Energy-saving operation control method and system of clean air conditioning system

Abstract

The present application relates to the technical field of control, in particular to an energy-saving operation control method and system of a clean air conditioning system, the method comprising: obtaining preset parameters and real-time operation data of the clean air conditioning system; identifying, based on the real-time operation data, whether the system is currently in a steady-state or transient-state operation mode; establishing a target function containing system energy consumption, filter loss and personnel exposure risk; solving, according to the operation mode, an optimal control sequence that minimizes the value of the target function in a future prediction time domain; issuing the first control instruction of the optimal control sequence to the corresponding execution mechanism, and periodically repeating the above steps to realize closed-loop optimization control of the clean air conditioning system. The present application can comprehensively improve the economy, reliability and safety of system operation.

Description

Technical Field

[0001] This invention relates to the field of control technology. More specifically, this invention relates to an energy-saving operation control method and system for a cleanroom air conditioning system. Background Technology

[0002] Cleanroom air conditioning systems are an indispensable key infrastructure in high-tech industries and fields such as semiconductor manufacturing, biopharmaceuticals, precision instruments, and medical and health care. Their core task is to provide and maintain a highly clean, constant temperature and humidity, and stable pressure gradient air environment for a specific space.

[0003] To meet these stringent environmental requirements, traditional cleanroom air conditioning systems typically employ a constant parameter control strategy based on maximum design load during design and operation. For example, the system sets constant air volume, fresh air ratio, and air exchange rate according to the maximum number of personnel, maximum equipment heat generation, and the most unfavorable outdoor weather conditions. While this control method ensures that the cleanroom's environmental parameters meet standards most of the time, its operation is static and inefficient. The system operates at high power for extended periods without considering the real-time dynamic changes in the cleanroom's load (such as heat, humidity, and dust sources), leading to significant energy waste. Statistics show that cleanroom air conditioning systems typically account for 30% to 70% of the total energy consumption of the entire production facility, indicating substantial potential for energy savings.

[0004] Another significant shortcoming of existing technologies lies in the limitations of their control logic. Traditional control systems mostly rely on simple setpoint control or PID feedback regulation, which have limited sensing capabilities. They typically only monitor macroscopic parameters such as temperature and humidity at a few points, failing to accurately capture the spatiotemporal distribution characteristics of pollutants in clean rooms and key disturbance factors such as personnel activity. This "passive" response mechanism results in significant lag in control behavior, making it difficult to cope with sudden pollution events or rapid changes in load. Furthermore, existing energy-saving attempts are often single-objective, such as reducing fan speed during off-peak hours, but this fails to establish a comprehensive optimization model. It ignores the inherent correlations and constraints between different operational objectives. For example, excessively reducing energy consumption may increase filter load, shorten their lifespan, or lead to pollutant accumulation in localized areas, thereby increasing the risk of personnel exposure.

[0005] Therefore, existing technologies lack an intelligent control method that can perceive in real time, predict dynamically, and perform multi-objective collaborative optimization, and cannot achieve the best balance between ensuring environmental quality, personnel safety, and equipment lifespan, thereby realizing truly refined energy-saving operation. Summary of the Invention

[0006] To address the technical problem of how to achieve truly refined energy-saving operation, this invention provides the following solution:

[0007] An energy-saving operation control method for a cleanroom air conditioning system includes the following steps: S1: Acquire preset parameters and real-time operating data of the cleanroom air conditioning system; the real-time operating data includes outdoor environmental parameters, temperature and humidity data, pressure difference data, particulate matter concentration data, airflow characteristic data, and personnel location and activity information of each indoor area; S2: Based on the real-time operating data, identify whether the system is currently in a steady-state or transient operating mode, and use a dynamic prediction model that is continuously corrected based on real-time data to predict the heat and humidity load, particulate matter concentration distribution, and personnel exposure risk of the cleanroom in the future time domain; S3: Establish an objective function that includes system energy consumption, filter wear, and personnel exposure risk; according to the operating mode, solve for the optimal control sequence that minimizes the value of the objective function in the future prediction time domain; S4: Issue the first control command of the optimal control sequence to the corresponding actuator, and periodically repeat the above steps to achieve closed-loop optimization control of the cleanroom air conditioning system.

