A fresh air temperature and humidity regulation method and system

Abstract

The present application provides a fresh air temperature and humidity regulation method and system, relating to the technical field of temperature and humidity regulation. The method includes steps such as data collection, first modeling, first model training, prediction, and regulation. The system includes a data collection module, a first modeling module, a first model training module, a prediction module, and a regulation module. The present application further adjusts the temperature and humidity parameters of the fresh air system by using environmental data, realizing intelligent management and optimized control of the fresh air system.

Description

Technical Field

[0001] This application relates to the technical field of temperature and humidity regulation, and particularly to a fresh air temperature and humidity regulation method and system. Background Art

[0002] With the development of Internet of Things technology and economy, most people have a higher pursuit of the quality of living environment. The quality of the indoor environment directly affects the physical health of the residents. Therefore, the monitoring and regulation of the indoor environment of houses have also become an important part of smart homes. In order to improve the quality of the indoor environment of houses, it is necessary to manage the monitoring and control of the indoor environment of houses. Currently, the monitoring of the indoor environment of houses mainly focuses on the corresponding air environment in the house, and then uses a fresh air system to regulate the air environment in the house to achieve the purpose of improving the indoor environment.

[0003] The Chinese invention patent with the publication date of February 13, 2024 and the publication number of CN117555376A provides an energy-saving regulation method and system for the living building environment based on big data. This patent imports the corresponding human body constitution data information into the constructed neural network model, outputs the appropriate temperature and humidity data of each person, obtains the critical illness value of the corresponding person and the calculated appropriate temperature and humidity data of the corresponding person, and imports them into the temperature and humidity value calculation strategy for the final calculation of temperature and humidity. The temperature and humidity controller in the building adjusts the living building environment to the final temperature and humidity in real time, analyzes the human body constitution characteristics in the living building, and constructs a deep learning neural network model to accurately calculate and export the appropriate temperature and humidity in the living environment. However, the indoor temperature and humidity data are also affected by environmental factors, such as light brightness, season, weather and other environmental factors. Therefore, this patent still has certain limitations in the temperature and humidity regulation of the indoor environment of houses. Summary of the Invention

[0004] In order to improve the accuracy of the temperature and humidity regulation of the fresh air system, this application provides a fresh air temperature and humidity regulation method and system.

[0005] In a first aspect, this application provides a fresh air temperature and humidity regulation method, adopting the following technical solution:

[0006] A fresh air temperature and humidity regulation method includes the following steps:

[0007] Data collection: including collecting real-time data and historical data;

[0008] Collecting real-time data: collecting real-time indoor temperature and humidity data, real-time environmental data, and real-time fresh air temperature and humidity data;

[0009] Collecting historical data: collecting historical indoor temperature and humidity data, historical environmental data, and historical fresh air temperature and humidity data;

[0010] First modeling: Based on machine learning algorithms, establish a BI-LSTM model;

[0011] First model training: Use historical indoor temperature and humidity data, historical environmental data, and historical fresh air temperature and humidity data to train the BI-LSTM model;

[0012] Prediction: Input real-time indoor temperature and humidity data and real-time environmental data into the trained BI-LSTM model, and output the predicted fresh air temperature and humidity data, denoted as the first data;

[0013] Adjustment: Adjust the real-time fresh air temperature and humidity data according to the first data.

[0014] By adopting the above technical solution, when predicting the fresh air temperature and humidity data, the present application takes into account environmental factors and improves the adjustment accuracy. During the prediction process, there are complex time-dependent relationships among environmental factors, and the BI-LSTM model can better capture these relationships, thereby improving the accuracy and reliability of the prediction. Through the training of historical data, the BI-LSTM model can learn the complex mapping relationship between environmental data and fresh air temperature and humidity. This training process enables the BI-LSTM model to predict the future fresh air temperature and humidity based on the current state, providing a scientific basis for subsequent adjustment. Input the real-time indoor temperature and humidity data and real-time environmental data into the trained BI-LSTM model, and the BI-LSTM model can output the predicted fresh air temperature and humidity data. Automatically adjust the real-time fresh air temperature and humidity data according to the fresh air temperature and humidity data predicted by the BI-LSTM model to achieve a more comfortable effect.

[0015] Optionally, in the step of data collection, it further includes:

[0016] Collect physical data: Collect current personnel physical data, historical personnel physical data, and historical fresh air temperature and humidity data suitable for personnel.

[0017] By adopting the above technical solution, after the present application takes into account the influence of environmental data on fresh air temperature and humidity data, it also pays attention to the physical data of personnel, and further provides a personalized adjustment method.

