Building ventilation regulation and control method and equipment based on multi-target space-time analysis

Abstract

The invention relates to the technical field of intelligent building environment control and energy management, and provides a building ventilation regulation and control method and equipment based on multi-target space-time analysis in order to solve the problems of high energy consumption, uncontrollable infection risk and response lag in the prior art. Original data and the spatial-temporal features are combined into a multi-modal fusion data set, an air quality change rule is learned based on an LSTM model, a future air quality state matrix is output, a multi-target dynamic optimization function is constructed, an optimal regulation and control strategy of a fan, fresh air and purification equipment is generated, and the optimal regulation and control strategy of the fan, the fresh air and the purification equipment is obtained. The response speed is improved while energy consumption is reduced and infection risks are controlled, and the system is particularly suitable for public places such as hospitals, schools and nursing homes with high requirements for indoor gas environments.

Description

Technical Field

[0001] This invention relates to the field of intelligent building environmental control and energy management technology, and in particular to a building ventilation control method based on multi-objective spatiotemporal analysis. Background Technology

[0002] In the field of building environmental control and energy management, a scientific balance between airtightness and ventilation performance is crucial. This contradiction is particularly pronounced in public places such as hospitals, schools, and nursing homes. Insufficient ventilation leads to the accumulation of pollutants such as CO2, pathogens, and VOCs, increasing the risk of infection, threatening human health, and potentially causing mold growth and structural corrosion. Conversely, excessive ventilation results in energy waste. Therefore, it is essential to accurately monitor and identify functional problems and health defects in the ventilation system to provide a basis for dynamic indoor air quality control, ensuring the health of indoor occupants and the energy efficiency and sustainable operation of buildings.

[0003] There are three existing technologies for intelligent control of building ventilation: The first is a control method based on traditional fixed thresholds. For example, Chinese invention patent CN201810651064.6, "An Indoor Air Quality Control Method and System," discloses an indoor air quality control method and system that can monitor indoor air parameters in real time, compare them with preset thresholds, and automatically activate purification equipment when standards are exceeded. The second is a dynamic control method based on sensor feedback. For example, Chinese invention patent CN120252150A, "An Intelligent Control System and Method for Air Conditioning and Fresh Air Based on Multi-Parameter Linkage," discloses an intelligent control system and method based on sensor feedback, which achieves comprehensive optimization control of the fresh air system according to changes in indoor environmental parameters. The third is an indoor air quality prediction and control method based on machine learning, which has been a research hotspot in recent years. For example, invention patent CN119146565A, "A Multi-Objective Intelligent Optimization Air Conditioning Control Method and System Based on Multi-Dimensional Machine Learning," discloses a multi-objective optimization control algorithm that mainly predicts and dynamically adjusts parameters such as temperature, humidity, and wind speed for air conditioning operation.

[0004] While existing technologies have achieved dynamic control of building ventilation to some extent, they still have two main problems: First, in terms of energy consumption, traditional methods often use fixed thresholds or simple feedback control, which cannot dynamically adapt to changes in population density and differences in building airtightness. This leads to over-ventilation at low loads or under-ventilation at high loads, resulting in energy waste. Furthermore, the lack of precise location of air infiltration points results in persistently high energy consumption for fresh air systems. Second, in terms of health monitoring, existing technologies have limited monitoring dimensions, primarily focusing on conventional environmental parameters such as CO2 and PM2.5, neglecting consideration of parameters related to human activity. Control strategies rely on fixed thresholds or rule triggers, failing to dynamically assess health risks in conjunction with human activity parameters, resulting in response lags and potentially exacerbating infection risks.

[0005] Therefore, developing a dynamic control method for indoor air quality that aims to reduce energy consumption, control infection risks, and provide dynamic response has become an urgent need in this field. Summary of the Invention

[0006] To address the problems of high energy consumption, uncontrollable infection risk, and slow response in existing technologies, this invention proposes a building ventilation control method and equipment based on multi-objective spatiotemporal analysis. By collecting comprehensive data on environmental, personnel, and equipment energy consumption, dynamic prediction is achieved based on an LSTM model. A multi-objective dynamic optimization function is constructed to generate optimal control strategies for fans, fresh air systems, and purification equipment, thereby reducing energy consumption, dynamically controlling infection risk, and improving response speed.

[0007] To achieve the above objectives, the present invention provides a building ventilation control method based on multi-objective spatiotemporal analysis, comprising the following steps: S1: Real-time collection of environmental parameters, personnel activity parameters, and equipment energy consumption data within the building as raw data; cleaning, smoothing, and normalization of the raw data, and calculation of spatiotemporal characteristics including the current infection risk coefficient and personnel spatial density; S2: Merge the original data with spatiotemporal features into a multimodal fusion dataset, and divide the multimodal fusion dataset into a training set and a validation set according to a certain ratio; S3: Extract raw data and spatiotemporal features from the training set to train the LSTM model to learn the air quality change pattern, predict and output the future air quality state matrix; extract raw data and spatiotemporal features from the validation set that have the same data dimension as the training set but are time-independent to evaluate the model performance. S4: Based on the future air quality state matrix output by the LSTM model, construct a multi-objective dynamic optimization function to generate the optimal control strategy for fans, fresh air and purification equipment; S5: Controls fan speed, fresh air volume and purification equipment in real time according to the optimal control strategy, and triggers graded responses.

