A method for calculating indoor space ventilation times based on pre-trained machine learning models and online indoor air monitoring data

Abstract

The present invention discloses a method for calculating the ventilation times of indoor spaces based on a pre-trained machine learning model and indoor air online monitoring data. In combination with the existing systems and data of the increasingly popular online monitoring of indoor air quality, online evaluation of ventilation efficiency is achieved through a pre-trained machine learning model. The model can not only calculate the ventilation times by utilizing the concentration changes of a certain air indicator monitored online, but also predict the concentration change trend by utilizing the same model, thereby providing a new and efficient artificial intelligence evaluation method for indoor environmental health.

Description

Technical Field

[0001] The present invention relates to the field of ventilation detection technology, and more specifically, to a method for calculating the number of ventilation changes in an indoor space based on a pre-trained machine learning model and indoor air online monitoring data. Background Art

[0002] Indoor ventilation efficiency is an important factor affecting indoor environmental health. The evaluation of indoor ventilation efficiency mainly uses indicators set by building or health industry standards, such as ventilation frequency or fresh air volume. The two indicators of ventilation frequency and fresh air volume are obtained through conversion relationships. The traditional method mainly relies on tracer gas detection experiments to calculate the ventilation frequency, and then calculates the fresh air volume through the ventilation frequency combined with the indoor space volume and the approved number of people.

[0003] The tracer gas method specified in the national standard GB / T18204.18-2000 Public Health Indoor Fresh Air Volume Determination Method is highly scientific and accurate, and can well reflect the actual ventilation effect of indoor space. However, this method has many difficulties in actual use and operation, so it is difficult to be promoted in daily public health inspections. For example:

[0004] 1. The injection and mixing stages of the tracer gas detection experiment require special environmental settings, such as closing doors and windows or closing all the air intake / exhaust ports of the fresh air system. This is actually difficult to operate during the actual detection process in public places and brings great human interference.

[0005] 2. In the actual testing process, the commonly used methods in large public places not only require a large amount of tracer gas, but also have high costs. It is also difficult to release enough tracer gas and mix it in the whole space in a short time (such as within 5 minutes). Moreover, if the injection and / or mixing time is too long, injection and leakage will occur at the same time, which will not only waste tracer gas, but also directly affect the detection accuracy or even invalidate the test.

[0006] 3. The equipment used in ventilation measurement experiments is usually expensive, large in size, and not easy to carry.

[0007] 4. The data obtained from the ventilation measurement experiment is only the ventilation frequency under ideal and fixed variables. It is not real-time and may differ from the actual situation due to human factors.

[0008] In summary, the tracer gas detection experimental method has good scientificity and accuracy, but the detection experiment is time-consuming, and the manpower and material costs are high. It cannot be used as a normalized detection method to cover a large area of ​​various indoor places. And because the detection experiment method is limited to certain specific working conditions (such as the status of doors and windows / fresh air equipment, etc.), the test results only reflect the static evaluation of the specific working conditions of the place. This static evaluation may be very different from the actual working conditions and cannot accurately reflect the actual ventilation efficiency of the space during actual use.

[0009] Therefore, the traditional evaluation of ventilation frequency and fresh air volume only uses very limited experimental test data, which cannot reflect the continuous change process and trend of ventilation efficiency during long-term use of the space, as well as the monitoring of sudden abnormal situations. Summary of the invention

[0010] The present invention provides a method for calculating the ventilation times of indoor spaces based on a pre-trained machine learning model and indoor air online monitoring data, thereby solving the technical problems existing in the traditional tracer gas experimental method.

[0011] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0012] The present invention provides a method for calculating the ventilation times of indoor space based on a pre-trained machine learning model and indoor air online monitoring data, comprising the following steps:

[0013] Collect tracer gas concentration data that varies with time under different spatial structures, different operating conditions, and different experimental conditions;

[0014] Calculate ventilation times according to the tracer gas concentration data of each experiment;

[0015] The ventilation frequency is merged with the tracer gas concentration data to construct a data set with a time series and a spatial series;

[0016] According to the data set, a preset machine learning model is trained to obtain a prediction model;

[0017] Acquiring indoor air online monitoring data, wherein the indoor air online monitoring data includes tracer gas concentration;

[0018] Calculate the concentration change available window within a preset period for the indoor air online monitoring data within the period;

[0019] The indoor air online monitoring data within the concentration change available window is input into the prediction model to obtain the ventilation frequency.