[0018] Optionally, after the step of performing the first model training and before the step of performing the prediction, it further includes:

[0019] Transfer learning: including second modeling and second model training;

[0020] Second modeling: Establish a prediction model, the prediction model includes an attention layer, a base model, a transfer model, and a fully connected layer, the base model is the trained BI-LSTM model, and the transfer model is a feedforward neural network model;

[0021] Second model training: Freeze the base model, and train the transfer model using historical personnel physical data and historical fresh air temperature and humidity data suitable for the personnel.

[0022] By adopting the above technical solution, the present application constructs a prediction model. The prediction model introduces an attention layer to endow the prediction model with the ability to focus on important parts in the input data, which helps the prediction model to more accurately understand complex data relationships. Especially when dealing with physical data and fresh air temperature and humidity data, the transfer model (feedforward neural network model) is used as a new learning component and can be optimized for a specific task (predicting suitable fresh air temperature and humidity based on personnel physical data). By freezing the base model (the trained BI-LSTM model), the knowledge already learned by the base model will not be changed when training the transfer model, thus ensuring the stability and accuracy of the prediction model when dealing with basic environmental data (such as indoor temperature and humidity, environmental data). Using historical personnel physical data and historical fresh air temperature and humidity data suitable for the personnel to train the transfer model enables it to learn the specific relationship between personnel physical data and the suitable fresh air temperature and humidity for the human body. By introducing transfer learning, historical data can be more effectively utilized, including physical data that may not have been fully utilized before, thereby improving the utilization value of the data and the prediction accuracy of the system.

[0023] Optionally, in the prediction step, input the real-time indoor temperature and humidity data, real-time environmental data, and current personnel physical data into the trained prediction model, and output new predicted fresh air temperature and humidity data.

[0024] By adopting the above technical solution, based on the processing and analysis of the input data, the prediction model will output a new predicted fresh air temperature and humidity data, which is the optimal or most suitable fresh air temperature and humidity data predicted by the prediction model according to the current environmental conditions and personnel physical data.

[0025] Optionally, the calculation model of the new predicted fresh air temperature and humidity data is as follows:

[0026] ;

[0027] ;

[0028] Wherein, is the new predicted fresh air temperature data; is the temperature data predicted by the base model; n is the number of indoor personnel; is the body temperature data of the i-th person; is the indoor temperature data; is the new predicted fresh air humidity data; Humidity data predicted by the base model; M is the human energy metabolic rate; is the convective mass transfer coefficient; P is the pressure formed by indoor water vapor, and the calculation model is as follows:

[0029] .

[0030] By adopting the above technical solution, based on the output data of the base model, the present application calculates the fresh air temperature and humidity according to the current environment and personnel physical data, enabling the prediction model to more accurately predict the fresh air temperature and humidity required by each person, thereby improving the comfort of the overall environment.

[0031] Optionally, after performing the step of the first model training and before performing the prediction step, it further includes:

[0032] Set the loss function: Set the loss function of the BI-LSTM model;

[0033] Set the evaluation matrix: Set the evaluation matrix, where each element in the evaluation matrix represents the probability that a sample belonging to the i-th fresh air temperature and humidity data is misclassified as the j-th fresh air temperature and humidity data, denoted as the first probability. The samples include: historical indoor temperature and humidity data, historical environmental data, and the diagonal elements of the evaluation matrix are 0;

[0034] Adjust the loss function: Adjust the weights in the loss function according to the evaluation matrix.

[0035] By adopting the above technical solution, the present application sets an evaluation matrix to quantify the accuracy of the BI-LSTM model. Each element in the evaluation matrix represents the probability that the BI-LSTM model should be classified as the i-th fresh air temperature and humidity data but is misclassified as the j-th fresh air temperature and humidity data, while the diagonal element is the probability that the i-th fresh air temperature and humidity data is misclassified as the i-th fresh air temperature and humidity data. At this time, it does not belong to the misclassification situation, so the diagonal element is set to 0. Adjusting the weights in the loss function according to the evaluation matrix can make the BI-LSTM model pay more attention to those samples or intervals that have a greater impact on the prediction performance during the training process. Optimizing the performance of the model on the evaluation matrix by adjusting the loss function helps to improve the generalization ability of the model on different data sets.

[0036] Optionally, after performing the step of setting the evaluation matrix and before performing the step of adjusting the loss function, it further includes:

[0037] Sum: Calculate the sum of the elements in each row of the evaluation matrix, denoted as the second data;

[0038] Normalize: Divide each element by the second data corresponding to its row to obtain a new evaluation matrix.

[0039] By adopting the above technical solution, each element in the evaluation matrix is divided by the "second data" of its row (i.e., the sum of that row), normalizing the elements of each row to the range of 0 to 1, or more generally, converting it into a ratio relative to the sum within the row, and then converting the evaluation matrix into a conditional probability matrix.