[0008] Furthermore, the environmental parameters within the building include the current indoor CO2 concentration and PM2.5 concentration. 2.5 The data collected includes concentration, temperature, and humidity; the personnel activity parameter is the real-time number of people within a designated area; the external meteorological data includes outdoor temperature, outdoor humidity, and outdoor wind speed; and the equipment energy consumption data includes the power of the air conditioning or fresh air system and the power consumption of the high-efficiency filter. This invention overcomes the limitations of existing technologies that rely solely on a single dimension like CO2 or PM2.5 by collecting environmental parameters, personnel activity, and equipment energy consumption data in real time, and by introducing infection risk coefficients and personnel spatial density as spatiotemporal characteristics. This provides a data foundation for subsequent prediction and control of future air quality.

[0009] Furthermore, the cleaning process removes outliers and fills in missing values ​​using linear interpolation; the smoothing process applies a moving average to high-frequency noise data within a uniform time window. This avoids erroneous adjustments triggered by instantaneous outliers or high-frequency noise, prevents LSTM model prediction distortion due to data gaps, and ensures the continuity of the infection risk coefficient.

[0010] Furthermore, the normalization process involves resampling and interpolation to align the sampling frequencies of all original data in time. This eliminates timing misalignments caused by different sensor models or network latency, ensuring sufficient data base for subsequent LSTM models and guaranteeing the integrity and temporal consistency of the model input vectors.

[0011] Specifically, the normalization process involves the following steps: A1: Set a uniform sampling frequency, generate time series with equal intervals, and use them as the target time for resampling; A2: Perform timestamp alignment on six types of raw data: CO2 concentration, PM2.5 concentration, temperature and humidity, number of people, external meteorological data, and equipment energy consumption; map the raw data to the nearest grid point: if a grid point already has an original value, retain it directly; if the grid point does not have an original value, proceed to the next step; A3: Select the completion strategy based on the duration of the missing information: If the duration of the missing value before and after the grid point is ≤ 5 times the uniform sampling frequency, then linear interpolation is used to fill in the missing value. If the time interval before and after the missing grid point is greater than 5 times the uniform sampling frequency and less than or equal to 15 times the uniform sampling frequency, then take 5 times the uniform sampling frequency before and after the missing grid point as the time window for mean filling. If the duration of the missing data before and after a grid point is greater than 15 times the uniform sampling frequency, it is marked as missing, and the data during the missing period will not be used for model training and prediction.

[0012] A4: Outputs equally spaced, continuous vectors with uniform dimensions.

[0013] Furthermore, the formula for calculating the current infection risk coefficient is as follows:

[0014]

[0015] Where DIRI(t) represents the current infection risk coefficient. The pollutant exposure coefficient, This indicates the personnel space density correction term. Let represent the environmental adaptive term, and further, wi(t) represent the hazard weights of various pollutants. λi represents the ratio of the current pollutant concentration to the national standard limit, λi represents the pollutant-specific cumulative coefficient, and Ti(t) represents the cumulative exposure time. The formula represents the cumulative toxicity coefficient of long-term exposure to pollutants; α represents the density influence factor; D(t)² represents the square of the real-time spatial density of people; vout(t) represents the outdoor wind speed; vcrit represents the critical wind speed; ΔH(t) represents the indoor-outdoor humidity difference; and Hstd is the standard humidity. This infection risk coefficient calculation formula uses exponential cumulative toxicity to reflect the hazards of long-term exposure, uses squared density correction to enhance the risk in high-density scenarios, and dynamically adjusts the environmental adaptive term through wind speed and humidity, thus simultaneously considering the sensitivity and scientific rigor of infection risk early warning in complex and ever-changing building environments. The design of this formula references the FMEA (Failure Mode and Effects Analysis) methodology widely used in hospital infection risk assessment. Its core idea is to quantify and weight the three dimensions of risk severity, probability of occurrence, and detectability to form a risk priority coefficient for graded response. In this invention, this method is applied to the indoor air quality scenario of buildings, constructing a spatiotemporal coupled model through pollutant exposure, spatial density of people, and environmental adaptive factors, achieving an improvement from static indicators to dynamic risk prediction.

[0016] Furthermore, the formula for calculating the spatial density D(t) of the personnel is as follows:

[0017]

[0018] Where N(t) represents the number of people in the area as monitored in real time, and A represents the effective area of ​​the area. Real-time calculation of personnel space density can accurately reflect the dynamic load relationship between people and space, providing a key correction factor for calculating the infection risk coefficient. It directly drives the adjustment of fresh air volume and purification strategy as needed, avoiding over-ventilation when there are few people or insufficient ventilation when there are many people, and simultaneously reducing energy consumption and health risks.

[0019] Furthermore, the multimodal fusion dataset consists of raw data and spatiotemporal features. The raw data consists of real-time collected environmental parameters, personnel activity parameters, and equipment energy consumption data within the building. The spatiotemporal features are calculated by cleaning, smoothing, and normalizing the raw data, and then calculating features that include the current infection risk coefficient and personnel spatial density.

[0020] Furthermore, the training set and validation set are two datasets with identical structures and independent time series. The identical structure means the validation set contains the same number of feature variables, time windows, and data format as the training set, and is standardized using the same scaling parameters as the training set. Furthermore, the ratio of the training set to the validation set is 8:2. Specifically, the first 80% of the data within a continuous time period is selected as the training set, and the last 20% of the data within the same continuous time period is selected as the validation set. These measures ensure that the model has sufficient training data to fully learn the patterns of air quality changes, and sufficient validation data to test the model's accuracy, thereby enabling precise parameter tuning and improving the predictive reliability of the system in actual operation.