[0008] Furthermore, the personnel activity information is quantified into three levels: stationary, slow-moving, and fast-moving.

[0009] Furthermore, the preset parameters include: cleanliness level, upper and lower limits of temperature and humidity, and minimum pressure difference between areas. The upper and lower limits of temperature and humidity include upper temperature limit, lower temperature limit, upper humidity limit, and lower humidity limit.

[0010] Furthermore, determining whether the identification system is currently in a steady-state or transient operating mode includes: calculating the average rate of change of particulate matter concentration data, the average rate of change of temperature, and the average rate of change of humidity over a set time period; when the average rate of change of particulate matter concentration is higher than a threshold for particulate matter concentration change, the average rate of change of temperature is higher than a threshold for temperature change, or the average rate of change of humidity is higher than a threshold for humidity change, the system is determined to be transient; otherwise, the system is determined to be steady-state.

[0011] Furthermore, the dynamic prediction model employs an LSTM model based on a long short-term memory network.

[0012] Furthermore, the objective function J is determined by the following formula: J = αE norm +βF norm +γR norm Specifically, the system energy consumption E, filter loss F, and personnel exposure risk R in the prediction time domain are normalized to obtain dimensionless E. norm F norm and R norm α, β, and γ are the weight coefficients for system energy consumption (E), filter loss (F), and personnel exposure risk (R), respectively.

[0013] Furthermore, the filter loss F is determined by the following formula:

[0014] The filter loss F is determined by the following formula:

[0015] ;

[0016] Among them, T p The length of the prediction time domain; Q(t) is the predicted air supply volume; C in (t) represents the particulate matter concentration upstream of the filter; η represents the filter efficiency;

[0017] or: ;

[0018] Among them, the filter cost coefficient k f The depreciation cost of the filter for every gram of particulate matter intercepted.

[0019] Furthermore, the optimal control sequence that minimizes the objective function value within a future prediction time domain is obtained by: setting the prediction time domain to 60 minutes, the control time domain to 15 minutes, and the control step size to 1 minute; at the beginning of each control cycle, using sequential quadratic programming or interior point method to obtain an optimal control sequence consisting of 15 control commands, specifically the fan operating frequency and fresh air valve opening degree per minute within the next 15 minutes.

[0020] Furthermore, finding the optimal control sequence that minimizes the objective function value in the future prediction time domain also includes the following constraints: temperature, humidity, pressure difference, and particulate matter concentration do not exceed preset ranges.

[0021] The present invention also provides an energy-saving operation control system for a clean air conditioning system, including a processor and a memory. The memory stores a computer program, and the processor executes the computer program to implement the above-mentioned energy-saving operation control method for the clean air conditioning system.

[0022] Compared with existing technologies, the beneficial effects of this invention include: By acquiring comprehensive operational information, including particulate matter concentration and personnel activity, this invention can more accurately grasp the actual load and pollution source status of the cleanroom. Based on predictions of environmental changes in future time periods, this control method can respond in advance, overcoming the adjustment deviations and energy waste caused by the lag in traditional control. A comprehensive optimization model, including system energy consumption, filter wear, and personnel exposure risk, is established, enabling the trade-off and coordination of various objectives under different operating scenarios. This allows the system to find the globally optimal operating strategy while meeting strict cleanliness and safety standards, achieving not only deep energy savings but also considering filter capacity or lifespan, as well as personnel occupational health, ultimately achieving a comprehensive improvement in the system's operational economy, reliability, and safety. Attached Figure Description

[0023] Figure 1 This is a schematic illustration of a cleanroom air conditioning system architecture according to an embodiment of the present invention;