[0040] Optionally, after performing the step of adjusting the loss function and before performing the step of prediction, it further includes:

[0041] Third modeling: Establish a Bayesian network model based on historical indoor temperature and humidity data, historical environmental data, historical fresh air temperature and humidity data, and the first probability;

[0042] Inference: Use the Bayesian network model for inference to predict the fresh air temperature and humidity data when given real-time indoor temperature and humidity data and real-time environmental data, denoted as the third data.

[0043] By adopting the above technical solution, the Bayesian network model can learn and simulate the causal relationship between indoor temperature and humidity, environmental data, and fresh air temperature and humidity. Using the Bayesian network model for inference can quickly and accurately predict the fresh air temperature and humidity data (denoted as the third data) based on real-time indoor temperature and humidity data and real-time environmental data. Since the Bayesian network model can capture the complex relationships and causal relationships between variables, its prediction results are often more accurate than simple statistical models or machine learning models, which helps to improve the overall performance and reliability.

[0044] Optionally, after performing the step of prediction and before performing the step of adjustment, it further includes:

[0045] Calculate data: Assign weights to the first data and the third data, and calculate the final fresh air temperature and humidity data, taking the final fresh air temperature and humidity data as the new predicted fresh air temperature and humidity data.

[0046] By adopting the above technical solution, assigning weights to the first data and the third data is actually identifying and distinguishing the relative importance of these two sets of data in the final prediction. By combining the first data and the third data and considering their weights, the calculated final fresh air temperature and humidity data is the result of integrating multiple information sources and prediction methods. This method usually can provide more accurate and reliable predictions because it reduces the biases and uncertainties that may be brought by a single data source or prediction method.

[0047] In a second aspect, the present application provides a fresh air temperature and humidity adjustment system, adopting the following technical solution:

[0048] A fresh air temperature and humidity adjustment system includes:

[0049] The data acquisition module includes a real-time data acquisition unit and a historical data acquisition unit;

[0050] The real-time data acquisition unit is used to acquire real-time indoor temperature and humidity data, real-time environmental data, and real-time fresh air temperature and humidity data;

[0051] The historical data acquisition unit is used to acquire historical indoor temperature and humidity data, historical environmental data, and historical fresh air temperature and humidity data;

[0052] The first modeling module is used to establish a BI-LSTM model;

[0053] The first model training module is communicatively connected to the historical data acquisition unit and the first modeling module, and is used to train the BI-LSTM model with historical indoor temperature and humidity data, historical environmental data, and historical fresh air temperature and humidity data;

[0054] The prediction module is communicatively connected to the real-time data acquisition unit and the first model training module, and is used to input real-time indoor temperature and humidity data and real-time environmental data into the trained BI-LSTM model to output predicted fresh air temperature and humidity data;

[0055] The adjustment module is communicatively connected to the real-time data acquisition unit and the prediction module, and is used to adjust the real-time fresh air temperature and humidity data according to the predicted fresh air temperature and humidity data.

[0056] By adopting the above technical solutions, the system integrates several modules of data acquisition, model training, prediction, and adjustment. Through the comprehensive acquisition of real-time and historical data, the bidirectional long short-term memory network (BI-LSTM) model is used for training to learn the environmental change rules, and then the accurate prediction of fresh air temperature and humidity is realized. The prediction result drives the adjustment module to respond quickly and intelligently adjust the real-time fresh air temperature and humidity, so as to effectively maintain the comfort and stability of the indoor environment, and at the same time improve the energy use efficiency and operation economy.

[0057] In summary, the present application includes at least one of the following beneficial technical effects:

[0058] 1. When predicting the fresh air temperature and humidity data, this application takes into account environmental factors, improving the adjustment accuracy. During the prediction process, there are complex time-dependent relationships among environmental factors, and the BI-LSTM model can capture these relationships well, thereby improving the accuracy and reliability of the prediction. Through the training of historical data, the BI-LSTM model can learn the complex mapping relationship between environmental data and fresh air temperature and humidity. This training process enables the BI-LSTM model to predict future fresh air temperature and humidity based on the current environmental state, providing a scientific basis for subsequent adjustments. By inputting real-time indoor temperature and humidity data and real-time environmental data into the trained BI-LSTM model, the BI-LSTM model can output the predicted fresh air temperature and humidity data. According to the predicted fresh air temperature and humidity data of the BI-LSTM model, the real-time fresh air temperature and humidity data can be automatically adjusted to achieve a more comfortable effect.