[0021] Furthermore, the future air quality status includes: the future infection risk coefficient, the future total power consumption of equipment, and the future indoor temperature.

[0022] Furthermore, the formula for calculating the future air quality state matrix using the LSTM model is as follows:

[0023]

[0024] The output of this matrix is:

[0025]

[0026] Where DIRI'(t) represents the predicted infection risk coefficient at time t in the future, P'(t) represents the predicted total power consumption of the device at time t in the future, and T in '(t) represents the predicted indoor temperature at time t in the future; meanwhile, x t This represents the input vector at time t, composed of multimodal fused data from the training set, including CO2 concentration and PM2.5 concentration. 2.5 Concentration, indoor temperature, indoor humidity, space density, current infection risk factor, and total power consumption of the equipment; h t Represents the hidden state at time t; h t-1 represents the hidden state at the previous time step; b represents the bias vector; and W represents the weight matrix.

[0027] The purpose of this method is to use LSTM to output three-dimensional prediction data of infection risk, energy consumption, and temperature at one time, and to integrate the traditionally separate health, energy consumption, and comfort goals into a unified indicator that can be predicted in advance. Based on this, the optimal control strategy for fans, fresh air and purification equipment can be generated and controlled, which can avoid the energy waste of post-event remediation and significantly reduce the fluctuation of comfort caused by sudden risks.

[0028] Furthermore, the calculation formula for the multi-objective dynamic optimization function is as follows:

[0029]

[0030] Where u(t) represents the control variables, including fan speed, fresh air volume, and the start / stop status of the purification equipment; min represents the control variable that minimizes the weighted sum within the square brackets; DIRI'(t) represents the infection risk coefficient at time t in the future; P'(t) represents the total power consumption of the equipment at time t in the future; |T in '(t)-T std | represents the indoor temperature T at time t in the future. in '(t) and comfort reference temperature T std The deviation values ​​are α(t), β(t), and γ(t), which represent the weights of health risk, energy consumption, and comfort, respectively. By constructing a dynamic optimization function with "health risk, energy consumption, and comfort" as the objectives, a balance among the three and real-time optimal solution are achieved, avoiding the extreme strategies of prioritizing energy saving or health in existing technologies, and minimizing energy consumption while ensuring comfort.

[0031] Furthermore, the optimal control strategy is as follows: If the infection risk coefficient is less than 1 and remains so for 30 minutes, it is considered low risk, and the fresh air volume is adjusted to 30% of the maximum volume. When 1 ≤ infection risk coefficient ≤ 2 and the personnel density in the space > 0.2 people / m², it is judged as medium risk. The fresh air volume is adjusted to 70% of the maximum air volume and the high-efficiency filter is turned on. When the infection risk coefficient is >2 or PM2.5 is >150μg / m³, it is judged as high risk. The fresh air volume is adjusted to the maximum air volume and the high-efficiency filter is turned on. At the same time, the presence of personnel is checked. If no one is present, the ultraviolet lamp is turned on. If personnel are present, the lamp is turned on later and the event is recorded.

[0032] The system automatically triggers a graded response mechanism based on the prediction results, enabling coordinated operation of equipment such as fans, fresh air systems, purifiers, and ultraviolet disinfection. This not only improves the system's response speed but also further protects human health and safety and reduces ineffective energy consumption through logic such as "ultraviolet disinfection when no one is present."

[0033] The present invention also provides an electronic device for intelligent control of building ventilation based on multi-objective spatiotemporal analysis, comprising: A processing unit, wherein the processing unit is used to implement each step of the intelligent control method for building ventilation based on multi-objective spatiotemporal analysis described above; A readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the various steps of the intelligent control method for building ventilation based on multi-objective spatiotemporal analysis described above.

[0034] Beneficial effects: As can be seen from the above technical solutions, this invention addresses the problems of high energy consumption, uncontrollable infection risk, and slow response in existing technologies. It proposes a building ventilation control method and equipment based on multi-objective spatiotemporal analysis. By collecting comprehensive data on environmental, personnel, and equipment energy consumption, dynamic prediction is achieved based on an LSTM model. A multi-objective dynamic optimization function is constructed to generate the optimal control strategies for fans, fresh air systems, and purification equipment. This reduces energy consumption, dynamically controls infection risk, and improves response speed. It is particularly suitable for public places with high requirements for indoor gas environment, such as hospitals, schools, and nursing homes.

[0035] The technical effects brought about by the technical solution described in this invention are as follows:

[0036] 1. Regarding energy consumption reduction, a dynamic optimization function targeting "health risk, energy consumption, and comfort" is constructed, achieving a balance and real-time optimal solution among the three, avoiding the extreme strategies of prioritizing energy saving or health in existing technologies. An automatic tiered response mechanism is triggered based on prediction results, enabling coordinated operation of equipment such as fans, fresh air systems, purifiers, and UV disinfection. This not only improves system response speed and execution accuracy but also further reduces ineffective energy consumption through logic such as "UV disinfection when unattended," thereby enhancing overall operational efficiency.

[0037] 2. In terms of controlling infection risk, this invention quantifies pollutant exposure, personnel space density, and environmental adaptive factors into dynamic infection risk coefficients in real time, realizing coupled monitoring of space and time, and breaking through the problem of existing technologies that only rely on a single dimension of CO2 or PM2.5; at the same time, it uses a graded response strategy to map risk values ​​to actions such as ventilation, filtration, and ultraviolet disinfection, which significantly improves the accuracy and response sensitivity of infection risk control.

