the prediction model, including the chilled water supply temperature set value, the cooling water supply temperature set value and / or the unit load rate.

[0044] In another embodiment of the present application, rolling optimization can also be further performed, which includes: predicting the cooling load of the future period by using the prediction model; based on the predicted cooling load, optimizing and determining the unit operation parameters of the current control period with the optimal unit energy efficiency as the target.

[0045] Embodiment 2 describes a specific implementation method of the time-space alignment and fusion basis of multi-source data. In particular, how to construct the mapping relationship between the equipment and the space, and how to specifically perform the cleaning and fusion of multi-source data are used to ensure the convergence of the subsequent inversion algorithm and the feasibility of the control strategy. In the embodiment, the implementation manner can be:

[0046] Step S201, collect the static building parameters of the station, and based on the spatial geometric information and the electromechanical system topology in the static building parameters of the station, construct a device-space grid-air conditioning partition multi-level mapping relationship containing the association of physical location coordinates and control logic.

[0047] In some embodiments, the spatial geometric information and the electromechanical system topology in the static building parameters of the station are analyzed to construct a device-space grid-air conditioning partition multi-level mapping relationship containing the association of physical location coordinates and control logic. A static physical index base is established.

[0048] Correspondingly, the system discretizes the planar area of the waiting hall and the platform into regular space grid units, and the size of the grid can be set according to the control accuracy requirement, for example, set to 2m x 2m. For each space grid, the system records the three-dimensional coordinates of its geometric center and the attribute of the functional area to which it belongs, such as belonging to the waiting area, the corridor or the commercial area. Then, the system analyzes the electromechanical system topology and identifies the installation position coordinates of each air conditioning terminal device such as the VAV box, the air supply port and the return air port.

[0049] On this basis, a multi-level mapping relationship is constructed, especially the influence of the device on the grid is quantified. Due to the diffusivity of air supply jet, one air supply port often affects multiple grids in the surrounding. In this embodiment, a distance attenuation model is used to calculate the influence weight. Correspondingly, the influence weight w _ij of the i-th device on the j-th grid is defined as:

[0050] w _ij = d _ij -α x cos(θ _ij );

[0051] where d _ij represents the Euclidean distance from the center of the air outlet of the device i to the center of the grid j; θ _ij represents the angle of the center of the grid j relative to the air supply axis of the device i; and a is an empirical attenuation coefficient, usually taking a value between 1 and 3.

[0052] Based on this formula, the system generates a weight matrix of devices to grids. Further, according to the physical partition wall and the design drawing, the attribution relationship of the grid to the physical air conditioning partition is determined, and a mapping matrix of the grid to the partition is generated. The mapping relationship table is persistently stored for subsequent data processing calls.

[0053] Step S202, according to the device-space grid-air conditioning partition multi-level mapping relationship, map the real-time device operation data and the real-time passenger flow perception data to a unified space grid unit, resample and align according to a preset time step, and generate a structured and unified spatio-temporal index multi-source fusion basic data set.

[0054] For real-time device operation data, such as real-time air volume feedback of VAV terminal, the system uses device-to-grid weight matrix to allocate the operation data of a single device to multiple space grids covered by the device. For example, the total air volume of a certain VAV terminal is V _total , the air volume V _j allocated to grid j is V _total × w _ij .

[0055] For real-time passenger flow perception data, such as personnel coordinate points identified by camera, the system directly matches the coordinate points to the corresponding space grid and counts the number of personnel in each grid to calculate the passenger flow density.

[0056] In terms of time alignment, since the sampling clocks of various subsystems are not synchronized, the system aligns them based on a preset time step t (e.g., 1 minute). For high-frequency data, such as temperature sensor values reported every 5 seconds, the system calculates the arithmetic mean value of the data in the t period as the value of the time step; for low-frequency or non-periodic data, such as train arrival signals, the system keeps the state value until the current time step. In addition, the system also performs data quality checks and interpolates missing values. For example, if the temperature data of a grid is missing, the system uses the average temperature of the spatially adjacent grids to perform spatial interpolation to fill in the missing values. The multi-source fusion basic data set is a structured data body, each record of which contains a timestamp t, a grid identifier ID, an environmental parameter set (temperature, humidity, CO2), a device state set (air volume, air temperature), and a passenger flow feature set (density, speed), realizing the deep fusion of multi-source information.

[0057] That is, this step belongs to the execution of real-time stream processing of dynamic data.

[0058] According to one aspect of the present application, the structure of the multi-source fusion basic data set D(t, z) can also be represented as:

[0059] D(t, z) = {T _air , H _air , N _occ , S _vav , Q _dev};

[0060] wherein t is a unified time step identifier, z is a unique identifier ID of a space grid, T _air is the air temperature in the grid, H _air is the relative humidity in the grid, N _occ is the passenger flow density in the grid, S _vav is the operation state of the associated VAV terminal, and Q _dev is the device operation parameter.

[0061] The embodiment 3 provides an optional implementation of online inversion and rolling prediction of partition equivalent heat and moisture loads. How to convert the load disturbance into quantifiable values by physical modeling and Kalman filtering algorithm to predict future trends is described.

[0062] Step S301, real-time passenger flow and behavior data and door and window opening states are extracted from the multi-source fusion basic data set, human sensible heat, human latent heat and penetration wind load of each physical partition are calculated, and a partition real-time internal disturbance input set is generated. The specific implementation is as follows:

[0063] Optionally, according to the passenger flow micro-behavior characteristics in the multi-source fusion basic data set, the on-site personnel in each physical partition are divided into static waiting, slow walking and fast passing categories, and the corresponding differentiated human metabolic rates are matched.

[0064] Among them, the sensible heat and latent heat disturbance brought by human is the main component of the passenger station air conditioning load. The traditional load calculation often takes the fixed human heat generation, and ignores the influence of passenger behavior state on heat dissipation. The system extracts the micro-behavior characteristics of passenger flow from the multi-source fusion basic data set, and the main basis is the average moving speed and stay time of passengers in the grid. The system sets a speed threshold to classify personnel: when the moving speed is less than 0.5 meters per second, it is determined as the static waiting category; when the moving speed is between 0.5 meters per second and 1.5 meters per second, it is determined as the slow walking category; when the moving speed is greater than 1.5 meters per second, it is determined as the fast passing category.

[0065] For each type of personnel, the system matches the corresponding differentiated human metabolic rate Met value. For example, the metabolic rate corresponding to the static waiting state is usually set to 1.0; the metabolic rate corresponding to the slow walking state is usually set to 1.2; and the metabolic rate corresponding to the fast passing state is usually set to 1.6. The numerical value refers to the ASHRAE standard or related measured data. By using the classification matching mechanism, the heat dissipation difference of personnel in different areas (such as the waiting area and the passage area) can be more truly reflected.

[0066] Optionally, the heat dissipation and moisture dissipation of each category of personnel are calculated based on the differentiated human metabolic rate, and the air penetration ventilation caused by the door opening frequency is combined to synthesize the total sensible heat disturbance and the total latent heat disturbance of each physical partition, and a partition real-time internal disturbance input set is generated.

