Dynamic scene energy consumption adjusting method based on deep learning
By establishing the relationship between indoor characteristics and equipment power through a deep learning-based TFT model, the problem of traditional systems being unable to respond to dynamic scene changes is solved, realizing intelligent management in teaching scenarios and improving the system's energy efficiency and comfort.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BUILDING DESIGN RES INST HARBIN INST OF TECH
- Filing Date
- 2026-02-04
- Publication Date
- 2026-05-19
AI Technical Summary
Traditional heating and ventilation systems struggle to respond in real time to dynamic changes, resulting in inefficient thermal comfort management and increased energy consumption. Existing forecasting methods lack accuracy and adaptability in dynamic scenarios, making it difficult to achieve a balance between comfort and energy conservation goals.
Using a deep learning-based TFT model, indoor data is collected and preprocessed to establish the relationship between indoor structural features, dynamic features, comprehensive thermal effects, and CO2 concentration. Control signals are then generated to adjust the power of heating and ventilation equipment, responding to environmental changes in real time.
It enables intelligent optimization of heating and ventilation systems in dynamic environments, improving the system's energy efficiency and operational efficiency. It can respond to environmental changes in teaching scenarios in real time, significantly improving the balance between comfort and energy consumption.
Smart Images
Figure CN122065244A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of building environment optimization and intelligent control. Background Technology
[0002] In dynamic scenarios, such as classroom teaching in primary and secondary schools, indoor thermal comfort needs and environmental conditions change dynamically over time, significantly influenced by factors such as student movement and ventilation during breaks. Traditional heating and ventilation systems are typically designed based on steady-state environments, making it difficult to respond in real time to changes in dynamic scenarios. This limitation leads to inefficient thermal comfort management, potentially causing student discomfort and increasing energy consumption. Furthermore, balancing overall thermal effects and energy consumption in dynamic environments remains a challenge for current technological development.
[0003] Existing research on comprehensive thermal effect prediction mostly focuses on steady-state environments, employing simple regression methods or basic machine learning models (such as LSTM and random forests), which struggle to effectively capture the nonlinear relationships of multivariate interactions in dynamic scenarios. These methods suffer from insufficient accuracy and adaptability when handling multi-time-step prediction tasks, particularly lacking the ability to comprehensively model time-series characteristics in dynamic scenarios. Furthermore, traditional methods typically apply prediction results in isolation, failing to integrate them with the dynamic optimization of heating and ventilation systems, making it difficult to achieve a balance between comfort and energy-saving goals. Summary of the Invention
[0004] The purpose of this invention is to address the problem that traditional heating and ventilation equipment is usually designed based on a steady-state environment and is difficult to respond to dynamic scene changes in real time. This invention proposes a method for adjusting energy consumption based on deep learning for dynamic scene changes.
[0005] A method for dynamically adjusting energy consumption based on deep learning, the method comprising the following:
[0006] Step 1: Collect historical indoor data, including indoor structural characteristics, indoor dynamic characteristics, comprehensive thermal effects, and CO2 concentration for each time period;
[0007] Step 2, Data Preprocessing: The collected data is processed sequentially by removing outliers, time-series processing, data standardization, and selection of important features to obtain preprocessed data;
[0008] Step 3: Use the preprocessed indoor structural features and indoor dynamic features of each time period as input data for one sample, and use the corresponding preprocessed comprehensive thermal effect and CO2 concentration as output data for one sample; use the input data and output data to train the TFT model to obtain the pretrained model.
[0009] Step 4: Collect indoor structural features and indoor dynamic features during the current time period, preprocess the indoor structural features and indoor dynamic features, and input the preprocessed data into the pre-trained model to predict the corresponding comprehensive thermal effect and CO2 concentration.
[0010] Step 5: Generate corresponding control signals based on the overall thermal effect and CO2 concentration to control the power of heating and ventilation equipment respectively.
[0011] Preferably, control signals are generated based on the combined thermal effect and CO2 concentration to control the power of the heating and ventilation equipment respectively, specifically:
[0012] When the overall thermal effect is less than the preset lower limit of the overall thermal effect and the CO2 concentration is less than the lower limit of the CO2 concentration, the heating equipment is controlled to operate at the preset maximum heating power; when the overall thermal effect is greater than or equal to the preset lower limit of the overall thermal effect and less than the first threshold of the overall thermal effect, and the CO2 concentration is less than the lower limit of the CO2 concentration, the heating equipment is controlled to operate at a higher heating power; when the overall thermal effect is greater than or equal to the first threshold of the overall thermal effect and less than the second threshold of the overall thermal effect, and the CO2 concentration is less than the lower limit of the CO2 concentration, the heating equipment is controlled to operate at a medium heating power; when the overall thermal effect is greater than or equal to the second threshold of the overall thermal effect and less than the third threshold of the overall thermal effect, and the CO2 concentration is less than the lower limit of the CO2 concentration, the heating equipment is controlled to operate at the preset minimum heating power.