[0038] 3. In terms of response speed, this invention learns the air quality change pattern through the LSTM model, predicts and outputs future infection risk, equipment power consumption and indoor temperature, realizing the transformation from "post-event response" to "pre-event intervention", breaking through the limitations of existing technology control strategies that rely on fixed thresholds or rule triggers, realizing a more proactive building ventilation control response, and significantly reducing energy waste and comfort fluctuations caused by delayed control.

[0039] It should be understood that all combinations of the foregoing concepts and the additional concepts described in more detail below can be considered part of the inventive subject matter of this disclosure, provided that such concepts do not contradict each other.

[0040] The foregoing and other aspects, embodiments, and features of the teachings of the present invention will be more fully understood from the following description in conjunction with the accompanying drawings. Other additional aspects of the invention, such as features and / or beneficial effects of exemplary embodiments, will become apparent from the following description or may be learned through practice of specific embodiments according to the teachings of the present invention. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0042] Figure 1 This is a flowchart of a building ventilation control method based on multi-objective spatiotemporal analysis disclosed in an embodiment of this application;

[0043] Figure 2 This is a flowchart of the normalization processing steps disclosed in the embodiments of this application;

[0044] Figure 3 This is a structural block diagram of the electronic device described in the embodiments of this application. Detailed Implementation

[0045] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0046] Before introducing this application, the relevant technologies of this application will be introduced first.

[0047] The infection risk assessment methodology, namely Failure Mode and Effects Analysis (FMEA), is a systematic risk assessment tool. Its core idea is to quantify and weight the severity, probability, and detectability of risk to form a risk priority coefficient, which is used to identify and control infection transmission routes in advance. This application draws on the infection risk assessment methodology to design an infection risk coefficient calculation method, quantifying factors such as pollutant exposure, personnel density, and environmental adaptation into dynamic risk coefficients. This elevates risk identification from a static threshold to a spatiotemporally coupled assessment, providing data support for subsequent LSTM models and multi-objective dynamic optimization functions.

[0048] LSTM models, or Long Short-Term Memory networks, are an improved type of recurrent neural network that selectively remembers or forgets historical information through gating mechanisms, making them adept at capturing long-term dependencies in time series data. In this application, the LSTM model is used to predict future air quality, including infection risk factors, total device power consumption, and indoor temperature.

[0049] The multi-objective dynamic optimization function is a mathematical model that weighs multiple conflicting objectives in real time. Through weight adjustment and constraint solving, it dynamically generates the optimal decision scheme. In this application, it is used to achieve a balance between energy consumption, health risks, and comfort operation of the ventilation system in different scenarios.

[0050] In the field of intelligent building environmental control, existing technologies for building ventilation control suffer from high energy consumption, uncontrollable infection risks, and slow response times. Therefore, this embodiment aims to propose a building ventilation control method and device based on multi-objective spatiotemporal analysis to simultaneously address the aforementioned technical problems.

[0051] The following description, in conjunction with the accompanying drawings, details a building ventilation control method and device based on multi-objective spatiotemporal analysis as described in this embodiment.

[0052] Combination Figure 1 As shown, the building ventilation control method based on multi-objective spatiotemporal analysis disclosed in this embodiment includes the following steps: S1: Real-time collection of environmental parameters, personnel activity parameters, and equipment energy consumption data within the building as raw data; cleaning, smoothing, and normalization of the raw data, and calculation of spatiotemporal characteristics including the current infection risk coefficient and personnel spatial density; S2: Merge the original data with spatiotemporal features into a multimodal fusion dataset, and divide the multimodal fusion dataset into a training set and a validation set according to a certain ratio; S3: Extract raw data and spatiotemporal features from the training set to train the LSTM model to learn the air quality change pattern, predict and output the future air quality state matrix; extract raw data and spatiotemporal features from the validation set that have the same data dimension as the training set but are time-independent to evaluate the model performance. S4: Based on the future air quality state matrix output by the LSTM model, construct a multi-objective dynamic optimization function to generate the optimal control strategy for fans, fresh air and purification equipment; S5: Controls fan speed, fresh air volume and purification equipment in real time according to the optimal control strategy, and triggers graded responses.

[0053] Specifically, in S1, this invention collects environmental parameters, personnel activities, and equipment energy consumption data in real time, and introduces infection risk coefficient and personnel spatial density as spatiotemporal characteristics. This breaks through the problem of existing technologies relying solely on a single dimension of CO2 or PM2.5, and provides a data foundation for subsequent prediction and control of future air quality.

[0054] Furthermore, taking a hospital as an example, the selection and deployment of sensors in this embodiment will be explained below:

[0055] The environmental parameters within the building include the current indoor CO2 concentration and PM2.5 concentration. 2.5 Concentration, temperature, and humidity; this embodiment uses a Honeywell HAQ-6 series integrated sensor module to simultaneously detect CO2 concentration, PM2.5, temperature, and humidity data. It is deployed at a height of 1.2m-1.5m on the wall of activity spaces such as wards, clinics, and lobbies, embedded in a wall junction box, at the same height as the human breathing zone, so that CO2 concentration and PM2.5 can be detected simultaneously. 2.5 Concentration readings are representative.

[0056] The personnel activity parameter is the real-time number of people in the set area; the number of people is monitored by an active infrared counting sensor. In this embodiment, the Hikvision DS-2CD6 series sensor is selected and installed on the wall or the top of the entrance / exit of the set area. It is detected by bidirectional infrared beams and does not rely on facial recognition, thus better protecting personnel privacy.