[0024] Figure 2 This is a schematic diagram illustrating the adaptive adjustment of the weights of the objective function according to an embodiment of the present invention;

[0025] Figure 3 This is a schematic diagram illustrating the scrolling optimization according to an embodiment of the present invention. Detailed Implementation

[0026] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0027] An energy-saving operation control method for a clean air conditioning system includes the following steps:

[0028] S1: Obtain the preset parameters and real-time operating data of the clean air conditioning system; the real-time operating data includes outdoor environmental parameters, temperature and humidity data of various indoor areas, pressure difference data, particulate matter concentration data and airflow characteristic data, as well as personnel location information and personnel activity information.

[0029] The preset parameters include: cleanliness level, upper and lower limits of temperature and humidity, and minimum pressure difference between areas. The upper and lower limits of temperature and humidity include upper temperature limit, lower temperature limit, upper humidity limit and lower humidity limit, which can be set and stored through a host computer or human-machine interface.

[0030] Deploy weather stations outside the cleanroom or connect to public meteorological data services to obtain the outdoor environmental parameters, including outdoor temperature data and outdoor air quality data.

[0031] Indoors, temperature and humidity sensors, differential pressure gauges, laser particle counters, and hot-wire anemometers are distributed and installed in key locations such as clean areas, changing areas, and return air ducts. Data is transmitted to a central controller via industrial buses such as Modbus or BACnet. Using ultra-wideband positioning base stations covering the entire area and positioning tags worn by personnel, the real-time three-dimensional coordinates of each person are acquired at high frequency, i.e., the personnel location information. Simultaneously, cameras deployed above key workstations utilize computer vision image recognition algorithms to analyze the number of personnel, their location, and their activity status, such as sitting, walking slowly, or operating rapidly, i.e., the personnel activity information. The temperature and humidity sensors are used for temperature and humidity data, the differential pressure gauges for pressure difference data, the laser particle counters for particulate matter concentration data, and the hot-wire anemometers for airflow characteristic data.

[0032] In one embodiment, for subsequent processing, personnel activity information needs to be quantified into three levels: stationary, slow-moving, and fast-moving. For example, a threshold for movement speed can be set, and the personnel activities identified by the computer vision image recognition algorithm can be quantified according to this threshold. The computer vision image recognition algorithm is prior art and unrelated to the inventive point of this invention, so it will not be described further. In other embodiments, multiple depth cameras can be installed in locations such as the ceiling of the clean area. These depth cameras can be, for example, Microsoft Kinect or Intel RealSense series cameras, which can simultaneously capture color images and depth information. Using advanced skeletal point recognition algorithms, the main joints of each worker's body, such as the head, shoulders, elbows, hips, and knees, can be identified in real time from the depth images. A human skeleton model can then be constructed, accurately calculating the three-dimensional spatial position of each person in the room's coordinate system.

[0033] Quantifying human activity identified by computer vision image recognition algorithms based on thresholds can be implemented as follows: Analyze the total displacement distance of a person's three-dimensional coordinates within a continuous short time window, such as the past five seconds. If an operator's total movement distance within five seconds is less than 0.1 meters, their activity is classified as stationary; if the movement distance is between 0.1 and 1 meter, it is classified as slow movement, which may correspond to routine operations or walking; if the movement distance exceeds 1 meter, it is classified as fast movement, which may indicate that the person is walking quickly or making large body movements. This quantification and classification is crucial for predicting particulate matter dispersion because fast movement will stir up more dust than a stationary state. The specific distance values ​​mentioned above are just examples and can be adjusted according to the specific circumstances in practical applications.

[0034] S2: Based on the real-time operating data, identify whether the system is currently in a steady-state or transient operating mode, and use a dynamic prediction model that is continuously corrected based on real-time data to predict the temperature, humidity, particulate matter concentration and personnel exposure risk in the clean area in the future time domain.