[0059] 2. This application constructs a prediction model that introduces an attention layer to endow the prediction model with the ability to focus on important parts of the input data, which helps the prediction model better understand complex data relationships. Especially when dealing with physical data and fresh air temperature and humidity data, the transfer model (feed-forward neural network model), as a new learning component, can be optimized for a specific task (predicting the appropriate fresh air temperature and humidity based on personnel physical data). By freezing the base model (the trained BI-LSTM model), the knowledge it has learned will not be changed during the training of the transfer model, thus ensuring the stability and accuracy of the prediction model when processing basic environmental data (such as indoor temperature and humidity, environmental data). The transfer model is trained using historical personnel physical data and historical fresh air temperature and humidity data suitable for personnel, enabling it to learn the specific relationship between personnel physical data and the fresh air temperature and humidity suitable for the human body. By introducing transfer learning, the system can make more effective use of historical data, including physical data that may not have been fully utilized before, thereby improving the data utilization value and the prediction accuracy of the system.

[0060] 3. This application sets an evaluation matrix to quantify the accuracy of the BI-LSTM model. Each element in the evaluation matrix represents the probability that the BI-LSTM model should classify the i-th fresh air temperature and humidity data but misclassifies it as the j-th fresh air temperature and humidity data, while the diagonal elements are the probabilities that the i-th fresh air temperature and humidity data is misclassified as the i-th fresh air temperature and humidity data. In this case, it does not belong to the misclassification situation, so the diagonal elements are set to 0. By adjusting the weights in the loss function according to the evaluation matrix, the BI-LSTM model can pay more attention to those samples or intervals that have a greater impact on performance during the training process. Optimizing the model's performance on the evaluation matrix by adjusting the loss function helps improve the generalization ability of the model on different datasets. Description of the Drawings

[0061] Figure 1 is the flowchart of Embodiment 1 of this application;

[0062] Figure 2 is the flowchart of Embodiment 2 of this application;

[0063] Figure 3 is the flowchart of Embodiment 3 of this application. Detailed implementation manners

[0064] The following Figures 1 to 3 further elaborates on this application in detail.

[0065] Embodiment 1: This embodiment discloses a fresh air temperature and humidity regulation method. Referring to Figure 1 , the regulation method includes: S1 data collection, S2 first modeling, S3 first model training, S4 prediction, and S5 regulation. First, data collection is performed, then a BI-LSTM model is established, then the BI-LSTM model is trained, then the trained BI-LSTM model is used to predict fresh air temperature and humidity data based on real-time indoor temperature and humidity data and real-time environmental data, and finally the real-time fresh air temperature and humidity data are regulated. This embodiment includes the following steps:

[0066] S1 data collection, including S11 collecting real-time data and S12 collecting historical data.

[0067] S11 collecting real-time data, collecting real-time indoor temperature and humidity data, real-time environmental data, and real-time fresh air temperature and humidity data.

[0068] Real-time indoor temperature and humidity data: The current indoor temperature and humidity information are collected in real time through temperature and humidity sensors installed indoors.

[0069] Real-time environmental data: including light intensity, CO2 concentration, PM2.5 concentration, etc. These data are important factors affecting indoor comfort and air quality and are monitored in real time through corresponding sensors.

[0070] Real-time fresh air temperature and humidity data: When the fresh air system introduces external air, its temperature and humidity are also key factors affecting the indoor environment. These real-time data are obtained through temperature and humidity sensors installed at the inlet of the fresh air system.

[0071] S12 collecting historical data, collecting historical indoor temperature and humidity data, historical environmental data, and historical fresh air temperature and humidity data.

[0072] Historical indoor temperature and humidity data: The indoor temperature and humidity data over a past period are collected. These data are used to analyze the long-term change trend and seasonal characteristics of the indoor environment.

[0073] Historical environmental data: Corresponding to the real-time environmental data, historical data of environmental parameters such as past light intensity, CO2 concentration, and PM2.5 concentration are also collected.

[0074] Historical fresh air temperature and humidity data: Record the temperature and humidity data of the fresh air system over a past period of time to help understand the changes in fresh air quality and its impact on the indoor environment.

[0075] S2 First modeling, based on machine learning algorithms, establish a BI-LSTM model.

[0076] S3 First model training, use the historical data collected in S12 (historical indoor temperature and humidity data, historical environmental data, historical fresh air temperature and humidity data) as the training data set to train the BI-LSTM model. During the training process, use the historical indoor temperature and humidity data and the historical environmental data collected at the same moment as the input data, and use the historical fresh air temperature and humidity data as the true label of the input data to train the BI-LSTM model. During the training process, the BI-LSTM model will learn the laws and patterns in these historical data so that it can make predictions on fresh air temperature and humidity data based on new input data in the future, that is, predict the labels of real-time indoor temperature and humidity data and real-time environmental data.