[0067] Correspondingly, after obtaining the metabolic rate, the system further calculates the total human load of each partition. Taking the sensible heat load as an example, the calculation formula of the human sensible heat load Q _sens_occ of the partition z is:

[0068] Q _sens_occ =Σ(N _m ×Met _m ×C_sens );

[0069] where m is the personnel behavior category (including sitting and waiting, slow walking, and fast passing) ; N _m is the number of personnel in the partition belonging to category m; Met _m is the differential human metabolic rate corresponding to category m; C _sens is the sensible heat conversion coefficient.

[0070] In a possible implementation, the partition human sensible heat load calculation formula can also be:

[0071] Q _sens_occ =∑(N _m ×Met _m ×(1-η)×C _sens );

[0072] where η is the radiation heat dissipation proportion coefficient.

[0073] Similarly, the human latent heat load is also calculated through the corresponding number of persons, metabolic rate, and latent heat conversion coefficient. In addition, the system also calculates the infiltration load caused by the opening of doors and windows. The system monitors the on-off state signal of the access control system, and estimates the infiltration air volume V _inf according to the indoor and outdoor pressure difference. As an example, the calculation of the partition infiltration sensible heat load can be described as the following formula:

[0074] Q _sens_inf =ρ×Cp×V _inf ×(T _out -T _in );

[0075] where Q _sens_inf is the infiltration sensible heat load of a physical partition; ρ is the air density; Cp is the air specific heat capacity; V _inf is the infiltration air volume caused by the opening of doors and windows; T _out is the outdoor temperature; T _in is the indoor temperature. That is, the infiltration sensible heat load = air density × air specific heat capacity × infiltration air volume × indoor and outdoor temperature difference.

[0076] On this basis, the human load and the infiltration load are added, that is, the total sensible heat disturbance and the total latent heat disturbance of the partition are obtained. The above disturbance data are organized in time series to form a real-time internal disturbance input set of the partition, which is used as a known input item to participate in subsequent model inversion.

[0077] In step S302, the static building parameters in the multi-source fusion basic data set are used to establish a partition heat and moisture balance discrete state space equation describing the evolution of the temperature and humidity content of each physical partition, and the equivalent sensible heat load and the equivalent latent heat load are defined as unknown state variables to be inverted. It is used to construct the mathematical model required for inversion.

[0078] Accordingly, the system establishes a thermal-hygroscopic equilibrium differential equation of the zone z based on the principle of energy conservation, and discretizes the thermal-hygroscopic equilibrium differential equation into a state space form. A specific state equation can be expressed as:

[0079] x _k+1 =A×x _k +B×u _k +G×q _k ;

[0080] wherein x _k+1 is a state vector at a next time k+1; A is a self-evolution matrix of the system, describing a heat capacity and a heat transfer characteristic; x _k is a state vector at a current time k, containing an air temperature and a humidity ratio of the zone; B is a control input matrix; u _k is a control input vector at the current time k, containing a supply air volume and a supply air temperature; G is a disturbance coefficient matrix; and q _k is an unknown load vector at the current time k to be inverted, containing an equivalent sensible heat load and an equivalent latent heat load.

[0081] Further, the matrix A describes a self-evolution characteristic of the system, elements of which are related to an air heat capacity of the zone and a building envelope heat transfer coefficient; the matrix B describes an influence of air conditioning supply air on an indoor state; and the matrix G is a coefficient matrix of a load disturbance. q _k not only includes a known internal disturbance, but also includes solar radiation, equipment heat dissipation error, and a residual term of a model not built. _k is defined as a part of the state variable (usually extended to the state vector), so that it can be estimated through an observation equation. Exemplarily, the observation equation is specifically as follows:

[0082] y _k =C×x _k ;

[0083] wherein y _k is an observation vector at the current time k, i.e., temperature and humidity data measured by a sensor; and C is an observation matrix.

[0084] According to one aspect of the present application, the zone thermal-hygroscopic equilibrium equation can be:

[0085] Cz×(dT _z / dt)=Q _sa +Q _env +Q _int +Q _eq ;

[0086] wherein Cz is an effective heat capacity of the zone, T _z is an air temperature of the zone, t is time, Q _saThe cold and heat provided for air supply, Q _env The heat transfer amount of the envelope structure, Q _int The internal known disturbance amount, Q _eq The equivalent load to be inverted.

[0087] Step S303, the real-time internal disturbance input set of the partition and the real-time environment and equipment operation data in the multi-source fusion basic data set are substituted into the discrete state space equation of the partition heat and humidity balance, the joint state parameter estimation algorithm is used, the current equivalent sensible and latent heat loads and the updated model parameter set are solved at the same time, and the current equivalent sensible and latent heat loads are stored in the partition equivalent heat and humidity load state set.

[0088] Correspondingly, the system adopts the joint state parameter estimation algorithm, which can be selected as the extended Kalman filter algorithm EKF. The system constructs an extended state vector z _k , which is spliced by the original state x _k (temperature and humidity), the parameter θ _k (such as the comprehensive heat transfer coefficient which is difficult to accurately obtain) and the unknown load q _k . At each time step k, the algorithm performs two stages of prediction and update. The prior value of the state at the current time is predicted by using the estimated value at the last time; the innovation (residual) is calculated by using the sensor observation value y _k at the current time; and the optimal posterior estimation value at the current time is obtained by correcting the prior value according to the Kalman gain matrix. Based on this, the extended state vector can be constructed as:

[0089] z _k =[x _k , θ _k , q _k ] T ;

[0090] Wherein, θ _k is the model parameter vector to be corrected, T and represents the vector transpose.

[0091] In this way, the system not only solves the current equivalent sensible and latent heat loads in real time, but also corrects the model parameters synchronously, so that the model can adapt to the slow change of the building performance (such as the change of resistance caused by filter blockage or the thermal inertia of the envelope structure). The load value obtained by solving is stored in the partition equivalent heat and humidity load state set. The load value in the data set is the net load, that is, the heat and humidity that must be removed or supplemented in order to maintain the current environment state unchanged, which is closer to the real physical demand of the system than the theoretical calculation value.

[0092] Step S304, according to the train operation plan and historical data in the multi-source fusion basic data set, the passenger flow change trend and outdoor meteorological parameters in the future prediction time domain are extrapolated to generate a partition short-term disturbance input prediction sequence.

[0093] After obtaining the current state, the system needs to make predictions for the future. The system utilizes the train timetable in the operation plan, combined with the passenger flow curve of the same type of train in history, to predict the passenger flow density trend of each subzone in the next 30 to 60 minutes. Further, meteorological forecast data is accessed to predict the changes in outdoor temperature and humidity. The system converts the predicted passenger flow into a human load prediction value, and converts the predicted meteorological parameters into a building envelope heat transfer prediction value, to generate a subzone short-term disturbance input prediction sequence.

[0094] Step S305, call the updated model parameter set, take the current time equivalent sensible and latent heat load as the initial state, combine the subzone short-term disturbance input prediction sequence, and perform forward rolling simulation on the subzone thermal and moisture balance discrete state space equation under the constraint of the preset comfort control target; deduce the required sensible and latent heat exchange to maintain the comfort control target at each future time step, generate a subzone short-term equivalent heat and moisture load prediction sequence, and update it to the subzone equivalent heat and moisture load state set. Used to complete the rolling prediction of the load.