[0057] The external meteorological data includes outdoor temperature, outdoor humidity, and outdoor wind speed; this data is obtained through API integration with the meteorological bureau. Compared to hardware such as anemometers, this method incurs zero hardware cost and low maintenance costs.

[0058] The energy consumption data of the equipment includes the power of the air conditioning or fresh air system and the power consumption of the high-efficiency filter. The energy consumption data of the equipment is obtained by connecting to the smart meter in the building. The smart meter measures the three-phase power through a current transformer and transmits the power consumption signal using the Modbus protocol.

[0059] Furthermore, the cleaning process employs the 3σ criterion to remove outliers and fills in missing values ​​using linear interpolation.

[0060] Furthermore, the smoothing process applies a moving average to the high-frequency noise data in the data using a uniform time window; the uniform time window is 5-10 minutes. This avoids erroneous adjustments triggered by instantaneous outliers or high-frequency noise, prevents LSTM model prediction distortion due to data gaps, and ensures the continuity of the infection risk coefficient.

[0061] Furthermore, the normalization process involves resampling and interpolation to process the original data, aligning the sampling frequencies of all original data in time.

[0062] Specifically, such as Figure 2 As shown, the normalization process involves the following steps: A1: Set a uniform sampling frequency; generate time series with equal intervals and use them as the target time for resampling; A2: Perform timestamp alignment on six types of raw data: CO2 concentration, PM2.5 concentration, temperature and humidity, number of people, external meteorological data, and equipment energy consumption; map the raw data to the nearest grid point: if a grid point already has an original value, retain it directly; if the grid point does not have an original value, proceed to the next step; A3: Select the completion strategy based on the duration of the missing information: If the duration of the missing value before and after the grid point is ≤ 5 times the uniform sampling frequency, then linear interpolation is used to fill in the missing value. If the time interval before and after the missing grid point is greater than 5 times the uniform sampling frequency and less than or equal to 15 times the uniform sampling frequency, then take 5 times the uniform sampling frequency before and after the missing grid point as the time window for mean filling. If the duration of the missing data before and after a grid point is greater than 15 times the uniform sampling frequency, it is marked as missing, and the data during the missing period will not be used for model training and prediction.

[0063] A4: Outputs equally spaced, continuous vectors with uniform dimensions.

[0064] Optionally, the uniform sampling frequency in A1 is 1-5 minutes.

[0065] In practical implementation, the sampling frequency for raw environmental parameters, including CO2 concentration, PM2.5, temperature, and humidity, is 1 minute / time; the sampling frequency for real-time personnel numbers is 5 seconds / time; the sampling frequency for external meteorological data, including outdoor temperature, outdoor humidity, and outdoor wind speed, is 5 minutes / time; and the sampling frequency for equipment energy consumption data is 30 seconds / time. This embodiment sets a uniform sampling frequency of 1 minute / time. For the six types of raw data—CO2 concentration, PM2.5 concentration, temperature and humidity, personnel numbers, external meteorological data, and equipment energy consumption—timestamp alignment is performed, missing durations are padded, and the output is a continuous vector with a sampling interval of 1 minute / time, which serves as the input vector for the subsequent LSTM model. Normalization processing eliminates data timing misalignment caused by different sensor models or network latency, ensuring a sufficient data base for the subsequent LSTM model and guaranteeing the integrity and timing consistency of the model input vector.

[0066] Furthermore, the formula for calculating the current infection risk coefficient is as follows:

[0067]

[0068] DIRI(t) represents the current infection risk coefficient, which is a relative risk value calculated by multiplying multiple unitless coefficients. It is similar to the common air quality index or ultraviolet index, and only indicates the level of risk without any unit.

[0069] Furthermore, The pollutant exposure coefficient is unitless; where wi(t) represents the hazard weight of various pollutants, also unitless. The value represents the ratio of the current pollutant concentration to the national standard limit, expressed in ppm / ppm or μg / m³ / μg / m³, with units offsetting each other; λi represents the pollutant-specific cumulative coefficient, expressed in minutes⁻¹; Ti(t) represents the cumulative exposure time, expressed in minutes. This indicates the cumulative toxicity coefficient of pollutants after long-term exposure, and has no unit.

[0070] Furthermore, The term represents the spatial density correction term for people, without units; where α represents the density influence factor, with units of (m² / person)², and D(t)² represents the square of the real-time spatial density of people, with units of (person / m²)².

[0071] Furthermore, The formula represents the environmental adaptive term, which is unitless; where vout(t) represents the outdoor wind speed in m / s, vcrit represents the critical wind speed in m / s; ΔH(t) represents the indoor-outdoor humidity difference in RH%, and Hstd represents the standard humidity in RH%. The design of this formula references the Failure Mode and Effects Analysis (FMEA) methodology widely used in hospital infection risk assessment. Its core idea is to quantify and weight the three dimensions of risk severity (S), probability of occurrence (P), and detectability (D) to form a risk priority coefficient for graded response. In this invention, this method is applied to the indoor air quality scenario of buildings. By constructing a spatiotemporal coupled model using pollutant exposure, personnel space density, and environmental adaptive factors, it achieves an improvement from static indicators to dynamic risk prediction.

[0072] Furthermore, the formula for calculating the spatial density D(t) of the personnel is as follows:

[0073]

[0074] The unit of D(t) is person / m 2 Where N(t) is the number of people in the area monitored in real time, in person; A is the effective area of ​​the area, in meters. 2Real-time calculation of personnel space density can accurately reflect the dynamic load relationship between people and space, providing a key correction factor for calculating the infection risk coefficient. It directly drives the adjustment of fresh air volume and purification strategy as needed, avoiding over-ventilation when there are few people or insufficient ventilation when there are many people, and simultaneously reducing energy consumption and health risks.