[0077] The historical fresh air temperature and humidity data mentioned in this step are the fresh air temperature and humidity data in a stable state. The fresh air temperature and humidity data collected during the adjustment process (that is, the fresh air temperature and humidity data in a changing process) will be filtered out and not used for the training of the BI-LSTM model.

[0078] S4 Prediction, input the real-time indoor temperature and humidity data and real-time environmental data into the trained BI-LSTM model. The trained BI-LSTM model will output the predicted fresh air temperature and humidity data, that is, the first data, according to these real-time data, combined with the historical laws and patterns it has learned.

[0079] S5 Adjustment, adjust the real-time fresh air temperature and humidity data according to the first data, that is: if the prediction result shows that the first data is not equal to the real-time fresh air temperature and humidity data, automatically adjust the settings of the fresh air system, and then adjust the temperature or humidity of the fresh air to make the indoor environment reach the best comfort level and air quality.

[0080] In this embodiment, by collecting real-time indoor temperature and humidity, real-time environmental data, and real-time fresh air temperature and humidity data, and using historical data (historical indoor temperature and humidity data, historical environmental data, and historical fresh air temperature and humidity data) to train the BI-LSTM model, this model can predict fresh air temperature and humidity based on real-time data. Then, according to the prediction result of the BI-LSTM model, the fresh air system is automatically adjusted to make the indoor environment reach the best comfort level, realizing intelligent environmental control and management.

[0081] Example 2: Refer to Figure 2 , the difference between this embodiment and Embodiment 1 is that in S1 data collection, it further includes:

[0082] S13 Collect physical data, including the current person's physical data, historical person's physical data, and the fresh air temperature and humidity data suitable for historical persons.

[0083] The purpose of this step is to collect the current person's physical data (such as age, gender, height, weight, etc.), historical person's physical data (i.e., the physical data of other persons), and the fresh air temperature and humidity data suitable for historical persons (i.e., the fresh air temperature and humidity data suitable for other persons). The fresh air temperature and humidity data suitable for historical persons is obtained based on statistical analysis. For example, the fresh air temperature and humidity data applicable to this person at different time periods stored in the database reflects the most suitable fresh air temperature and humidity data for persons with different physiques under different conditions.

[0084] When the quantity of physical data of persons is insufficient, a generative adversarial network is also used to generate similar data to enrich the quantity and types of physical data of persons.

[0085] After performing S3 first model training and before performing S4 prediction, it further includes:

[0086] S31 Transfer learning, including S311 second modeling and S312 second model training.

[0087] S311 Second modeling, establish a prediction model. The prediction model includes an attention layer, a base model, a transfer model, and a fully connected layer. The base model is a trained BI-LSTM model, and the transfer model is a feedforward neural network model.

[0088] The input of the attention layer is used to receive input data. The output of the attention layer is respectively connected to the input of the base model and the input of the transfer model. The input of the fully connected layer is respectively connected to the output of the base model and the output of the transfer model. The fully connected layer outputs the final prediction result. The attention layer is used to highlight important information (by assigning different weights to different input data). The base model provides basic time series analysis capabilities and provides a predicted value of fresh air temperature and humidity data based on time patterns. The transfer model is used to capture the influence of physical data on the fresh air temperature and humidity requirements and provides a predicted value of fresh air temperature and humidity data based on physical data. The fully connected layer is responsible for integrating all this information to output the final prediction result.

[0089] S312 Second model training: Freeze the base model, that is, the parameters of the base model remain unchanged during training to avoid destroying the patterns it has learned about time series data (various historical data). Then, use the historical personnel physical data and the historical fresh air temperature and humidity data suitable for the personnel to train the transfer model so that it can learn how the physical data affects the suitability of fresh air temperature and humidity. During the training of the transfer model, the historical personnel physical data is used as the input data of the transfer model, and the historical fresh air temperature and humidity data suitable for the personnel is used as the true label of the historical personnel physical data.

[0090] In S4 prediction, input the real-time indoor temperature and humidity data, real-time environmental data, and current personnel physical data into the trained prediction model together. The prediction model will comprehensively consider these input information and, through the coordinated action of the attention layer, base model, transfer model, and fully connected layer, output the new predicted fresh air temperature and humidity data. The calculation model of the new predicted fresh air temperature and humidity data is as follows:

[0091] ;

[0092] ;

[0093] Among them, is the new predicted fresh air temperature data; is the temperature data predicted by the base model; n is the number of indoor personnel; is the body temperature data of the i-th person, obtained through wearable devices such as smart bracelets, thermometers, etc., or obtained non-contact through an infrared thermometer; is the indoor temperature data, measured by arranging temperature sensors indoors; is the new predicted fresh air humidity data; is the humidity data predicted by the base model.