[0035] In an optional embodiment, the identification system's current operating mode (steady-state or transient) is determined by: calculating the average rate of change of particulate matter concentration data, the average rate of change of temperature, and the average rate of change of humidity at multiple indoor points over the past 5 minutes; if the average rate of change of particulate matter concentration is higher than 15% / minute, and the average rate of change of temperature or humidity is higher than 10% / minute, the system is determined to be transient; otherwise, the system is determined to be steady-state. Specifically, the system can continuously monitor and record 0.5-micron particulate matter concentration, temperature, and humidity data at multiple locations within the clean area.

[0036] The aforementioned dynamic prediction model employs an LSTM model based on a Long Short-Term Memory (LSTM) network. In one embodiment, this LSTM model takes real-time operational data from the past 30 minutes as input and outputs predicted values ​​for particulate matter concentration, temperature, humidity, and personnel location at each monitoring point in the clean area for the next 60 minutes. Every 10 minutes, the network weights of the LSTM model are updated using an online learning algorithm based on the latest collected real-time operational data. LSTM models are highly adept at processing and predicting time-series data. In actual operation, the system packages all relevant data collected within the past 30 minutes into an input sequence. This includes, but is not limited to, particulate matter concentration, temperature, humidity, and differential pressure data measured by all sensors, the operating frequency of the air conditioning system fans, the opening degree of the fresh air valve and return air valve, and the real-time location and activity level of all personnel captured by cameras.

[0037] After receiving 30 minutes of historical data, the LSTM model uses its complex internal network structure to generate a detailed prediction for the next 60 minutes. This prediction is not a single value, but a dynamic dataset containing particulate matter concentration, temperature, humidity, and personnel location data for each monitoring point every minute. To ensure continuous accuracy, the model is not static. Every ten minutes, the system uses newly collected data from the past ten minutes to fine-tune the LSTM model's internal network weights through an online learning algorithm. This process allows it to continuously adapt to subtle changes that may occur in the cleanroom environment, thus maintaining high-precision prediction capabilities. Regarding the LSTM model, it is a current technology. By collecting relevant input and output data for training, a trained LSTM model acquires the ability to reason based on input information. Its specific capabilities and architecture will not be elaborated upon here.

[0038] Personnel exposure risk, denoted by R, is a quantitative indicator. Its core idea is to estimate the total amount of particulate matter inhaled by all workers in the cleanroom over a future period. A higher R value indicates a greater health risk to personnel. Since the LSTM model can predict particulate matter concentration and personnel location, combining these factors yields the personnel exposure risk, the specific calculation method of which will be revealed below. Furthermore, the exposure dose mentioned below refers to the total amount of particulate matter an individual worker is exposed to within a specific time period. This is an individual-specific quantitative indicator used to assess the potential health risks faced by individuals in a cleanroom environment.

[0039] S3: Establish an objective function that includes system energy consumption, filter loss, and personnel exposure risk; based on the operating mode, solve for the optimal control sequence that minimizes the value of the objective function in the future prediction time domain.

[0040] Wherein, the objective function J=αE norm+βF norm +γR norm The system energy consumption E, filter loss F, and personnel exposure risk R in the prediction time domain are normalized to obtain dimensionless E. norm F norm and R norm α represents the system energy consumption weight, β represents the filter loss weight, and γ represents the personnel exposure risk weight.

[0041] This method aims to optimize three objectives simultaneously: system energy consumption E, typically in kilowatt-hours; filter loss F, in mass or equivalent monetary units; and personnel exposure risk R, in cumulative particulate matter quantity. Since these three indicators have vastly different dimensions and numerical ranges, direct addition is meaningless; therefore, normalization is required first. For example, the predicted value of each indicator is linearly mapped to the interval between 0 and 1, resulting in a dimensionless E. norm F norm and R norm This allows them to be compared and weighted on the same scale.

[0042] In one embodiment, the system energy consumption E can be calculated based on the fan power and cooling water pump power, using the following formula:

[0043] ;where P fan (t) represents the instantaneous power P of the wind turbine at time t. pump (t) represents the instantaneous power of the cooling water pump at time t; the power of the fan and cooling water pump, which belong to the clean air conditioning system, is also included in the real-time operating data.