[0075] Furthermore, taking hospital wards as an example, referring to GB / T 18883-2022 "Indoor Air Quality Standard", the personnel and environmental parameters are as follows:

[0076] The current effective area of ​​the ward is A=30m² 2 ;

[0077] Indoor pollutant concentration at current time t: CO2 concentration C1(t) = 800 ppm, national standard limit C 1,safe =1000 ppm; PM2.5 concentration C2(t) = 50 μg / m3, national standard limit C 2,safe =75 μg / m3;

[0078] Pollutant hazard weights: CO2 weight w1(t) = 0.6, PM2.5 weight w2(t) = 0.4;

[0079] Cumulative exposure time: T1(t) = 2 hours, T2(t) = 1.5 hours;

[0080] Pollutant-specific cumulative coefficients: λ1=0.1, λ2=0.15;

[0081] The real-time number of personnel is N(t) = 6.

[0082] Density influence factor α = 0.05;

[0083] Outdoor wind speed v out (t) = 3 m / s, critical wind speed v crit =5 m / s;

[0084] Indoor-outdoor humidity difference ΔH(t) = 10%, standard humidity H std =50%.

[0085] Based on the above data, the calculation steps for the current infection risk coefficient DIRI(t) are as follows:

[0086] D1: First, calculate the spatial density of people: D(t) = N(t) / A = 6 / 30 = 0.2 people / m² 2 .

[0087] D2: Then calculate the pollutant exposure factor:

[0088] CO2 exposure items are:

[0089]

[0090] PM2.5 exposure items are:

[0091]

[0092] Total pollutant exposure factor:

[0093]

[0094] D3: Calculate the personnel space density correction term:

[0095]

[0096] D4: Computational Environment Adaptation Items:

[0097]

[0098] D5: Calculate the current infection risk coefficient DIRI(t):

[0099]

[0100] The current infection risk coefficient DIRI(t) is 0.553.

[0101] In the above calculations, this invention uses exponential cumulative toxicity to reflect the long-term exposure hazards, uses squared density correction to enhance the risks in high-density scenarios, and dynamically adjusts the environmental adaptive term through wind speed and humidity, thereby taking into account both the sensitivity and scientific nature of infection risk early warning in complex and ever-changing building environments.

[0102] Furthermore, the multimodal fusion dataset comprises raw data and spatiotemporal features. The raw data consists of real-time collected environmental parameters, personnel activity parameters, and equipment energy consumption data within the building. The spatiotemporal features are calculated by cleaning, smoothing, and normalizing the raw data, and then determining the calculated features that include the current infection risk coefficient and personnel spatial density. The multimodal fusion dataset aligns and integrates real-time environmental, personnel, energy consumption, infection risk, and spatial density spatiotemporal features, providing the model with comprehensive, low-noise, and same-dimensional input, significantly improving prediction accuracy and control response speed.

[0103] Furthermore, the training set and validation set are two datasets with identical structures and independent time series. The identical structure means the validation set contains the same number of feature variables, time windows, and data format as the training set, and is standardized using the same scaling parameters as the training set. Furthermore, the ratio of the training set to the validation set is 8:2. Specifically, the first 80% of the data within a continuous time period is selected as the training set, and the last 20% of the data within the same continuous time period is selected as the validation set. These measures ensure that the model has sufficient training data to fully learn the patterns of air quality changes, and sufficient validation data to test the model's accuracy, thereby enabling precise parameter tuning and improving the predictive reliability of the system in actual operation.

[0104] Furthermore, the future air quality status includes: the future infection risk coefficient, the future total power consumption of equipment, and the future indoor temperature.

[0105] Furthermore, the formula for calculating the future air quality state matrix using the LSTM model is as follows:

[0106]

[0107] The output of this matrix is:

[0108]

[0109] Where DIRI'(t) represents the predicted infection risk coefficient at time t in the future, P'(t) represents the predicted total power consumption of the device at time t in the future, and T in '(t) represents the predicted indoor temperature at time t in the future; meanwhile, x t This represents the input vector at time t, composed of multimodal fused data from the training set, including CO2 concentration and PM2.5 concentration. 2.5 Concentration, indoor temperature, indoor humidity, space density, current infection risk factor, and total power consumption of the equipment; h t Represents the hidden state at time t; h t-1 represents the hidden state at the previous time step; b represents the bias vector; and W represents the weight matrix.

[0110] Furthermore, taking a hospital ward as an example, the following is a 30-minute historical data collected at 09:30:00 on Day 1 in this embodiment, which serves as the input vector x of the LSTM model. t :

[0111] Time CO2 concentration PM2.5 concentration Indoor temperature Indoor humidity Person density DIRI Device power consumption Day 1 09:00:00 650 ppm 20 μg / m3 23.5℃ 38% 0.15 person / m2 0.45 1.23 kW Day 1 09:01:00 655 ppm 22 μg / m3 23.5℃ 42% 0.16 person / m2 0.46 1.28 kW Day 1 09:02:00 660 ppm 23 μg / m3 23.6℃ 46% 0.18 person / m2 0.47 1.35 kW ... ... ... ... ... ... ... ... Day 1 09:29:00 670 ppm 22 μg / m3 23.7℃ 30% 0.17 person / m2 0.49 1.42 kW Day 1 09:30:00 680 ppm 25 μg / m3 23.8℃ 28% 0.18 person / m2 0.50 1.50 kW