[0094] M is the human energy metabolism rate. When the indoor personnel are in an active state, its calculation model is as follows:

[0095] ;

[0096] is the oxygen consumption, obtained through respiratory measurement; is the oxygen thermal equivalent, and the specific value is obtained by referring to the oxygen thermal equivalent table corresponding to NPRQ. NPRQ is the non-protein respiratory quotient, and its calculation formula is as follows:

[0097] ;

[0098] Among them, is the carbon dioxide production, obtained through respiratory measurement.

[0099] When the indoor personnel are in an inactive state, that is, a resting state, the calculation model of the human body energy metabolism rate M is as follows:

[0100] ;

[0101] Among them, TZ represents body weight, in kg; SG represents height, in cm; NL represents age, in years.

[0102] is the convective mass transfer coefficient, in (W·m) / kJ, and the convective mass transfer coefficient has the following calculation model:

[0103] ;

[0104] Among them, is the fluid density, in kg / m 3 , and the value range is [1.1, 1.3]; is the specific heat capacity at constant pressure, in kJ / (kg·K), with a value of 1.005; h is the convective heat transfer surface coefficient, in W / (m 2 ·K), and the value range is [5, 50].

[0105] In other embodiments, the convective mass transfer coefficient can also be calculated by the Buzzi formula or estimated by using the Sherwood criterion.

[0106] P is the pressure formed by indoor water vapor, and the calculation model is as follows:

[0107] .

[0108] This embodiment adds the collection of current and historical personnel physical data and the recording of the appropriate fresh air temperature and humidity for historical personnel. After the BI-LSTM model is trained, through transfer learning techniques (including establishing a prediction model combining an attention layer, a frozen BI-LSTM base model, a transfer model, and a fully connected layer, and training with physical and historical data), the prediction ability of the prediction model is further improved. Then, in the S4 prediction, the real-time environment, indoor temperature and humidity, and current personnel physical data are input into the prediction model to predict more personalized and accurate fresh air temperature and humidity data.

[0109] Embodiment 3: Refer to Figure 3 , the difference between this embodiment and Embodiment 1 is that after performing the first model training in S3 and before performing the S4 prediction, it further includes:

[0110] S61 Set the loss function, and set the loss function of the BI-LSTM model. The loss function is the key to optimizing the BI-LSTM model. It measures the difference between the predicted value and the actual value, and trains the BI-LSTM model by minimizing this difference.

[0111] S62 Set the evaluation matrix, and set an evaluation matrix. Each element in the evaluation matrix represents the probability that a sample belonging to the i-th fresh air temperature and humidity data is misclassified as the j-th fresh air temperature and humidity data, denoted as the first probability. The samples include historical indoor temperature and humidity data and historical environmental data. The diagonal elements of the evaluation matrix are 0.

[0112] Each element in the evaluation matrix represents the probability that a sample belonging to the i-th fresh air temperature and humidity data is misclassified as the j-th fresh air temperature and humidity data. That is, the first probability represents the probability when the prediction result of the BI-LSTM model is a certain wrong result.

[0113] S63 Sum, calculate the sum of the elements in each row of the evaluation matrix, denoted as the second data.

[0114] S64 Normalize, divide each element by the second data of the corresponding row respectively to achieve normalization and obtain a new evaluation matrix.

[0115] S65 Adjust the loss function, and adjust the weight in the loss function according to the evaluation matrix.

[0116] The calculation model for setting the loss function is as follows:

[0117] ;

[0118] where L is the first loss function, m is the number of input data of the BI-LSTM model, is the true label of the i-th input data (i.e., the historical fresh air temperature and humidity data), is the predicted label of the BI-LSTM model for the i-th input data (i.e., the predicted fresh air temperature and humidity data).

[0119] The calculation model for the loss function after adjusting the weight is as follows:

[0120] ;

[0121] m is the number of input data of the BI-LSTM model; when i = j, ; is the value of the element that classifies the predicted label of the i-th input data as the predicted label of the j-th input data in the evaluation matrix; is the predicted label of the BI-LSTM model for (i.e., the wrong label).

[0122] S66 Third modeling, based on historical indoor temperature and humidity data, historical environmental data, historical fresh air temperature and humidity data, and the first probability, a Bayesian network model is established. The nodes of the Bayesian network model are: historical indoor temperature and humidity data, historical environmental data, historical fresh air temperature and humidity data, and the value of the directed edge of the Bayesian network model is equal to one minus the first probability.