[0044] In one embodiment, the filter loss F is determined by the following formula:

[0045] ;

[0046] Among them, T p The length of the prediction time domain; Q(t): the predicted air supply volume; C in (t): Particle concentration upstream of the filter; η: Filter efficiency. Here, filter loss F physically represents the total mass of particles trapped by the filter within the prediction time domain.

[0047] In one embodiment, a filter cost factor k can also be added. f The corresponding formula is:

[0048] .

[0049] The filter cost factor represents the depreciation cost of the filter for every gram of particulate matter intercepted. It can be obtained by summing the purchase price of a new filter and the labor cost of replacement, then dividing by the filter's designed total fouling capacity. The filter cost factor converts filter loss F from "the mass of trapped particulate matter" to "monetary cost." A higher value indicates a more expensive replacement, and the optimization algorithm will be more inclined to adopt operating strategies that extend the filter's lifespan.

[0050] The risk of human exposure, R, is determined by the following formula:

[0051] ;

[0052] Where N is the total number of people in the clean area; x i (t) represents the predicted spatial location of the i-th person at time t; C(x) i (t),t) represents the predicted position at time t and location x. i (t) is the particulate matter concentration; the personnel exposure risk R is the risk of all personnel in the prediction time domain T. p The cumulative exposure dose within the body.

[0053] The core idea behind the personnel exposure risk R is to calculate the total amount of particulate matter inhaled by all workers within the predicted time domain. The system uses a prediction model to obtain the three-dimensional location x of the i-th worker at each future time t. i (t), and the predicted particulate matter concentration C(x) at the same location at the same time. i (t),t). By summing this concentration value over the entire prediction time domain, the total particulate exposure dose for the worker can be obtained. Finally, by summing the exposure doses of all workers in the clean area, the total personnel exposure risk index R is obtained. The higher this index, the greater the health risk to the personnel.

[0054] The weighting coefficients α, β, and γ reflect the relative importance of different objectives, and their settings are dynamically adaptive. When the system is determined to be in a steady state, such as during a dust-free night or a period of stable production, the primary objectives are energy saving and reducing long-term costs. Therefore, the energy consumption weight α is set to a relatively high 0.6, the filter loss weight β to 0.25, and the personnel risk weight γ to a relatively low 0.15. However, once the system enters a transient state, such as when personnel enter or a contamination event occurs, the focus of the control strategy immediately shifts. At this time, the weights are adjusted so that α equals 0.2, β equals 0.1, and the personnel risk weight γ increases significantly to 0.7. This forces the optimization algorithm to prioritize all necessary measures, even if it consumes more energy, to quickly reduce the particulate matter concentration in the area where personnel are located, ensuring personnel safety.

[0055] Finding the optimal control sequence that minimizes the objective function value within a future prediction time domain includes: using numerical optimization algorithms such as sequential quadratic programming to find a series of combinations of control variables that minimize the total value of the objective function J within the next hour prediction time domain (this is a nonlinear optimization problem). In this embodiment, the control variables include the fan operating frequency and the fresh air valve opening degree, for example, the sequence of fan operating frequency and fresh air valve opening degree for the next twelve five-minute time steps.

[0056] Specifically, the prediction time domain can be set to 60 minutes, the control time domain to 15 minutes, and the control step size to 1 minute. At the beginning of each control cycle, the nonlinear optimization problem is solved by sequential quadratic programming or interior point method to obtain an optimal control sequence consisting of 15 control commands. The commands are specifically the fan operating frequency and fresh air valve opening degree per minute in the next 15 minutes.

[0057] Sequential quadratic programming is a very mature and classic existing technique, which will not be elaborated on here.