[0112] The above input vector x t Converted into a matrix, it includes 1 sample, 30 time steps, and 7 features:

[0113]

[0114] After 30 time steps of sequence computation, LSTM updates the hidden state h through a gating mechanism. t :

[0115]

[0116] The weight matrix W and bias vector b of the model, trained using historical data, are as follows:

[0117] ,

[0118] The model predicts the air quality state matrix for the next 15 minutes as follows:

[0119]

[0120] The result is as follows:

[0121] Predicted time point Infection risk index (DIRI) Total device power consumption (P) Indoor temperature (T) t+15 0.58 1.65 kW 24.1°C

[0122] The purpose of this method is to use LSTM to output three-dimensional prediction data of infection risk, energy consumption, and temperature at one time, and to integrate the traditionally separate health, energy consumption, and comfort goals into a unified indicator that can be predicted in advance. Based on this, the optimal control strategy for fans, fresh air and purification equipment can be generated and controlled, which can avoid the energy waste of post-event remediation and significantly reduce the fluctuation of comfort caused by sudden risks.

[0123] Furthermore, the calculation formula for the multi-objective dynamic optimization function is as follows:

[0124]

[0125] Where u(t) represents the control variables, including fan speed, fresh air volume, and the start / stop status of the purification equipment; min represents the control variable that minimizes the weighted sum within the square brackets; DIRI'(t) represents the infection risk coefficient at time t in the future; P'(t) represents the total power consumption of the equipment at time t in the future; |T in '(t)-T std | represents the indoor temperature T at time t in the future. in '(t) and comfort reference temperature T std The deviation values ​​are α(t), β(t), and γ(t), which represent the weights of health risk, energy consumption, and comfort, respectively.

[0126] Furthermore, taking a hospital ward as an example, the air quality status predicted by the above LSTM model for the next 15 minutes is substituted into the multi-objective optimization function:

[0127]

[0128] In this embodiment, the weights of infection risk DIRI', energy consumption P', and comfort T' are set as α=0.6, β=0.3, and γ=0.1, respectively. The optimization objective is then:

[0129]

[0130] Optionally, in specific implementation, the weighting strategies for infection risk DIRI', energy consumption P', and comfort T' are as follows:

[0131] Time period α β Gamma Strategy tendency Daytime 0.6 0.3 0.1 Infection risk > power consumption > comfort Nighttime 0.3 0.6 0.2 Power consumption > infection risk > comfort Extreme weather 0.5 0.1 0.4 Comfort ≈ infection risk > power consumption

[0132] By constructing a dynamic optimization function with "health risk, energy consumption, and comfort" as the objectives, a balance among the three and real-time optimal solution are achieved, avoiding the extreme strategies of prioritizing energy saving or health in existing technologies, and minimizing energy consumption while ensuring comfort.

[0133] Furthermore, the optimal control strategy is as follows:

[0134] If the infection risk coefficient is less than 1 and remains so for 30 minutes, it is considered low risk, and the fresh air volume is adjusted to 30% of the maximum volume.

[0135] When 1 ≤ infection risk coefficient ≤ 2 and the personnel density in the space > 0.2 people / m², it is judged as medium risk. The fresh air volume is adjusted to 70% of the maximum air volume and the high-efficiency filter is turned on.

[0136] When the infection risk coefficient is >2 or PM2.5 is >150μg / m³, it is judged as high risk. The fresh air volume is adjusted to the maximum air volume and the high-efficiency filter is turned on. At the same time, the presence of personnel is checked. If no one is present, the ultraviolet lamp is turned on. If personnel are present, the lamp is turned on later and the event is recorded.

[0137] Furthermore, the calculation results of the above multi-objective optimization function are substituted into the optimal control strategy:

[0138] Based on the risk level assessment, since DIRI'=0.58<1 and has lasted for 30 minutes, it is determined to be low risk.

[0139] Energy consumption is constrained. Since P'=1.65kW, which is close to the equipment's rated upper limit of 2.0kW, overload must be avoided.

[0140] For comfort compensation, since T'=24.1℃, the deviation from the standard value of 24℃ is only 0.1℃, and the impact is negligible.

[0141] The start-up control actions are as follows: maintain the fresh air volume at the current speed of 30%, keep the HEPA filter on, keep the humidifier on, and keep the UV lamp off.

[0142] This invention automatically triggers a graded response mechanism based on prediction results, enabling coordinated operation of equipment such as fans, fresh air systems, purifiers, and ultraviolet disinfection. This not only improves the system's response speed but also further protects human health and safety and reduces ineffective energy consumption through logic such as "ultraviolet disinfection when no one is present."

[0143] In addition, the present invention also provides an electronic device for intelligent control of building ventilation based on multi-objective spatiotemporal analysis, such as... Figure 3 As shown, it includes:

[0144] A processing unit, wherein the processing unit is used to implement each step of the intelligent control method for building ventilation based on multi-objective spatiotemporal analysis described above;

[0145] A readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the various steps of the intelligent control method for building ventilation based on multi-objective spatiotemporal analysis described above.

[0146] Since the electronic device is built based on the method and is used to implement the method, it can effectively address the problems of high energy consumption, uncontrollable infection risk, and slow response in existing technologies in practical applications. It can achieve dynamic intelligent control of building ventilation, effectively reduce energy consumption, dynamically control infection risk, and improve response speed.

[0147] While the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Those skilled in the art can make various modifications and refinements without departing from the spirit and scope of the invention. Therefore, the scope of protection of the present invention shall be determined by the claims.