[0123] S67 Inference, using the Bayesian network model for inference, predicting the fresh air temperature and humidity data when given real-time indoor temperature and humidity data and real-time environmental data, denoted as the third data.

[0124] Then execute S4 prediction, and after executing S4 prediction, this embodiment further includes:

[0125] S7 Calculate data: By assigning weights to the prediction result (the first data) of the BI-LSTM model and the inference result (the third data) of the Bayesian network model, and calculating the weighted average, the final real-time fresh air temperature and humidity data is obtained, and the final fresh air temperature and humidity data is used as the new predicted fresh air temperature and humidity data to execute S5 adjustment.

[0126] The calculation model of the final real-time fresh air temperature and humidity data is as follows:

[0127] ;

[0128] ;

[0129] Where, is the final real-time fresh air temperature and humidity data, is the weight of the first data, is the value of the first data; is the weight of the third data, is the value of the third data.

[0130] In this embodiment, after the BI-LSTM model is trained, by setting and adjusting the loss function, constructing an evaluation matrix and analyzing prediction errors, introducing the Bayesian network model for independent inference, and finally fusing the prediction results of the two models, the prediction process of fresh air temperature and humidity is adjusted and optimized in a more refined manner, aiming to improve the accuracy and reliability of the prediction.

[0131] Embodiment 4: This embodiment discloses a fresh air temperature and humidity regulation system, and the system includes:

[0132] A data acquisition module, including a real-time data acquisition unit and a historical data acquisition unit.

[0133] The real-time data acquisition unit is used for capturing real-time data, mainly including real-time indoor temperature and humidity data, real-time environmental data, and real-time fresh air temperature and humidity data. The real-time indoor temperature and humidity data reflect the current indoor temperature and humidity data, while the real-time environmental data (such as outdoor temperature, humidity, air pressure, etc.) provides external factors that may affect the indoor environment. These data are obtained in real time through hardware devices such as sensors, ensuring the timeliness and accuracy of the data.

[0134] The historical data acquisition unit is used for collecting data over a past period of time, including historical indoor temperature and humidity data, historical environmental data, and historical fresh air temperature and humidity data. Historical data is crucial for the training of the model because it provides rich samples to help the model learn and identify complex relationships between data. Through the long-term accumulated historical data, the model can predict future states more accurately.

[0135] The first modeling module is used to establish a BI-LSTM model. The BI-LSTM model is a special recurrent neural network that can handle long-term dependence problems in sequence data and consider the context information of the sequence at the same time. In this embodiment, the BI-LSTM model is used to learn the complex relationships between indoor temperature and humidity, environmental data, and fresh air temperature and humidity.

[0136] The first model training module is communicatively connected to the historical data acquisition unit and the first modeling module, and is used to train the BI-LSTM model using the historical data provided by the historical data acquisition unit. Through continuous iteration and optimization, the BI-LSTM model can gradually learn the internal laws and patterns between various data, thereby improving the prediction accuracy. After training, the BI-LSTM model will be capable of predicting fresh air temperature and humidity based on real-time data.

[0137] The prediction module is communicatively connected to the real-time data acquisition unit and the first model training module, and is used to input the real-time indoor temperature and humidity data and real-time environmental data into the trained BI-LSTM model. The BI-LSTM model calculates the predicted value of fresh air temperature and humidity based on the input data and outputs this value to subsequent modules.

[0138] The adjustment module is communicatively connected to the real-time data acquisition unit and the prediction module, and is used to adjust the real-time fresh air temperature and humidity data according to the predicted fresh air temperature and humidity data.

[0139] The data acquisition module in this embodiment is responsible for collecting real-time and historical data, including indoor temperature and humidity, environment, and fresh air temperature and humidity. The first modeling module establishes a BI-LSTM model, and the first model training module trains this model using historical data. The prediction module inputs real-time data into the trained model and outputs the predicted value of fresh air temperature and humidity. The adjustment module automatically adjusts the real-time fresh air temperature and humidity data according to the prediction result to achieve intelligent environmental control.

[0140] The above are all the preferred embodiments of the present application, and do not limit the protection scope of the present application accordingly. Therefore, all equivalent changes made according to the structure, shape and principle of the present application shall be covered within the protection scope of the present application.