[0020] In the above-mentioned technical means, the traditional experimental method for calculating the ventilation rate is changed, and the ventilation rate is directly calculated by using the data of online indoor air quality monitoring combined with the machine learning model. Based on the fragmentary data of the actual changes in gas concentration in the space, a highly accurate ventilation efficiency evaluation can be obtained.

[0021] Furthermore, the collection of tracer gas concentration data varying with time under different spatial structures, different working conditions and different experimental conditions includes:

[0022] A tracer gas ventilation experimental detection method is adopted, and multiple detection devices are set under different space structures, different working conditions and different experimental conditions to obtain tracer gas concentration data. The tracer gas concentration data includes a timestamp at each moment and the tracer gas concentration obtained by each detection device at that moment.

[0023] Further, the ventilation times are calculated according to the tracer gas concentration data, including:

[0024]

[0025] In the formula, Indicates the ventilation times. Indicates the background value of the tracer gas, Indicates the tracer gas concentration at the start of the measurement, represents the tracer gas concentration at the measurement time t, Indicates the measurement time.

[0026] Furthermore, the ventilation frequency and the tracer gas concentration data are fused to construct a data set with time series and space series, including:

[0027] The ventilation times calculated at each moment are associated and fused with the tracer gas concentration to form a data set including timestamps, tracer gas concentrations and ventilation times records in different locations, different working conditions and different experimental conditions.

[0028] Furthermore, the concentration change available window within a preset period is calculated for the indoor air online monitoring data. When the continuous indoor air online monitoring data in a certain period meets the following conditions, the period is regarded as a preliminary concentration change available window:

[0029]

[0030] In the formula, Indicates the number of continuous indoor air online monitoring data in a certain period of time. Sampling point data, represents the number of samples, Indicates the concentration drop threshold, Indicates the maximum sampling point data of continuous indoor air online monitoring data in a certain period of time. Indicates the minimum sampling point data of continuous indoor air online monitoring data in a certain period of time;

[0031] Take all the preliminary concentration changes in the available window The largest available window is as a concentration change.

[0032] Furthermore, after the indoor air online monitoring data within the concentration change available window is input into the prediction model and the ventilation times are obtained, the following steps are also included:

[0033] According to the ventilation times obtained by the prediction model, the whole process change of the tracer gas concentration drop in the current place is calculated:

[0034]

[0035] In the formula, is the predicted tracer gas concentration at time t, is the last tracer gas concentration within the available window for the concentration change. and When the difference is lower than the preset threshold, ventilation is considered complete.

[0036] Furthermore, the predicted tracer gas concentration at time t is The tracer gas concentration at time t during the actual descent process Comparison is performed to obtain error data, and the prediction model is iterated according to the error data.

[0037] Furthermore, after the indoor air online monitoring data within the concentration change available window is input into the prediction model and the ventilation times are obtained, the following steps are also included:

[0038] The indoor air online monitoring data and the corresponding ventilation times are visualized.

[0039] Furthermore, the prediction model is used to make a long-term prediction of the ventilation frequency of the indoor space, and the Arima model is used to model the long-term predicted ventilation frequency to obtain the periodic changes of the ventilation frequency and the estimation of the future ventilation frequency.

[0040] Furthermore, the machine learning model includes a regression model.

[0041] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0042] The present invention can construct a data set for training the ventilation calculation model using laboratory simulation conditions. After completing the model training, it can calculate the ventilation times of different places in combination with the data of the online monitoring equipment, without the need for tracer gas method experiments. This greatly improves the ability to detect the ventilation times of places and can achieve low-cost and large-scale ventilation efficiency evaluation. At the same time, the model can also predict the trend of concentration changes in combination with the gas concentration data monitored online, thereby achieving the purpose of early prediction and warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A flow chart of a method for calculating the number of ventilation changes in an indoor space based on a pre-trained machine learning model and indoor air online monitoring data provided by an embodiment of the present invention;

[0044] Figure 2 A schematic diagram of the tracer gas concentration prediction result provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The accompanying drawings are only used for illustrative purposes and are not to be construed as limiting the present application;

[0046] In order to better illustrate the present embodiment, some parts in the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product;