[0095] Correspondingly, the system takes the current state and the corrected parameters as the starting point, uses the future disturbance sequence as input, and evolves the model forward on a virtual time axis. During the simulation process, the system sets a comfort control target, such as requiring the indoor temperature at the future time to be maintained near the set value. The system uses the model equation to solve backward, i.e., in order to make the temperature at the next time equal to the target value, how much cooling or heating is required at the current time.

[0096] This calculation process is calculated step by step in the prediction time domain, to obtain a series of future load demand values. The values constitute a subzone short-term equivalent heat and moisture load prediction sequence. This sequence not only reflects the influence of external disturbances, but also contains the influence of system thermal inertia. For example, before the passenger flow peak arrives, the predicted load will rise in advance, prompting the system to need to precool or increase the supply air. The sequence is updated to the subzone equivalent heat and moisture load state set as the input boundary for subsequent virtual subzone optimization and distribution.

[0097] Embodiment 4, describes an exemplary scheme of virtual subzone and VAV dynamic distribution based on passenger flow streamline. In particular, how to use a dynamic clustering algorithm to reorganize the physical grid into virtual subzones, based on which to build an optimization model for air volume distribution, and realize control that follows passenger flow. Accordingly, the present scheme includes:

[0098] Step S401, extract the real-time passenger flow density thermal map of the waiting room from the multi-source fused basic data set, map it to the preset spatial grid unit, combine the subzone equivalent heat and moisture load state set, and construct a grid-level comprehensive feature vector containing passenger flow density, load intensity, and functional attributes.

[0099] Based on this, the system reads a real-time passenger flow density heatmap of the waiting hall, which is typically generated by a video analytics system and mapped onto a unified spatial grid. For clustering, the system needs to construct a multi-dimensional grid-level comprehensive feature vector V for each spatial grid i. _i The feature vector V _i Specifically, it can include three components. The first is the normalized passenger flow density, which represents the number of people per unit area and reflects the current level of population density. The second is the load intensity, which is derived from the zonal equivalent heat and humidity load state set. This value is obtained by allocating the zonal load to the grid level according to the area weight, reflecting the heat and humidity demand at that location. The third is the functional area code, for example, coding the waiting area as 1, the passageway as 0.5, and the commercial area as 0, to distinguish the service priority of different areas.

[0100] Optionally, the grid-level synthesized feature vector can be defined as:

[0101] V _i =[ρ _i ΔQ _i Type _i ];

[0102] Where, ρ _i The normalized passenger flow density, ΔQ _i For the normalized load strength, Type _i Encode the functional areas of the grid.

[0103] Furthermore, when constructing the feature vector, the system normalizes each component to eliminate dimensional differences. For example, it divides passenger flow density by the historical maximum density and load intensity by the equipment's rated cooling capacity density. In addition, the system can introduce time-dimensional features; for instance, it can include the predicted passenger flow density increment for the next 15 minutes as a dimension of the feature vector, giving the subsequently generated virtual partitions a certain degree of foresight. It should be understood that this step is the data preparation stage for virtual partition identification.

[0104] Step S402: Based on the grid-level integrated feature vector, a clustering algorithm with spatial connectivity constraints is used to aggregate spatially adjacent and feature-similar grid cells to form an initial region division.

[0105] Unlike K-Means clustering in traditional data mining, the clustering in this embodiment must guarantee physical spatial connectivity. That is, grids within the same virtual partition must be spatially adjacent, and enclaves cannot exist. In other words, isolated regions cannot exist where some grids within a partition are physically isolated from other grids and not adjacent to each other. The system employs a region growing algorithm with spatial connectivity constraints or a constrained spectrum clustering algorithm.

[0106] Taking the region growing algorithm as an example, the system selects several grid points with the maximum modulus of feature vectors, i.e., the highest passenger flow density or the highest load, as seed points. Next, the system iteratively checks the neighborhood grid points around the seed points, and calculates the Euclidean distance or the cosine similarity between the feature vectors of the neighborhood grid points and the seed points.

[0107] If the similarity between the neighborhood grid point and the seed point is higher than a preset merging threshold, and the grid point has not been classified, it is merged into the cluster to which the seed point belongs, and the average feature vector of the cluster is updated. The growing process continues until all grid points are assigned. In order to avoid generating too fine partitions, the system can set a minimum cluster area constraint, for example, requiring each virtual partition to contain at least 10 grid points. Through this process, grid points that are physically adjacent and have similar passenger flow load characteristics are aggregated together to form an initial regional division. For example, around a ticket gate where passengers are checking tickets, the high-density queuing area in front of the ticket gate will automatically be aggregated into a virtual partition, while the surrounding idle seat area will be aggregated into another virtual partition.

[0108] Step S403, morphological smoothing processing is performed on the initial regional division to eliminate isolated grid points and correct the boundary shape, generating a set of virtual partitions that span the physical air conditioning partition boundaries. This step specifically includes:

[0109] Optionally, a time smoothing penalty term is introduced to calculate the ownership change cost of each grid unit in the initial regional division relative to the virtual partition result of the previous time, and to suppress the regional jump caused by short-term passenger flow fluctuations.

[0110] Specifically, if the boundary of a virtual partition fluctuates dramatically over time, it will cause the air conditioning terminal equipment to adjust frequently, causing system oscillation. Therefore, the system introduces a time smoothing mechanism during clustering or post-processing. The system defines an ownership change cost function. For each grid, if its belonging partition label at the current time is different from the label at the previous time, a penalty value is generated.

[0111] Further, the system corrects the initial division result by optimizing a total objective function that includes a similarity error term and an ownership change cost term. Only when the grid feature changes significantly, so that the similarity benefit brought by re-division exceeds the ownership change cost, the system allows the grid to change its belonging virtual partition. This mechanism filters the noise caused by short-term movement of personnel, and helps to ensure the continuity and stability of the virtual partition.

[0112] Optionally, spatial morphological opening and closing operations are performed on the initial regional division to merge small isolated regions with an area below a preset threshold to adjacent dominant regions, smooth the jagged structure of the regional boundary, and generate a set of virtual partitions that are stable in time domain and continuous in space.

[0113] Where, the initial division may produce some irregular edges or tiny holes contained in the large area. The system uses morphological operations in image processing to process the division matrix. Accordingly, the closed operation, i.e. first inflation and then corrosion, is performed to fill the holes inside the division and connect the adjacent broken areas; the open operation, i.e. first corrosion and then inflation, is performed to eliminate small protrusions and isolated noise grids.

[0114] On this basis, the system can also set an area threshold, for example, 20 square meters. For the tiny division whose area is still smaller than the threshold after the operation, the system will force it to merge into the adjacent dominant division with the longest contact boundary or the most similar characteristics.

[0115] After the above processing, a virtual division set is generated, which is composed of several regions with smooth edges, moderate areas and uniform internal characteristics. Each virtual division is assigned a unique ID and records all the grid indexes it contains as an independent unit for subsequent VAV control.