Claims

1. A fresh air temperature and humidity adjustment method, characterized in that, Including: Data collection: including collecting real-time data and historical data; Collecting real-time data: collecting real-time indoor temperature and humidity data, real-time environmental data, real-time fresh air temperature and humidity data; Collecting historical data: collecting historical indoor temperature and humidity data, historical environmental data, historical fresh air temperature and humidity data; First modeling: Based on machine learning algorithms, establish a BI-LSTM model; First model training: Use historical indoor temperature and humidity data, historical environmental data, and historical fresh air temperature and humidity data to train the BI-LSTM model; Set the loss function: Set the loss function of the BI-LSTM model; Set the evaluation matrix: Set the evaluation matrix, where each element in the evaluation matrix represents the probability that a sample belonging to the i-th fresh air temperature and humidity data is misclassified as the j-th fresh air temperature and humidity data, denoted as the first probability. The samples include: historical indoor temperature and humidity data, historical environmental data. The diagonal elements of the evaluation matrix are 0; Summation: Calculate the sum of the elements in each row of the evaluation matrix, denoted as the second data; Normalization: Divide each element by the second data of the corresponding row to obtain a new evaluation matrix; Adjust the loss function: Adjust the weights in the loss function according to the evaluation matrix; Prediction: Input the real-time indoor temperature and humidity data and real-time environmental data into the trained BI-LSTM model, and output the predicted fresh air temperature and humidity data, denoted as the first data; Adjustment: Adjust the real-time fresh air temperature and humidity data according to the first data.

2. The fresh air temperature and humidity adjustment method according to claim 1, wherein In the step of data collection, it also includes: Collecting physical data: Collecting current personnel physical data, historical personnel physical data, and historical fresh air temperature and humidity data suitable for personnel.

3. The fresh air temperature and humidity adjustment method according to claim 2, characterized in that After executing the step of the first model training and before executing the step of prediction, it also includes: Transfer learning: including second modeling and second model training; Second modeling: Establish a prediction model, which includes an attention layer, a base model, a transfer model, and a fully connected layer. The base model is the trained BI-LSTM model, and the transfer model is a feedforward neural network model; Second model training: Freeze the base model and use historical personnel physical data and historical fresh air temperature and humidity data suitable for personnel to train the transfer model.

4. The fresh air temperature and humidity adjustment method according to claim 3, characterized in that, In the step of prediction, input the real-time indoor temperature and humidity data, real-time environmental data, and current personnel physical data into the trained prediction model, and output the new predicted fresh air temperature and humidity data.

5. The fresh air temperature and humidity adjustment method according to claim 4, wherein The calculation model of the new predicted fresh air temperature and humidity data is as follows: ； ； Among them, is the newly predicted fresh air temperature data; is the temperature data predicted by the base model; n is the number of indoor people; is the body temperature data of the i-th person; is the indoor temperature data; is the newly predicted fresh air humidity data; is the humidity data predicted by the base model; M is the human energy metabolism rate; is the convective mass transfer coefficient; P is the pressure formed by indoor water vapor, and the calculation model is as follows: 。 6. The fresh air temperature and humidity adjustment method according to claim 1, characterized in that After executing the step of adjusting the loss function and before executing the step of prediction, it also includes: Third modeling: Based on historical indoor temperature and humidity data, historical environmental data, historical fresh air temperature and humidity data, and the first probability, establish a Bayesian network model; Inference: Use the Bayesian network model for inference to predict the fresh air temperature and humidity data when given real-time indoor temperature and humidity data and real-time environmental data, denoted as the third data.

7. The fresh air temperature and humidity adjustment method according to claim 6, characterized in that, After executing the step of prediction and before executing the step of adjustment, it also includes: Calculate data: Assign weights to the first data and the third data, and calculate the final fresh air temperature and humidity data, and use the final fresh air temperature and humidity data as the new predicted fresh air temperature and humidity data.

8. A fresh air temperature and humidity regulation system, the system being applicable to the method according to any one of claims 1-7, characterized in that, Including: The data acquisition module includes a real-time data acquisition unit and a historical data acquisition unit; The real-time data acquisition unit is used to acquire real-time indoor temperature and humidity data, real-time environmental data, and real-time fresh air temperature and humidity data; The historical data acquisition unit is used to acquire historical indoor temperature and humidity data, historical environmental data, and historical fresh air temperature and humidity data; The first modeling module is used to establish a BI-LSTM model; The first model training module is communicatively connected to the historical data acquisition unit and the first modeling module, and is used to train the BI-LSTM model with historical indoor temperature and humidity data, historical environmental data, and historical fresh air temperature and humidity data; The prediction module is communicatively connected to the real-time data acquisition unit and the first model training module, and is used to input the real-time indoor temperature and humidity data and the real-time environmental data into the trained BI-LSTM model and output the predicted fresh air temperature and humidity data; The adjustment module is communicatively connected to the real-time data acquisition unit and the prediction module, and is used to adjust the real-time fresh air temperature and humidity data according to the predicted fresh air temperature and humidity data.

Images