[0058] This method does not calculate the control strategy for a long future timeframe all at once, but instead employs a more flexible and robust rolling optimization approach. The system looks ahead to the next sixty minutes, the prediction time domain, to assess the long-term impact of control decisions. However, it only develops a detailed control plan for the next fifteen minutes, known as the control time domain. This plan is broken down into minute increments, with a control step size of one minute. Therefore, at each decision point, such as 10:00 AM, the optimization algorithm aims to find an optimal sequence of control commands.

[0059] The calculation process requires determining the weighting coefficients, which necessitates identifying the operating mode—either stable or transient—to determine the weighting coefficients. For example, multiple sets of weighting coefficients can be designed, such as at least two sets. Figure 2 As shown, when the system is in steady state, the weighting coefficients (α, β, γ) are set to (0.6, 0.25, 0.15); when the system is in transient state, the weighting coefficients (α, β, γ) are set to (0.2, 0.1, 0.7). The specific values ​​can be adjusted according to actual needs.

[0060] In another embodiment, the calculation process can be further constrained by conditions such as ensuring that temperature, humidity, pressure differential, and particulate matter concentration do not exceed preset ranges. This differs from the weighting coefficients mentioned above; these constraints are hard constraints. This approach is more rigorous and secure. It first ensures all the most basic environmental quality and safety requirements (hard constraints), and then, while meeting these requirements, seeks the optimal balance between system energy consumption, filter wear, and personnel exposure risk. This ensures that the system never sacrifices necessary cleanliness or environmental stability for energy conservation.

[0061] S4: Issue the first control command of the optimal control sequence to the corresponding actuator, and repeat the above steps periodically to achieve closed-loop optimization control of the clean air conditioning system.

[0062] Suppose that the optimal control sequence obtained by solving S3 represents the fan operating frequency setpoints every five minutes for the next hour, at 80%, 82%, 81%, etc. The system only outputs the first setpoint, 80%, to the fan's frequency converter via digital or analog signal, and simultaneously sends the corresponding valve opening command to the fresh air valve actuator. After five minutes of executing this command, the system does not execute the second command in the sequence, but returns to S1 to re-collect the latest full data, and then completely executes the prediction of S2 and the optimization solution of S3 again to obtain a new optimal control sequence calculated based on the latest information. Similarly, only the first control command of this new sequence is executed. This process repeats cyclically every five minutes, forming a rolling optimization, ensuring that control decisions are always based on the latest field conditions, thereby providing a timely and optimal response to various disturbances and changes.

[0063] Based on the example above, a sequence contains fifteen instruction pairs, each specifying the precise operating frequency of the fan and the exact opening degree of the fresh air valve for each minute from 10:00 to 10:14. To find this optimal sequence, the system employs efficient nonlinear programming algorithms, such as sequential quadratic programming, to solve for the minimum value of the aforementioned objective function J. Figure 3 As shown, after the calculation is complete, the system does not execute all fifteen instructions, but only the first instruction, which is the setting for the current time k, i.e., the minute from 10:00 to 10:01. Then, at the next time k+1, i.e., at 10:01, the system collects the latest data, re-performs prediction and optimization calculations, obtains a new fifteen-minute control sequence, and again executes only the first instruction. This continuously rolling decision-making process allows the system to continuously adjust its control behavior based on the latest situation, thereby making a fast and accurate response to various disturbances.

[0064] This invention also relates to an energy-saving operation control system for a cleanroom air conditioning system, including a processor and a memory. The memory stores a computer program, and the processor can interact with the memory and call the computer program (e.g., via a bus). The processor then executes the computer program, and when the computer program is executed by the processor, it implements the energy-saving operation control method for the cleanroom air conditioning system described in the above embodiments. Figure 1 As shown, the system also needs to include an actuator and a data acquisition mechanism. The actuator includes at least the aforementioned fan and fresh air valve. The data acquisition mechanism has been specifically described in step S1 above and will not be repeated here.