[0116] Step S404, according to the equipment space layout information in the multi-source fusion basic data set, calculate the air supply influence weight of each VAV terminal on each region in the virtual division set, and construct the VAV terminal and virtual division coupling relationship matrix.

[0117] Since the boundary of the virtual division is dynamically changing, and the physical location of the VAV terminal is fixed, a dynamic many-to-many relationship is presented between them. The system traverses each VAV terminal, and based on the device-to-grid weight matrix, it summarizes and calculates the sum of the influence weight of the VAV terminal on all grids in a certain virtual division Vz.

[0118] Specifically, the coupling coefficient c _kz represents the air supply contribution rate of the kth VAV terminal to the zth virtual division. c _kz is equal to the weight accumulation value of all grids covered by the VAV and falling within the virtual division z. The matrix composed of all c _kz is the VAV terminal and virtual division coupling relationship matrix. This matrix describes which virtual divisions will be affected by adjusting the air volume of a VAV, and the degree of the influence, which can be used as a constraint coefficient in the subsequent optimization model.

[0119] Step S405, taking the temperature deviation in the virtual division set and the weighted sum of system energy consumption as the objective function, taking the air volume distribution relationship defined by the VAV terminal and virtual division coupling relationship matrix and the minimum fresh air volume of the virtual division as the constraint condition, combining the short-time load prediction sequence of the division in the equivalent thermal and humidity load state set, and constructing a VAV air volume distribution optimization model.

[0120] Accordingly, the system establishes a quadratic programming model with all VAV terminal air supply quantities as decision variables. The objective function J includes three parts, including: a comfort penalty term, i.e., the square sum of the difference between the predicted temperature and the set temperature of each virtual subzone; an energy consumption term, which is approximately the cubic sum or square sum of all VAV air quantities; a regulation smoothing term, i.e., the square sum of the difference between the current air quantity and the air quantity at the last time; and the corresponding weights of the three terms.

[0121] The constraint conditions of the model include physical constraints, i.e., the air quantity of each VAV must be between the minimum air quantity and the maximum air quantity; balance constraints, i.e., the effective cooling capacity obtained by all virtual subzones must be greater than or equal to the predicted load demand, wherein the effective cooling capacity obtained by the virtual subzone is calculated by multiplying the VAV terminal and the virtual subzone coupling relationship matrix by the VAV air quantity vector and combining the supply air temperature difference; and health constraints, i.e., the total fresh air quantity obtained by the virtual subzone must meet the minimum fresh air quantity standard per capita for the predicted number of people in the region.

[0122] According to one aspect of the present application, the optimization objective function formula can also be:

[0123] minJ=Σ(α×(T _pred -T _set ) 2 +β×(V _fresh_req -V _fresh_act ) 2 +γ×P _fan );

[0124] Wherein, J is the total optimization objective value, T _pred is the predicted temperature of the virtual subzone, T _set is the set temperature, V _fresh_req is the minimum fresh air quantity required by the virtual subzone, V _fresh_act is the actual obtained fresh air quantity, P _fan is the energy consumption of the fan operation, and α, β, and γ are the weighting coefficients of the comfort penalty term, the regulation smoothing term, and the energy consumption term, respectively.

[0125] Step S406, the rolling solution is performed on it (VAV air quantity distribution optimization model) to calculate the air quantity distribution proportion and the supply air temperature of each VAV terminal in the future prediction time domain, and a set of air conditioning system control trajectories is generated.

[0126] That is, the periodic cyclic solution is performed to obtain the air quantity distribution proportion and the supply air temperature of each VAV terminal in the future prediction time domain, and a set of air conditioning system control trajectories containing the VAV terminal air quantity and the supply air temperature set value is generated.

[0127] This step involves solving and outputting the model. In each control cycle, for example every 5 minutes, the system reads the latest load forecast and zoning status, and instantiates the aforementioned optimization model. Using a standard quadratic programming solver or interior-point solver, it calculates the optimal VAV flow sequence for each time step within the future forecast time domain, such as 30 minutes.

[0128] Although the solution yields the future sequence, the system typically only issues the setpoint for the first time step as the current control command; this is the rolling optimization mechanism of Model Predictive Control (MPC). The output air conditioning system control trajectory set includes the airflow setpoints for hundreds of VAV terminals throughout the station. For systems with supply air temperature regulation capabilities, the model can also jointly optimize the supply air temperature of the air conditioning unit (AHU), further reducing energy consumption by utilizing natural cooling sources during transitional seasons by increasing the supply air temperature.

[0129] According to another aspect of this application, the clustering objective function can be calculated using the following formula:

[0130] J _cluster =Σ||V _i -C _k || 2 +λ×ΣI(C _k t ≠C _k t-1 );

[0131] Among them, J _cluster For the total cost, V _i Let C be the eigenvector of grid i. _k Let ||.|| be the center vector of the k-th cluster. 2 Let λ represent the square of the Euclidean distance, λ be the time smoothing weighting coefficient, I(.) be the indicator function, and C be the value of the Euclidean distance. _k t C is the cluster center at the current moment. _k t-1 It is the cluster center at the previous moment.

[0132] Example 5 provides an optional implementation method for platform-waiting hall linkage control driven by train timetable. It describes how to utilize railway-specific timetable data to address passenger surges and thermal shocks caused by trains entering and leaving stations, achieving coordinated control of the platform and waiting hall. Accordingly, this can be achieved through the following steps:

[0133] Step S501: Parse the train arrival and departure times and train formation information from the multi-source fusion basic dataset to generate the corresponding passenger surge event sequence, and define a linkage time window covering the pre-gathering, boarding / alighting and disembarking stages for each surge event.

[0134] This step converts the operation plan into a trigger event for the control system. The system scans the train timetable in real time and extracts the train arrival and departure information within the next hour. For each train, the system generates a passenger flow surge event object, which contains the train number, planned time, train type, number of cars, and estimated passenger load. The system defines a linkage time window based on the train type. For example, for a starting train, passenger flow mainly gathers before departure, and the time window is defined as 40 minutes before departure to the departure time; for a terminating train, passenger flow mainly disperses after arrival, and the time window is defined as 10 minutes before arrival to 20 minutes after arrival; for a passing train, the time window covers from arrival to departure.

[0135] If train delay information is received, the system will dynamically adjust the time window. For example, if the train is expected to be delayed by 15 minutes, the system will shift the original time window by 15 minutes, mark the event as a delay event, and facilitate subsequent strategies that may strengthen cold supply to cope with additional metabolic heat dissipation caused by passenger impatience.

[0136] Step S502, within the linkage time window, the historical passenger flow pattern and real-time station observation are combined to calculate the migration passenger flow between the platform and the waiting hall, and the additional sensible and latent heat loads of the corresponding virtual partitions are converted.

[0137] In this step, the heat shock caused by passenger flow migration is quantified. Further, the system maintains a historical passenger flow pattern library, which stores typical boarding and alighting passenger flow curve templates for different trains and different time periods. Within the linkage time window, the system calculates a scaling factor based on the current real-time observed number of station gates to correct the amplitude of the historical template; at the same time, the template is time-shifted according to the real-time delay time.