Claims

1. A method for intelligent control of building ventilation based on multi-objective spatiotemporal analysis, characterized in that, Includes the following steps: S1: Real-time collection of environmental parameters, personnel activity parameters, and equipment energy consumption data within the building as raw data; cleaning, smoothing, and normalization of the raw data, and calculation of spatiotemporal characteristics including the current infection risk coefficient and personnel spatial density; S2: Merge the original data with spatiotemporal features into a multimodal fusion dataset, and divide the multimodal fusion dataset into a training set and a validation set according to a certain ratio; S3: Extract raw data and spatiotemporal features from the training set to train the LSTM model to learn the air quality change pattern, predict and output the future air quality state matrix; extract raw data and spatiotemporal features from the validation set that have the same data dimension as the training set but are time-independent to evaluate the model performance. S4: Based on the future air quality state matrix output by the LSTM model, construct a multi-objective dynamic optimization function to generate the optimal control strategy for fans, fresh air and purification equipment; S5: Controls fan speed, fresh air volume and purification equipment in real time according to the optimal control strategy, and triggers graded responses.

2. The intelligent control method for building ventilation based on multi-objective spatiotemporal analysis according to claim 1, characterized in that, The normalization process involves resampling and interpolation to process the original data, aligning the sampling frequencies of all original data in time.

3. The intelligent control method for building ventilation based on multi-objective spatiotemporal analysis according to claim 1, characterized in that, The formula for calculating the current infection risk coefficient is as follows: ， Where DIRI(t) represents the current infection risk coefficient. The pollutant exposure coefficient, This indicates the personnel space density correction term. This represents the environment adaptation term, and further w i (t) represents the hazard weight of various pollutants. λ represents the ratio of the current pollutant concentration to the national standard limit. i T represents the pollutant-specific cumulative coefficient. i (t) Cumulative exposure time, α represents the cumulative toxicity coefficient of pollutants after long-term exposure; α represents the density influence factor, D(t). 2 v represents the square of the real-time spatial density of people; out (t) represents the outdoor wind speed, v crit H represents the critical wind speed, ΔH(t) represents the indoor-outdoor humidity difference, and H represents the critical wind speed. std This is the standard humidity.

4. The intelligent control method for building ventilation based on multi-objective spatiotemporal analysis according to claim 2, characterized in that, The formula for calculating the spatial density D(t) of the personnel is as follows: ， Where N(t) is the number of people in the area monitored in real time, and A is the effective area of ​​the area.

5. The intelligent control method for building ventilation based on multi-objective spatiotemporal analysis according to claim 1, characterized in that, The training set and the validation set are two datasets with the same structure and independent time series; the same structure means that the validation set contains the same number of feature variables, time window and data format as the training set, and the validation set is standardized using the same scaling parameters as the training set.

6. The intelligent control method for building ventilation based on multi-objective spatiotemporal analysis according to claim 1, characterized in that, The formula for calculating the future air quality state matrix using the LSTM model is as follows: ， The output of this matrix is: ， Where DIRI'(t) represents the predicted infection risk coefficient at time t in the future, P'(t) represents the predicted total power consumption of the device at time t in the future, and T in '(t) represents the predicted indoor temperature at time t in the future; meanwhile, x t This represents the input vector at time t, composed of multimodal fused data from the training set, including CO2 concentration and PM2.5 concentration. 2.5 Concentration, indoor temperature, indoor humidity, space density, current infection risk factor, and total power consumption of the equipment; h t Represents the hidden state at time t; h t-1 represents the hidden state at the previous time step; b represents the bias vector; and W represents the weight matrix.

7. The intelligent control method for building ventilation based on multi-objective spatiotemporal analysis according to claim 1, characterized in that, The calculation formula for the multi-objective dynamic optimization function is as follows: ， Where u(t) represents the control variables, including fan speed, fresh air volume, and the start / stop status of the purification equipment; min represents the control variable that minimizes the weighted sum within the square brackets; DIRI'(t) represents the infection risk coefficient at time t in the future; P'(t) represents the total power consumption of the equipment at time t in the future; |T in '(t)-T std | represents the indoor temperature T at time t in the future. in '(t) and comfort reference temperature T std The deviation values ​​are α(t), β(t), and γ(t), which represent the weights of health risk, energy consumption, and comfort, respectively.

8. The intelligent control method for building ventilation based on multi-objective spatiotemporal analysis according to claim 1, characterized in that, The optimal control strategy is as follows: If the infection risk coefficient is less than 1 and remains so for 30 minutes, it is considered low risk, and the fresh air volume is adjusted to 30% of the maximum volume. When 1 ≤ infection risk coefficient ≤ 2 and the personnel density in the space > 0.2 people / m², it is judged as medium risk. The fresh air volume is adjusted to 70% of the maximum air volume and the high-efficiency filter is turned on. When the infection risk coefficient is >2 or PM2.5 is >150μg / m³, it is judged as high risk. The fresh air volume is adjusted to the maximum air volume and the high-efficiency filter is turned on. At the same time, the presence of personnel is checked. If no one is present, the ultraviolet lamp is turned on. If personnel are present, the lamp is turned on later and the event is recorded.

9. An electronic device for intelligent control of building ventilation based on multi-objective spatiotemporal analysis, characterized in that, Include: A processing unit, wherein the processing unit is used to implement each step of the intelligent control method for building ventilation based on multi-objective spatiotemporal analysis as described in any one of claims 1 to 8; A readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the various steps of a building ventilation intelligent control method based on multi-objective spatiotemporal analysis as described in any one of claims 1 to 8.

Images