[0047] It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0048] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0049] Example

[0050] This embodiment provides a method for calculating the ventilation times of indoor spaces based on a pre-trained machine learning model and indoor air online monitoring data. Figure 1 As shown, the following steps are included:

[0051] Collect tracer gas concentration data that varies with time under different spatial structures, different operating conditions, and different experimental conditions;

[0052] Calculate ventilation times according to the tracer gas concentration data of each experiment;

[0053] The ventilation frequency is merged with the tracer gas concentration data to construct a data set with a time series and a spatial series;

[0054] According to the data set, a preset machine learning model is trained to obtain a prediction model;

[0055] Acquiring indoor air online monitoring data, wherein the indoor air online monitoring data includes tracer gas concentration;

[0056] Calculate the concentration change available window within a preset period for the indoor air online monitoring data within the period;

[0057] The indoor air online monitoring data within the concentration change available window is input into the prediction model to obtain the ventilation frequency.

[0058] The embodiment of the present invention first collects the entire attenuation data of the tracer gas concentration during the experiment, trains a prediction model, and then combines the existing system and data of the increasingly popular online monitoring of indoor air quality to realize the online evaluation of ventilation efficiency through the pre-trained prediction model. The model can calculate the ventilation times by using the concentration change of a certain indicator of the air monitored online.

[0059] In a further embodiment, the collecting of tracer gas concentration data varying with time under different spatial structures, different operating conditions and different experimental conditions comprises:

[0060] A tracer gas ventilation experimental detection method is adopted, and multiple detection devices are set under different space structures, different working conditions and different experimental conditions to obtain tracer gas concentration data. The tracer gas concentration data includes a timestamp at each moment and the tracer gas concentration obtained by each detection device at that moment.

[0061] In this embodiment, tracer gas concentration data under different spatial structures, different working conditions and different experimental conditions are collected to cover as many places with different spatial structures and ventilation efficiencies as possible to ensure the wide applicability and representativeness of the data.

[0062] At the same time, in order to improve the accuracy of model prediction, the tracer gas concentration sampling rate is required to reach the second level during the tracer gas ventilation test to ensure high temporal resolution of the data, thereby capturing the subtle dynamics of concentration changes and providing higher quality training data for model optimization. To ensure the accuracy of ventilation test detection, each test experiment deploys multiple tracer gas concentration detectors according to the spatial structure of the site to collect and record concentration change data in real time.

[0063] This embodiment also provides examples of tracer gas concentration data, as shown in Table 1.

[0064] Table 1

[0065]

[0066] In a further embodiment, calculating the ventilation rate according to the tracer gas concentration data includes:

[0067]

[0068] In the formula, Indicates the ventilation times. Indicates the background value of the tracer gas, Indicates the tracer gas concentration at the start of the measurement, in mg / m³ or %. Indicates the tracer gas concentration at the measurement time t, in mg / m³ or %. Indicates the measurement time.

[0069] In a further embodiment, before calculating the ventilation times, a data cleaning step is also included:

[0070] For the tracer gas concentration data collected in each experiment, the tracer gas concentration is resampled at the specified sampling rate to ensure that there are no null values ​​in each record. The resampled data is cleaned, including missing value processing, outlier detection / deletion / replacement, data format standardization, data deduplication and other steps. The cleaned data of each experiment is saved according to the experiment sequence number.

[0071] In a further embodiment, the ventilation frequency is fused with the tracer gas concentration data to construct a data set with a time series and a spatial series, including:

[0072] After the ventilation times calculated at each moment are associated and fused with the tracer gas concentration, a data set including timestamps, tracer gas concentrations and ventilation times records in different locations, different working conditions, and different experimental conditions is formed. The specific samples are shown in Table 2.

[0073] Table 2

[0074]

[0075] In a specific embodiment, the data of all experiments are fused to form a data set with time series and space series, including timestamps, tracer gas concentration data and calculated ventilation times. The data set is constructed by staggered sampling. The data set contains complete ventilation data of all experiments, including concentration changes and corresponding ventilation times in different locations, different working conditions, and different experimental conditions (such as peak concentration, etc.), which is highly representative.

[0076] In a further embodiment, the data set is divided into a training set and a test set by random sampling to ensure the effectiveness and reliability of model training and testing.