[0138] On this basis, the system calculates the migration passenger flow between the platform and the waiting hall. For the ticket checking process, the migration passenger flow is represented by the number of people in the corresponding ticket checking area of the waiting hall decreasing at a certain rate, while the number of people in the corresponding area of the platform increases at the same rate. The system multiplies the migration number flow by the human metabolic rate to calculate the load reduction of the virtual partition in the waiting hall and the load increase of the platform area, i.e. the additional sensible and latent heat loads. The above calculation clearly shows the transfer process of the load in space, rather than disappearing or appearing out of thin air.

[0139] An example, the passenger flow curve correction formula can be described as:

[0140] Curve _pred (t)=κ×Curve _template (t-Δt);

[0141] Where Curve _pred (t) is the corrected predicted passenger flow curve, κ is the scaling factor based on the current measured passenger flow intensity, Curve_template This is a template for a typical historical passenger flow curve. Δt is the translation time based on train delay information; in other words, t is the predicted time.

[0142] Step S503: The additional sensible and latent heat loads of the human body are superimposed onto the baseline predicted values ​​of the zoned equivalent heat and humidity load state set to generate a short-term equivalent heat and humidity load prediction sequence for the platform and waiting hall that reflects the driving characteristics of train operation. This is used to correct the load prediction.

[0143] The baseline load forecast is primarily based on inertial extrapolation from the current moment, resulting in a lag in response to sudden train arrivals and departures. This step adds the additional load generated by passenger flow migration to the baseline load in a time-aligned manner. For example, five minutes before a train arrives, the predicted load in the platform area will spike due to the added expected disembarking passenger load. The corrected short-term equivalent thermal and humidity load forecast sequence for the platform and waiting hall can anticipate upcoming passenger surges. This sequence not only includes numerical values ​​but also characterizes the arrival time and duration of load peaks, providing a targeted approach for subsequent feedforward control.

[0144] Step S504: Based on the short-term equivalent heat and humidity load prediction sequence of the platform and waiting hall, identify the load ramp-up period before the arrival of passenger surges, and calculate the pre-cooling and pre-heating correction amount for the corresponding virtual zone of the waiting hall. This is used to execute the feedforward control strategy on the waiting hall side.

[0145] Specifically, the system analyzes the corrected load forecast curve and identifies periods when the load ramp-up rate exceeds a preset threshold. To offset the sudden temperature rise caused by large passenger flows, the system employs a pre-cooling or pre-heating strategy. Under summer conditions, the system calculates the pre-cooling correction amount. This correction amount is proportional to the predicted load surge intensity.

[0146] Specifically, within the time window before the surge arrives (e.g., 15 minutes in advance), the system will set the air conditioning temperature T for the corresponding virtual zone of the waiting hall. _set Temporarily reduce the pre-cooling correction (e.g., reduce by 1 to 2 degrees Celsius). This utilizes the building structure and air's heat capacity to store cold air in advance. When peak passenger flow actually arrives, although the cooling load increases dramatically, the peak indoor temperature will not exceed the comfort limit because the ambient base temperature has already decreased, achieving a peak-shaving and valley-filling effect.

[0147] According to one aspect of this application, the temperature correction amount can be set using the following calculation formula:

[0148] ΔT _set =-K _p ×ΔQ _surge ;

[0149] Where, ΔT _setK is the adjustment value of the temperature setting _p ΔQ is the pre-cooling gain coefficient _surge is the predicted load surge intensity.

[0150] Step S505, according to the station area load demand in the station and waiting hall linkage short-time equivalent heat and humidity load prediction sequence, deduce the station ventilation equipment start-stop and operation gear that meets the station comfort and fresh air requirements. Used to execute the control strategy on the station side.

[0151] In this embodiment, the station environment is usually semi-open or large space, and the control focus is on ventilation and exhaust. The system calculates the minimum fresh air volume required to maintain the carbon dioxide concentration below 1000ppm according to the predicted number of people gathered in the station. At the same time, according to the predicted heat load, the ventilation volume required to remove excess heat is calculated. The system takes the maximum value of the two as the target ventilation volume. According to the target ventilation volume, the system consults the performance curve of the station fan equipment to determine the number of fans that need to be turned on and the operation frequency gear of the variable frequency fan.

[0152] For example, during the train station stop, the system automatically instructs the station exhaust fan to run at high speed to remove the heat and humidity of the train braking and the people gathering; after the train leaves the station, the system instructs the fan to slow down or enter the stop mode to save energy.

[0153] Step S506, integrate the waiting hall pre-cooling and pre-heating correction amount and the station ventilation equipment start-stop and operation gear, combined with the basic air volume setting of the VAV terminal, to generate a set of air conditioning system control trajectories containing the execution instructions of the whole system of the waiting hall and the station. Used to summarize and generate control instructions.

[0154] Correspondingly, the system applies the waiting hall temperature setting correction amount to the VAV air volume optimization result, which may trigger re-optimization or directly bias the setting value. At the same time, the station equipment instructions are combined. The generated air conditioning system control trajectory set is an instruction package containing the future action sequence of the whole station air conditioning and ventilation equipment. The system issues instructions through the BAS interface to coordinate and schedule the AHU, VAV of the waiting hall and the fan system of the station, and completes the active heat and humidity environment protection for the whole process of train entering and leaving the station.

[0155] Embodiment 6, provides an exemplary scheme of algorithm optimization and system hardware implementation, especially about data quality control, optimization model solution acceleration and passenger flow template construction, and also provides a physical hardware implementation architecture of the method of the application. On this basis, the scheme includes:

[0156] Step S601, perform multi-source consistency check and outlier identification based on physical constraint rules and statistical methods, and generate data quality labels.

[0157] On the basis of data alignment, the system further performs consistency verification. Correspondingly, the system applies rule checking based on physical constraints. For example, for temperature sensor data, the system sets a reasonable change rate threshold, and if the temperature of a certain measuring point changes by more than 2 degrees Celsius within 1 minute, it is marked as a physically infeasible anomaly; for relative humidity data, the system verifies it in combination with the outdoor dew point temperature, and if the indoor humidity reading is lower than the theoretical limit of the outdoor air after being cooled and dehumidified, it is determined to be a sensor failure.

[0158] Next, the system applies statistical methods for outlier detection. For multiple redundant temperature sensors in the same air conditioning partition, the system calculates the mean and standard deviation of their readings, and uses the 3-sigma rule to mark readings deviating from the mean by more than 3 times the standard deviation as outliers.

[0159] In addition, for passenger flow data, the system compares video recognition counting with gate card data, and if the deviation exceeds a preset ratio such as 20%, the data for that period is marked as low confidence. The system attaches a quality label to each data point, classified as normal, suspicious, and invalid, with a confidence score, which can be used as a weighting basis for the observation noise covariance matrix in subsequent Kalman filter inversion.

[0160] It should be understood that this step belongs to the quality control link in data preprocessing.

[0161] Step S602, construct train type-passenger flow pattern library, based on train operation plan to perform passenger flow template retrieval and splicing.