[0065] Those skilled in the art will conceive of many modifications, alterations, and alternatives without departing from the spirit and essence of this invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. An energy-saving operation control method for a clean air conditioning system, characterized in that, Includes the following steps: S1: Obtain the preset parameters and real-time operating data of the clean air conditioning system; the real-time operating data includes outdoor environmental parameters, temperature and humidity data of various indoor areas, pressure difference data, particulate matter concentration data, airflow characteristic data, and personnel location and activity information; S2: Based on the real-time operating data, identify whether the system is currently in a steady state or a transient operating mode, and use a dynamic prediction model that is continuously corrected based on real-time data to predict the heat and humidity load, particulate matter concentration distribution and personnel exposure risk of the clean area in the future time domain. S3: Establish an objective function that includes system energy consumption, filter loss, and personnel exposure risk, including: J=αE norm +βF norm +γR norm The system energy consumption E, filter loss F, and personnel exposure risk R in the prediction time domain are normalized to obtain the dimensionless E. norm F norm and R norm α, β, and γ are the weight coefficients of system energy consumption E, filter loss F, and personnel exposure risk R, respectively; based on the operating mode, the optimal control sequence that minimizes the objective function value in the future prediction time domain is solved; The system energy consumption E satisfies: P fan (t) represents the instantaneous power of the wind turbine at time t, P pump (t) represents the instantaneous power of the cooling water pump at time t; The filter loss F satisfies: or T p Let Q(t) be the predicted air volume, and C be the length of the prediction time domain. in (t) represents the particulate matter concentration upstream of the filter, and η represents the filter efficiency; k f The filter cost factor is the depreciation cost of the filter for every gram of particulate matter intercepted. The personnel exposure risk R satisfies: N is the total number of people in the clean area, x i (t) represents the predicted spatial location of the i-th person at time t, C(x) i (t),t) represents the predicted position at time t and location x. i The particulate matter concentration (t) represents the risk of human exposure R for all individuals in the prediction time domain T. p The cumulative exposure dose within; S4: Issue the first control command of the optimal control sequence to the corresponding actuator, and repeat the above steps periodically to achieve closed-loop optimization control of the clean air conditioning system.

2. The method according to claim 1, characterized in that, The personnel activity information is quantified into three levels: stationary, slow-moving, and fast-moving.

3. The method according to claim 1, characterized in that, The preset parameters include: cleanliness level, upper and lower limits of temperature and humidity, and minimum pressure difference between areas. The upper and lower limits of temperature and humidity include the upper limit of temperature, the lower limit of temperature, the upper limit of humidity, and the lower limit of humidity.

4. The method according to claim 1, characterized in that, The identification system is currently in a steady-state or transient operating mode, including: calculating the average rate of change of particulate matter concentration data, the average rate of change of temperature, and the average rate of change of humidity over a set period of time; when the average rate of change of particulate matter concentration is higher than a threshold for particulate matter concentration change, the average rate of change of temperature is higher than a threshold for temperature change, or the average rate of change of humidity is higher than a threshold for humidity change, the system is determined to be in a transient state; otherwise, the system is determined to be in a steady state.

5. The method according to claim 1, characterized in that, The dynamic prediction model uses an LSTM model based on a long short-term memory network.

6. The method according to claim 1, characterized in that, Finding the optimal control sequence that minimizes the objective function value within a future prediction time domain includes: setting the prediction time domain to 60 minutes, the control time domain to 15 minutes, and the control step size to 1 minute; at the beginning of each control cycle, using sequential quadratic programming or interior point method to obtain an optimal control sequence consisting of 15 control commands, specifically the fan operating frequency and fresh air valve opening degree per minute within the next 15 minutes.

7. The method according to claim 1, characterized in that, Finding the optimal control sequence that minimizes the objective function value in the next prediction time domain also includes the following constraints: Temperature, humidity, pressure difference, and particulate matter concentration all remain within preset ranges.

8. An energy-saving operation control system for a clean air conditioning system, characterized in that, It includes a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the energy-saving operation control method for the clean air conditioning system as described in any one of claims 1 to 7.

Images