[0077] The data set construction process of this embodiment ensures the comprehensiveness, accuracy and representativeness of the data set, and provides a solid foundation for subsequent model optimization and prediction.

[0078] In a further embodiment, an indoor air quality online monitoring system is used to continuously collect tracer gas concentration data in the air. The data sampling frequency is ensured to be close to the sampling frequency during machine learning model training to ensure the accuracy of the prediction results. The tracer gas concentration data collected in real time is preprocessed, including removing noise, processing missing values ​​and outliers, and data smoothing, to ensure that the data quality meets the model input requirements.

[0079] In a further embodiment, in this embodiment, the prediction model is trained based on the tracer gas concentration change process. When using this prediction model in combination with the online air quality monitoring data to calculate the ventilation times, there are certain requirements for the concentration change rate of the online monitoring data. Since the ventilation times are determined by the speed and time of the gas concentration change, and the gas concentration in the daily environment does not change much, it cannot be used for ventilation prediction. Therefore, it is necessary to select conditions that simultaneously meet: a downward trend, large downward fluctuations, and an initial value greater than a certain value, that is, the continuously monitored concentration is required to have a certain rate of change, and the air quality monitoring data of most indoor places can meet this concentration change requirement, and the use of the prediction model to calculate the ventilation times only requires a window segment of concentration change data, which generally only needs to be at the minute level, but the calculation window is required to have certain typicality and regularity, and can better characterize the ventilation effect of the place; in order to achieve the best calculation effect, the embodiment of the present invention compares and identifies the online monitoring data according to a certain period range (such as daily or weekly) and calibrates the best available concentration change available window;

[0080] The indoor air online monitoring data is used to calculate the concentration change available window within a preset period. When the continuous indoor air online monitoring data in a certain period meets the following conditions, the period is regarded as the preliminary concentration change available window:

[0081]

[0082] In the formula, Indicates the number of continuous indoor air online monitoring data in a certain period of time. Sampling point data, represents the number of samples, Indicates the concentration drop threshold, Indicates the maximum sampling point data of continuous indoor air online monitoring data in a certain period of time. Indicates the minimum sampling point data of continuous indoor air online monitoring data in a certain period of time;

[0083] Take all the preliminary concentration changes in the available window The largest available window is as a concentration change.

[0084] In this embodiment, for the online monitoring data processing cycle, the lowest value of the gas concentration in the cycle is obtained as the background value, and the maximum value is the peak value. The concentration drop threshold and time window are automatically evaluated based on the change process of the background value and the peak value. After the concentration change available window is identified, the data of this section is taken out for subsequent prediction calculation. In order to be more appropriate for the pre-experiment of training, in the multiple preliminary concentration change available windows identified, select The largest one is selected as the available concentration change window.

[0085] In a further embodiment, the prediction model can be combined with the data of the current concentration change available window to predict the real-time change of the tracer gas concentration in the current space. By predicting the change of the tracer gas concentration, the sudden harmful factors of indoor air quality can be predicted and warned, specifically:

[0086] After inputting the indoor air online monitoring data within the concentration change available window into the prediction model to obtain the ventilation times, the following steps are also included:

[0087] According to the ventilation times obtained by the prediction model, the whole process of tracer gas concentration decrease in the current place is calculated in reverse:

[0088]

[0089] In the formula, is the predicted tracer gas concentration at time t, The last tracer gas concentration within the available window for the concentration change, is the expected background value, generally around 600, and the above formula is iterated until and When the difference is lower than the preset threshold (usually around 15), ventilation is considered completed.

[0090] In a further embodiment, the predicted tracer gas concentration at time t is The tracer gas concentration at time t during the actual descent process Compare and obtain the error data, and combine it with other factors to bring it into the following formula for continuous iterative optimization:

[0091]

[0092] In the formula, , represents the weight, , Indicates the corresponding influencing factor value, To a small constant value, initialize , The value of is 0. , is the factor value, such as time, whether it is a working day, error value, etc.;

[0093] The prediction model is iterated according to the error data so that the prediction model will automatically adjust its value in the long term to better fit the actual situation of the scene.

[0094] In a further embodiment, after the indoor air online monitoring data within the concentration change available window is input into the prediction model to obtain the ventilation times, the following steps are further included:

[0095] The indoor air online monitoring data and the corresponding ventilation times are visualized to view the past ventilation efficiency change trends.