[0162] The system uses historical long-term operation data in the offline stage to construct a train type-passenger flow pattern library. The system clusters historical trains by train type such as high-speed rail, conventional speed, and day type such as weekday, holiday, and extracts typical boarding and alighting passenger flow curve templates for each type of train in different platform segments and waiting hall areas. The template describes the normalized passenger flow intensity as a function of time. In the online running stage, for each train event in the future prediction time domain, the system retrieves the matching typical passenger flow curve template from the pattern library according to the train attributes in the received train operation plan.

[0163] Further, the system performs template splicing operation to align each retrieved single train template to the global time axis according to its actual planned departure time. For overlapping time periods, the system linearly superimposes the passenger flow intensities of multiple templates to generate an initial global passenger flow prediction curve. Based on the template splicing method, the passenger flow waveform synthesis problem in the multi-train concurrent and frequent late departure scenario is solved.

[0164] It should also be understood that this step expands the template construction mechanism in passenger flow prediction.

[0165] Step S603, linearization or convexity processing is performed on the VAV air volume distribution optimization model, and a quadratic programming problem is constructed to adapt to real-time rolling solution.

[0166] Since the physical process of the air conditioning system contains nonlinear links, such as the cubic relationship between fan energy consumption and air volume, and the nonlinear coupling between temperature response and air volume, it takes a long time to directly solve the nonlinear programming problem, which is difficult to meet the real-time control requirements. In this embodiment, a linearization and convexity approximation strategy is adopted. The system selects the current working point as the reference, expands the Taylor series of the nonlinear temperature response equation near the working point, and retains the first-order term to realize linearization.

[0167] For the fan energy consumption term, the system approximates it as a quadratic function of air volume to ensure the convexity of the objective function. After the above processing, the originally complex nonlinear optimization model is converted into a standard quadratic programming QP problem or a second-order cone programming SOCP problem. This enables the system to use a solver to complete the air volume distribution calculation of hundreds of VAV terminals in the station within milliseconds, ensuring the real-time and stability of the model predictive control MPC rolling mechanism.

[0168] It should also be understood that this step illustrates an optional solution acceleration strategy for the optimization model.

[0169] According to one aspect of the present application, the partial method of the present application can also be performed in the following way:

[0170] Optionally, using the air conditioning partition-function area mapping relationship and space topology information, for each air conditioning partition, the boundary envelope structure (outer wall, inner wall, roof, floor, door and window) and adjacent space are retrieved, and the area, construction level (heat transfer coefficient U obtained from the material library) and orientation of the envelope structure are extracted from the BIM model. Determine the heat exchange relationship with the outdoor or non-air conditioning space. Based on the above information, generate a heat balance and moisture balance topology structure for each partition, i.e. use a simplified model of a single node air + equivalent heat capacity, air temperature and humidity content as state variables; envelope heat transfer, permeation ventilation, supply air / return air, human body heat dissipation, etc. as input items or disturbance items; form a topology graph of the partition-adjacent environment / system, indicating the flow paths of heat and moisture. On this basis, store the topology structure in the form of a graph or a matrix (such as an adjacency matrix, a parameter table).

[0171] According to the topology graph, define for each air conditioning partition: a set of state variables, including the average air temperature, specific humidity or absolute humidity in the partition; a set of control input variables, including the supply air mass flow, supply air temperature a, and fresh air ratio; a set of disturbance input variables, including human body sensible heat / potential heat disturbance, envelope heat gain (which can be estimated by combining weather and envelope parameters), and permeation ventilation load (which can be estimated by door and window opening and passenger flow disturbance); a set of output variables, including measured quantities such as partition measurement point temperature and humidity.

[0172] On this basis, the observation corresponding to each partition is one-to-one mapped with the above output variable, and an observation variable-model output mapping table is established; the mapping table can be used to compare the measured value with the model simulation value for subsequent state and parameter estimation; it can also be used to map the model output back to the actual physical quantity for prediction.

[0173] Further, according to the equipment rated parameters and the building envelope information, the initial physical parameters of each partition are estimated, for example, the air mass and the equivalent heat capacity, which are estimated by using the partition volume, air density and specific heat, and the equivalent heat capacity of furniture / building interior; the equivalent heat loss coefficient is obtained by comprehensively considering the total heat transfer coefficient of the building envelope and the area; the air infiltration ventilation rate is set as the initial value according to the number of doors and windows and the passenger flow level classification by referring to the design value and the empirical coefficient.

[0174] Correspondingly, it is assumed that the air in the partition is fully mixed and the radiation non-uniformity is ignored, and the complex internal load (lighting, equipment) is preliminarily combined as an equivalent internal sensible heat disturbance, and the assumptions are recorded for limiting the use range of the model. The above initial values of the parameters are arranged into a parameter vector, together with the model topology and variable definition, to constitute a basic model of thermal and humidity balance of the partition.

[0175] Optionally, the passenger flow density, throughput, residence time and other data arranged according to the grid, and the grid-air conditioning partition mapping relationship are read. At each time step, for each air conditioning partition, the passenger flow count, residence time and other indicators of all the grids covered by the partition are summed / averaged according to the area and weight to obtain the partition-level passenger flow indicators, including real-time presence number estimation, average residence time, passing rate (person / minute), activity intensity proxy indicators (such as walking speed, crowding degree). The operation labels such as weekdays / weekends / holidays are associated to provide scene information (such as clothing, activity mode difference) for subsequent human load mapping.

[0176] Optionally, based on the human heat balance model and the pre-stored human metabolic rate-activity intensity-clothing parameter library (as system preset parameters), the passenger flow behavior characteristics are converted into the human thermal and humidity load of each partition. Correspondingly, according to the season, outdoor temperature and operation scene, the appropriate clothing thermal resistance range is selected; according to the residence time and walking speed, the on-site personnel are divided into behavior categories such as sitting / waiting, slow walking and fast passing, which correspond to different metabolic rates. For each type of population, the sensible heat / latent heat dissipation rate per person is calculated, multiplied by the real-time number, to obtain the sensible heat / latent heat load of each type of population in the partition; the time series of the total human sensible heat load and latent heat load of the partition are obtained, which can be used as part of the disturbance input.

[0177] Optionally, the permeation air exchange rate of the subzone is estimated by using the door area passenger flow, door passage opening state (if there is a monitoring point), combined with static information such as door and window area and sealing level. For each entrance, the instantaneous permeation air volume is calculated according to the unit passenger flow caused by the air exchange rate formula or the simplified flow model; combined with the indoor and outdoor temperature and humidity difference (which can be obtained from the outdoor meteorological data), the sensible heat and latent heat load time series caused by permeation are obtained.

[0178] Optionally, combined with the equipment running state and design specification document, the internal sensible heat load of lighting, equipment (such as advertising screen, elevator, escalator) is estimated. According to the rated power and load rate curve of the equipment, the running state is converted into equivalent sensible heat output; according to the area attribution, the equipment load is aggregated to the corresponding air conditioning subzone. The human body load, permeation load, lighting and equipment load and the like are summarized to form the total internal disturbance load time series at the subzone level, that is, the sensible heat disturbance and the latent heat disturbance.

[0179] Illustratively, taking the subzone heat and moisture balance basic model as the starting point, the continuous time heat and moisture balance equation is discretized according to a uniform time step, for example, for air temperature, the following processing is performed:

[0180] Cz×(T _z_k+1 +1-T _z_k ) / Δt=Q _sa_k +Q _env_k +Q _int_k +Q _eq_k ; the moisture content can also be processed similarly;

[0181] Wherein, T _z_k corresponds to the indoor temperature at a certain time, T _z_k+1 corresponds to the temperature at the next time, Δt is the time step, Q _sa_k , Q _env_k , Q _int_k , Q _eq_k are the cold and heat provided by the supply air at k time, the building envelope heat transfer, the internal known disturbance, and the equivalent load to be inverted, respectively.

[0182] Another example is to arrange the above equation into a discrete state space form, specifically:

[0183] x _k+1 =A(θ)×x _k +B(θ)×u _k +E(θ)×d _k +G(θ)×q _k ;

[0184] y _k =C×x _k ;

[0185] Above, d _kA(θ), B(θ), E(θ), G(θ) are the system evolution matrix, control input matrix, known disturbance coefficient matrix, unknown load disturbance matrix, respectively, depending on parameter θ, corresponding to known disturbances (human load, infiltration load, etc.).

[0186] In another example, an extended state vector is constructed, and model parameters and equivalent loads are introduced into a joint estimation framework, as shown in the following equation:

[0187] z _k =[x _k , θ _k , q _k ] T ;

[0188] Further, according to the degree of nonlinearity of the system and the real-time requirement, a suitable online estimation method is selected, such as constrained extended Kalman filter (cEKF), unscented Kalman filter (UKF), ensemble Kalman filter (EnKF), or constrained least squares estimation based on sliding time window.

[0189] In another example, data quality labels / confidence is introduced into the estimation algorithm as a basis for setting observation noise covariance or weight. For example, observations with high confidence have a greater impact on updates; for abnormal or low-confidence observations, their weight in estimation updates is reduced.

[0190] In another example, constraints are set for each variable to prevent divergence of inversion results and parameter estimation or to comply with physical laws. For example, the upper and lower limits of parameters such as heat transfer coefficient and infiltration ventilation rate are determined according to the design value plus or minus a certain percentage range; the range of equivalent load is jointly limited by internal load and equipment rated capacity, such as not exceeding the maximum cooling capacity of the unit, not appearing long-term large sensible heat, etc.; the temperature and humidity state variables are limited to a reasonable comfort range to avoid numerical instability.

[0191] Further, in the filtering or least squares iteration process, the estimated values that exceed the physically feasible region are projected or constrained to shrink back into the feasible region, ensuring that the model always remains physically interpretable.

[0192] In another example, at each time step k, the current partition observation y _k and control input u _k are read, the disturbance d _k is read, the estimation algorithm is initialized from the initial parameters or the parameters estimated in the previous step; the joint estimation algorithm is run to obtain the updated state, parameters, and equivalent load. The updated equivalent load is split into equivalent sensible heat load and equivalent latent heat load, and a short-time-window smoothing filter (such as moving average) is performed to suppress jitter caused by measurement noise. The updated parameters are stored in the model parameter library.

[0193] In one possible implementation, a uniform time step (such as 1 minute) is selected, and a rolling prediction time domain length (such as the next 30-60 minutes) is defined. The state estimation and model parameters at the current time are read as the prediction starting point and parameters. The disturbance time series of human load, infiltration wind, outdoor weather, etc. in the prediction time domain are read; the current control settings (air volume, supply air temperature, etc.) or the planned control trajectory (such as the optimization results of the last rolling period) are collected, and the prediction stage control assumption is set, that is, if there is no update control plan, it is assumed that the control quantity remains unchanged at the current value in the prediction time domain; if a preset trajectory has been output, the trajectory is used as the prediction input.

[0194] For each air conditioning partition, the partition heat and humidity balance basic model and the corrected parameters are used to perform discrete-time forward simulation in the prediction time domain. At each future time step, the predicted disturbance and control assumption are input, and the predicted state is calculated according to the state space equation.

[0195] On the basis of simulation, the equivalent load required to maintain or approach the expected temperature and humidity trajectory is backstepped, and according to the energy balance, the known items such as building heat transfer, infiltration load, internal disturbance are calculated; the target temperature / humidity (usually the comfort set value or the slow-changing trajectory based on the current temperature) is used to constrain the state evolution, and the net equivalent load required by the air conditioning system under this condition is solved. The equivalent load sequence in the prediction time domain is obtained, which reflects the load level required to meet the comfort target under the current / expected control strategy.

[0196] Further, the predicted equivalent load sequence is post-processed. Accordingly, the physical rationality (such as the instantaneous load does not exceed the maximum capacity of the equipment, and does not appear large oscillation) is checked; the short-time pulse load is smoothed (for example, low-pass filtering or constraining the change rate) to make it more suitable for subsequent optimization calculation.

[0197] According to the partition-grid-equipment mapping, the partition-level load is aggregated downward / upward. Specifically, downward, the partition load is split according to the grid area or weight, and the virtual partition level is used to provide fine-grained load characteristics; upward, a plurality of partition loads are aggregated into the prediction load at the device level such as AHU / VAV, to provide constraints and targets for optimization control.

[0198] In the present application, the real-time passenger flow thermal map is collected and dynamic clustering with spatial constraints is performed, the virtual partition changing with the passenger flow streamline is constructed, the dynamic coupling weight matrix of the VAV terminal and the virtual partition is established, and the problems of uneven cooling and heating and energy waste caused by the fixed civil partition that cannot adapt to the tidal passenger flow distribution are solved.

[0199] Further, the discrete state space equation and the extended Kalman filtering algorithm are used to realize online decoupling inversion of the equivalent sensible heat and latent heat loads in the partition, forward rolling simulation is carried out based on multi-source data, and the traditional temperature lag feedback is replaced by the predictive control based on the physical model, so that the control instability problem caused by the thermal lag of the large space is solved.

[0200] In addition, the embodiment parses the train working diagram into passenger flow surge events, quantifies the transfer load between the platform and the waiting hall, generates the feedforward control instruction containing pre-cooling and pre-heating, and solves the problem that the instant heat shock caused by the train entering and leaving the station cannot be coped with in the independent control mode.

[0201] The above describes the optional embodiments of the application in detail, but the application is not limited to the specific details in the above embodiments, and various equivalent transformations can be made to the technical solutions of the application within the technical concept range of the application, and these equivalent transformations all belong to the protection range of the application.