[0096] In a further embodiment, the prediction model is used to make long-term predictions of the ventilation frequency of indoor spaces, and the Arima model is used to model the long-term predicted ventilation frequency to obtain periodic changes in the ventilation frequency and estimates of future ventilation frequency, so as to further evaluate and analyze the ventilation change trends of the venue and provide effective prediction analysis and intervention measures.

[0097] In this embodiment, Arima (Autoregressive Integrated Moving Average Model) is a statistical model used for time series forecasting and analysis. It combines the autoregressive model (AR) and the moving average model (MA) and has a difference operation (I, representing difference).

[0098] The parameters of the ARIMA model include three parts: p, d and q.

[0099] p represents the order of the autoregressive term, which represents the relationship between the current value and the value at p moments in the past.

[0100] d represents the order of the time series, that is, the number of differences, which is used to make the time series stationary.

[0101] q represents the order of the moving average term, which represents the linear combination of the current value and the error at the past q moments.

[0102] The ARIMA model selects appropriate p, d, and q values ​​by analyzing the autocorrelation and partial autocorrelation of the time series, and then fits the model using historical data and makes predictions.

[0103] The AR model believes that "history determines the future", so it largely ignores the complexity of the actual situation and the unpredictable impact of the factors that really affect the label. In contrast, the MA model believes that "time series are relatively stable, and the fluctuations of time series are determined by accidental factors", but it is difficult for time series in reality to always maintain the assumption of "stability". Based on this, the ARIMA model "combines" the above two models. The basic idea of ​​the ARIMA model is that the label value at a point in time is affected by both the label value in the past period of time and the accidental events in the past period of time. That is, the ARIMA model assumes that the label value fluctuates around the general trend of time, where the trend is affected by historical labels, the fluctuation is affected by accidental events in a period of time, and the general trend itself is not necessarily stable. Guided by this central idea, the formula of the ARIMA model is expressed as:

[0104]

[0105] In this formula, the meaning of each variable is exactly the same as in the MA and AR models, and the solution method of each variable is basically the same as the original MA and AR models. The first half of the formula is the AR model, and the second half is the "volatility" part of the MA model. The mean μ representing the long-term trend in the MA model does not exist in the formula of the ARIMA model, because the "predicting long-term trends" part of the ARIMA model is performed by the AR model, so the AR model replaces the original μ. In the ARIMA model, β0 can be 0.

[0106] The above model is called the ARIMA (p, d, q) model, where the meanings of p and q are exactly the same as in the original MA and AR models, and p and q can be set to different values, and d is the order of difference required by the ARIMA model.

[0107] In a further embodiment, regarding the type of tracer gas, as long as the tracer gas method and online monitoring can be met and the gas can be obtained, the method of this embodiment can be used.

[0108] In a further embodiment, the machine learning model comprises a regression model.

[0109] In a specific embodiment, an experimental data set is used to train a machine learning model to fit the correlation between concentration changes and ventilation times. By comparing the fitting effects of different machine learning models, the embodiment of the present invention selects a Bagging regression model as an implementation description; other types of machine learning models also belong to the implementation extension of the present invention under the condition of satisfying the fitting effect.

[0110] Implementation process: Import the preprocessed data set into the Bagging regression model, set different sampling intervals, basic models, number of sub-models, and sample size of sub-models for cyclic training, and obtain the prediction model of ventilation times. Evaluate the model by mean square error, and improve the accuracy of model prediction by adjusting parameters and optimizing training set sampling, and finally save the model locally.

[0111] The Bagging regression model is an integrated learning method that improves the accuracy and stability of the overall prediction by combining the predictions of multiple base regression models. It randomly extracts multiple sample sets from the original data set through the self-service sampling technique, trains a base model for each sample set, and then performs simple averaging or other forms of aggregation on the prediction results of these base models to obtain the final prediction results. The Bagging method can effectively reduce the variance of the model, combat overfitting, and is applicable to a variety of regression problems, especially when a single model is susceptible to random fluctuations in the data. By adjusting the number and type of base models, Bagging regression can flexibly adapt to different data sets and prediction tasks.

[0112] The training steps of the Bagging regression model are:

[0113] Self-service sampling: n samples are randomly selected from the original data set repeatedly to construct a new training set. This process is repeated k times to generate k training sets. k is a custom preset value that determines the number of base regressors constructed. In this model, n is the size of the source data set and k is 100.