Claims

1. A method for dynamic optimization control of a railway station air conditioning system based on multi-source perception fusion, characterized in that, Comprise: Call multi-source fusion basic data set, including passenger flow spatial distribution characteristics, train operation plan, equipment spatial layout information, static building parameters, real-time environment and equipment operation data, operation plan and historical data, passenger flow micro-behavior characteristics; Based on the multi-source fusion basic data set, a simplified heat and moisture balance model is constructed, and the sensible heat and latent heat exchange of each physical partition is inversely calculated online to generate a partition equivalent heat and moisture load state set; Extract the passenger flow spatial distribution characteristics, combine the partition equivalent heat and moisture load state set to perform spatial dynamic clustering, identify and generate a virtual partition set; According to the virtual partition set and the train operation plan, the linkage load demand of the platform and the waiting hall is predicted, the VAV terminal is optimized and solved, and the air conditioning system control trajectory set is generated; Wherein, extracting passenger flow spatial distribution characteristics, combining partition equivalent heat and moisture load state set to perform spatial dynamic clustering, identifying and generating virtual partition set, comprising: extracting waiting hall real-time passenger flow density thermal map from multi-source fusion basic data set, mapping it to the preset spatial grid unit, combining partition equivalent heat and moisture load state set, constructing grid-level comprehensive feature vector containing passenger flow density, load intensity and function attribute; Based on the grid-level comprehensive feature vector, the clustering algorithm with spatial connectivity constraint is used to aggregate the spatial adjacent and similar grid units to form an initial regional division; Perform morphological smoothing processing on the initial regional division to eliminate isolated grids and correct the boundary shape to generate a virtual partition set containing several virtual partitions across the physical air conditioning partition boundary; Wherein, based on the multi-source fusion basic data set, a simplified heat and moisture balance model is constructed, and the sensible heat and latent heat exchange of each physical partition is inversely calculated online to generate a partition equivalent heat and moisture load state set, comprising: using static building parameters, establishing a partition heat and moisture balance discrete state space equation describing the evolution of temperature and humidity content of each physical partition, defining equivalent sensible heat load and equivalent latent heat load as unknown state variables to be inverted; Extract real-time passenger flow and behavior data and door and window opening state from multi-source fusion basic data set, calculate human sensible heat, human latent heat and infiltration wind load of each physical partition, generate partition real-time internal disturbance input set; Substitute the partition real-time internal disturbance input set and the real-time environment and equipment operation data into the partition heat and moisture balance discrete state space equation, and use the joint state parameter estimation algorithm to solve the equivalent sensible heat and latent heat load at the current time and update the model parameter set, and store it in the partition equivalent heat and moisture load state set.

2. The method of claim 1, wherein, According to the virtual partition set and the train operation plan, the linkage load demand of the platform and the waiting hall is predicted, the VAV terminal is optimized and solved, and the air conditioning system control trajectory set is generated, comprising: According to the equipment spatial layout information, calculate the air supply influence weight of each VAV terminal to each region in the virtual partition set, and construct the VAV terminal and virtual partition coupling relationship matrix; A VAV air volume distribution optimization model is constructed, with a temperature deviation in a virtual partition set and a system energy consumption weighted sum as an objective function, an air volume distribution relationship defined by a VAV terminal and a virtual partition coupling relationship matrix and a minimum fresh air volume of the virtual partition as constraint conditions, and a partition short-time load prediction sequence in a partition equivalent heat and moisture load state set. The model is solved in a rolling manner to calculate air volume distribution proportions and supply air temperatures of the VAV terminals in a future prediction time domain, and to generate an air conditioning system control trajectory set.

3. The method of claim 1, wherein, The method further includes: According to an operation plan and historical data, a passenger flow change trend and an outdoor meteorological parameter in the future prediction time domain are extrapolated to generate a partition short-time disturbance input prediction sequence; An updated model parameter set is called to perform forward rolling simulation on a partition heat and moisture balance discrete state space equation under a preset comfort control target constraint, with the current time equivalent sensible and latent heat loads as initial states and in combination with the partition short-time disturbance input prediction sequence; The required sensible and latent heat exchange amounts for maintaining the comfort control target at each time step in the future are deduced in reverse to generate a partition short-time equivalent heat and moisture load prediction sequence, which is updated to the partition equivalent heat and moisture load state set.

4. The method of claim 1, wherein, According to the virtual partition set and a train operation plan, a linkage load demand of a platform and a waiting hall is predicted, including: Train arrival and departure time and marshalling information are parsed from a multi-source fusion basic data set to generate a corresponding passenger flow surge event sequence, and a linkage time window covering pre-aggregation, boarding and alighting and dispersing stages is defined for each surge event; In the linkage time window, in combination with a historical passenger flow mode and real-time in-out station observations, a platform and a waiting hall transfer passenger flow is calculated, which is converted into additional human body sensible and latent heat loads of corresponding virtual partitions; The additional human body sensible and latent heat loads are superimposed on baseline predicted values in the partition equivalent heat and moisture load state set to generate a platform and a waiting hall linkage short-time equivalent heat and moisture load prediction sequence reflecting train operation driving characteristics.

5. The method of claim 4, wherein, The VAV terminal is optimized to generate an air conditioning system control trajectory set, including: Based on the platform and the waiting hall linkage short-time equivalent heat and moisture load prediction sequence, a load climbing period before a passenger flow surge arrives is identified, and a waiting hall pre-cooling and pre-heating correction amount for corresponding virtual partitions is calculated; According to a platform area load demand in the platform and the waiting hall linkage short-time equivalent heat and moisture load prediction sequence, a platform ventilation equipment start-stop and operation gear that meets platform comfort and fresh air requirements is deduced; In combination with a basic air volume setting of the VAV terminal, the waiting hall pre-cooling and pre-heating correction amount and the platform ventilation equipment start-stop and operation gear are comprehensively considered to generate an air conditioning system control trajectory set containing a waiting hall and a platform full-system execution instruction.

6. The method of claim 1, wherein, The multi-source fusion basic data set is constructed, specifically including: Station static building parameters are collected, and a device-space grid-air conditioning partition multi-level mapping relationship containing physical location coordinates and control logic association is constructed based on spatial geometric information and mechanical and electrical system topology therein; According to the multi-level mapping relationship of equipment-space grid-air conditioning zoning, the real-time equipment operation data and real-time passenger flow perception data are mapped to a unified space grid unit, resampled and aligned according to a preset time step, and a structured and unified spatiotemporal index multi-source fusion basic data set is generated.

7. The method of claim 1, wherein, Perform morphological smoothing processing on the initial regional division, eliminate isolated grids and correct the boundary shape, generate a virtual zoning set containing several virtual zones that cross the physical air conditioning zoning boundaries, including: Introduce a time smoothing penalty term, calculate the change cost of each grid cell in the initial regional division relative to the virtual zoning result of the previous time, and suppress the regional jump caused by short-term passenger flow fluctuations; Perform spatial morphological opening and closing operation on the initial regional division, merge small isolated regions with an area below a preset threshold to adjacent dominant regions, smooth the jagged structure of the regional boundary, and generate a virtual zoning set that is stable in the time domain and continuous in the space.

8. The method of claim 1, wherein, Calculate the sensible heat, latent heat and permeation wind load of each physical zone, generate a real-time internal disturbance input set, including: According to the micro-behavior characteristics of the passenger flow, the people in each physical zone are divided into categories such as sitting and waiting, slow walking and fast passing, and the corresponding differential human metabolic rates are matched respectively; Based on the differential human metabolic rate, the heat dissipation and moisture dissipation of each category of personnel are calculated, combined with the air permeation ventilation volume caused by the door opening frequency, and the total sensible heat disturbance and total latent heat disturbance of each physical zone are synthesized respectively to generate a real-time internal disturbance input set.

Images