[0114] Training base regressors: For each new training set generated by sampling, a multivariate linear regression model is trained independently. That is, when there are k training sets, k linear regression models will be trained, and each linear regression is regarded as a base regressor.

[0115] Aggregate prediction: The prediction results of all base regressors are calculated in parallel and their arithmetic mean is used as the final prediction result. This method reduces the variance of the model and makes the final model more robust to small fluctuations in the training data.

[0116] The Bagging regression model has the following advantages:

[0117] 1) Low risk of overfitting: The Bagging model constructs a data set through multiple sampling for independent training, and comprehensively evaluates and predicts all results, effectively reducing the risk of overfitting of the model.

[0118] 2) High stability: Even if the original data changes slightly, the final prediction result will be more stable by averaging the prediction results of multiple models.

[0119] 3) High accuracy: An ensemble of multiple models usually performs better than a single model.

[0120] The specific steps of using the Bagging regression model to predict ventilation times are as follows:

[0121] Using the sample data in the available window of concentration change calibrated above, divide it one by one according to the time window required by model training, stagger sampling to obtain nk groups of data (n is the amount of available data, k is the amount of data required for model prediction), and import the data into the model one by one. The model will generate a ventilation frequency based on each group of data, and average all predicted ventilation frequencies. The ventilation frequency obtained is used to calculate the ventilation frequency before reaching the background value. The comparison diagram between the predicted results and the actual results is shown in the figure below. Figure 2 The model calculation formula is as follows:

[0122]

[0123] in It is a simplified form of the Bagging model, n is the amount of available data, and k is the amount of data required for model prediction.

[0124] In this way, the online monitoring of indoor air quality and ventilation calculation can be integrated to achieve online evaluation of ventilation efficiency. There is no need for special ventilation detection experiments, which greatly reduces manpower and material resources and improves efficiency. In addition, in this way, the ventilation times can be calculated continuously to form a time series change data set, which can further evaluate the changing trend of the ventilation effect of the indoor space and the cause correlation analysis.

[0125] The method of this embodiment uses the data of online monitoring of indoor air quality to evaluate ventilation efficiency, replacing the traditional tracer gas test method or the evaluation method based on indoor wind speed measurement, and has the following advantages:

[0126] 1) Method innovation: Change the traditional experimental method to calculate the ventilation rate, and directly use the data of online indoor air quality monitoring combined with artificial intelligence model to calculate the ventilation rate. Based on the fragmentary data of the actual gas concentration changes in the space, a highly accurate ventilation efficiency evaluation can be obtained.

[0127] 2) Low sample volume requirement: The ventilation frequency of the site can be predicted with only a few minutes of data, and the concentration trend of the complete experiment can be calculated based on the currently collected tracer gas concentration data

[0128] 3) Real-time: The resource overhead required for the prediction process is extremely low. When new collected data is added, the model prediction results can be automatically corrected based on the new data and displayed in real time.

[0129] 4) Predictiveness: The model can not only calculate the tracer gas concentration and ventilation frequency of the current data, but also perform predictive processing. When the data deviates from the range, the prediction system can sense and issue an alarm.

[0130] 5) Visualization: The system can generate a trend graph of the predicted tracer gas concentration trend.

[0131] 6) Comprehensive evaluation models based on multiple machine learning methods: including methods for identifying available windows for concentration changes and ventilation efficiency evaluation models.

[0132] 7) Automatic iteration: Based on the long-term online ventilation efficiency evaluation results, the model is automatically optimized and iterated for different scenarios.

[0133] 8) High accuracy: After verification and comparison with the tracer gas laboratory method, the accuracy rate is over 80%.

[0134] The same or similar reference numerals correspond to the same or similar components;

[0135] The terms used to describe the positional relationship in the drawings are only used for illustrative purposes and should not be construed as limiting the present application;

[0136] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A method for calculating the number of ventilation changes in indoor space based on a pre-trained machine learning model and indoor air online monitoring data, characterized in that: The following steps are involved: Collect tracer gas concentration data that varies with time under different spatial structures, different operating conditions, and different experimental conditions; Calculate ventilation times according to the tracer gas concentration data of each experiment; The ventilation frequency is merged with the tracer gas concentration data to construct a data set with a time series and a spatial series; According to the data set, a preset machine learning model is trained to obtain a prediction model; Acquiring indoor air online monitoring data, wherein the indoor air online monitoring data includes tracer gas concentration; Calculate the concentration change available window within a preset period for the indoor air online monitoring data within the period; Inputting the indoor air online monitoring data within the concentration change available window into the prediction model to obtain ventilation times; The concentration change available window within the preset period is calculated for the indoor air online monitoring data. When the continuous indoor air online monitoring data in a certain period meets the following conditions, the period is regarded as the preliminary concentration change available window: In the formula, Indicates the number of continuous indoor air online monitoring data in a certain period of time. Sampling point data, represents the number of samples, Indicates the concentration drop threshold, Indicates the maximum sampling point data of continuous indoor air online monitoring data in a certain period of time. Indicates the minimum sampling point data of continuous indoor air online monitoring data in a certain period of time; Take all the preliminary concentration changes in the available window The largest available window is as a concentration change.

2. The method for calculating the number of ventilation changes in indoor space based on a pre-trained machine learning model and indoor air online monitoring data according to claim 1, characterized in that: The collection of tracer gas concentration data varying with time under different spatial structures, different working conditions and different experimental conditions includes: A tracer gas ventilation experimental detection method is adopted, and multiple detection devices are set under different space structures, different working conditions and different experimental conditions to obtain tracer gas concentration data. The tracer gas concentration data includes a timestamp at each moment and the tracer gas concentration obtained by each detection device at that moment.

3. The method for calculating the number of ventilation changes in indoor space based on a pre-trained machine learning model and indoor air online monitoring data according to claim 2, characterized in that: According to the tracer gas concentration data of each experiment, the ventilation air exchange rate is calculated, including: In the formula, Indicates the ventilation frequency. Indicates the background value of the tracer gas, Indicates the tracer gas concentration at the start of the measurement, represents the tracer gas concentration at the measurement time t, Indicates the measurement time.

4. The method for calculating the number of ventilation changes in indoor space based on a pre-trained machine learning model and indoor air online monitoring data according to claim 3, characterized in that: The ventilation frequency and the tracer gas concentration data are fused to construct a data set with a time series and a spatial series, including: The ventilation times calculated at each moment are associated and fused with the tracer gas concentration to form a data set including timestamps, tracer gas concentrations and ventilation times records in different locations, different working conditions and different experimental conditions.

5. The method for calculating the number of ventilation changes in indoor space based on a pre-trained machine learning model and indoor air online monitoring data according to claim 4, characterized in that: After inputting the indoor air online monitoring data within the concentration change available window into the prediction model to obtain the ventilation times, the following steps are also included: According to the ventilation times obtained by the prediction model, the whole process change of the tracer gas concentration drop in the current place is calculated: In the formula, represents the expected background value, The last tracer gas concentration within the available window for the concentration change, is the predicted tracer gas concentration at time t, Indicates ventilation times Multiply by the measurement time ,when and When the difference is lower than the preset threshold, ventilation is considered complete.

6. The method for calculating the number of ventilation changes in indoor space based on a pre-trained machine learning model and indoor air online monitoring data according to claim 5, characterized in that: The predicted tracer gas concentration at time t The tracer gas concentration at time t during the actual descent process Comparison is performed to obtain error data, and the prediction model is iterated according to the error data.

7. The method for calculating indoor space ventilation times based on a pre-trained machine learning model and indoor air online monitoring data according to claim 6, characterized in that: After inputting the indoor air online monitoring data within the concentration change available window into the prediction model to obtain the ventilation times, the following steps are also included: The indoor air online monitoring data and the corresponding ventilation times are visualized.

8. The method for calculating indoor space ventilation times based on a pre-trained machine learning model and indoor air online monitoring data according to claim 7, characterized in that: The prediction model is used to make a long-term prediction of the ventilation frequency of indoor space, and the Arima model is used to model the long-term predicted ventilation frequency to obtain the periodic change of ventilation frequency and the estimation of future ventilation frequency.

9. The method for calculating the ventilation times of indoor spaces based on a pre-trained machine learning model and indoor air online monitoring data according to any one of claims 1 to 8, characterized in that: The machine learning model includes a regression model.

Images