[0013] When the CO2 concentration is within the first CO2 concentration threshold range and the comprehensive thermal effect is greater than or equal to the third comprehensive thermal effect threshold, the ventilation equipment is controlled to operate at the preset minimum ventilation frequency; when the CO2 concentration is within the second CO2 concentration threshold range and the comprehensive thermal effect is greater than or equal to the third threshold, the ventilation equipment is controlled to operate at the preset medium ventilation frequency; when the CO2 concentration is within the third threshold range and the comprehensive thermal effect is greater than or equal to the third CO2 concentration threshold, the ventilation equipment is controlled to operate at the preset maximum ventilation frequency.
[0014] Preferably, the interior structural features include interior area, floor height, and orientation.
[0015] Preferably, indoor dynamic characteristics include indoor temperature and humidity, wind speed, number of students, and type of activity.
[0016] Preferably, the preset lower limit value of the comprehensive thermal effect is -0.3, the first threshold value of the comprehensive thermal effect is -0.2, the second threshold value of the comprehensive thermal effect is -1.0, and the third threshold value of the comprehensive thermal effect is -0.5;
[0017] The maximum heating capacity is 0.75 times the maximum heating capacity.
[0018] Medium heating capacity is 0.5 times the maximum heating capacity;
[0019] The minimum heating power is 0.25 times the maximum heating power.
[0020] Preferably, the lower limit of CO2 concentration is 450 ppm;
[0021] The first CO2 concentration threshold range is [450ppm, 600ppm].
[0022] The second CO2 concentration threshold range is [600ppm, 800ppm].
[0023] The third CO2 concentration threshold range is greater than or equal to 800 ppm.
[0024] The beneficial effects of this invention are:
[0025] This invention establishes the relationship between indoor structural features, indoor dynamic features, overall thermal effect, and CO2 concentration using a TFT model. Based on the indoor structural and dynamic features at the current time period, it predicts the corresponding overall thermal effect and CO2 concentration, and then controls the power of appropriate heating and ventilation equipment according to these results. Compared to traditional formula-based methods, this approach better reflects the complex nonlinear relationships between multiple variables in a dynamic environment and adapts to transient changes in dynamic scenarios.
[0026] This invention is applicable to indoor comfort and energy consumption optimization management in dynamic environments and can be widely used in education, office, conference and other scenarios, providing technical support for intelligent management of building environments.
[0027] This invention optimizes heating and ventilation strategies based on a TFT thermal comfort prediction model, successfully achieving a balance between comfort and energy consumption in dynamic environments. Through dynamic feedback and intelligent optimization algorithms, this invention can respond in real time to environmental changes in teaching scenarios, significantly improving the system's energy-saving effect and operational efficiency. Attached Figure Description
[0028] Figure 1 This is a flowchart of a method for regulating energy consumption for thermal comfort in a deep learning-based dynamic teaching model. Detailed Implementation
[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0030] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the scope of the invention.
[0031] Example:
[0032] A method for regulating energy consumption for thermal comfort in a dynamic teaching model based on deep learning, the method comprising the following:
[0033] Step 1: Collect historical indoor data, including indoor structural characteristics, indoor dynamic characteristics, comprehensive thermal effects, and CO2 concentration for each time period;
[0034] Step 2, Data Preprocessing: The collected data is processed sequentially by removing outliers, time-series processing, data standardization, and selection of important features to obtain preprocessed data;
[0035] Step 3: Use the preprocessed indoor structural features and indoor dynamic features of each time period as input data for one sample, and use the corresponding preprocessed comprehensive thermal effect and CO2 concentration as output data for one sample; use the input data and output data to train the TFT model to obtain the pretrained model.
[0036] Step 4: Collect indoor structural features and indoor dynamic features during the current time period, preprocess the indoor structural features and indoor dynamic features, and input the preprocessed data into the pre-trained model to predict the corresponding comprehensive thermal effect and CO2 concentration.
[0037] Step 5: Generate corresponding control signals based on the overall thermal effect and CO2 concentration to control the power of heating and ventilation equipment respectively.
[0038] Further specifying, based on the combined thermal effect and CO2 concentration, corresponding control signals are generated to control the power of heating and ventilation equipment respectively, specifically:
[0039] When the overall thermal effect is less than the preset lower limit of the overall thermal effect and the CO2 concentration is less than the lower limit of the CO2 concentration, the heating equipment is controlled to operate at the preset maximum heating power; when the overall thermal effect is greater than or equal to the preset lower limit of the overall thermal effect and less than the first threshold of the overall thermal effect, and the CO2 concentration is less than the lower limit of the CO2 concentration, the heating equipment is controlled to operate at a higher heating power; when the overall thermal effect is greater than or equal to the first threshold of the overall thermal effect and less than the second threshold of the overall thermal effect, and the CO2 concentration is less than the lower limit of the CO2 concentration, the heating equipment is controlled to operate at a medium heating power; when the overall thermal effect is greater than or equal to the second threshold of the overall thermal effect and less than the third threshold of the overall thermal effect, and the CO2 concentration is less than the lower limit of the CO2 concentration, the heating equipment is controlled to operate at the preset minimum heating power.
[0040] When the CO2 concentration is within the first CO2 concentration threshold range and the comprehensive thermal effect is greater than or equal to the third comprehensive thermal effect threshold, the ventilation equipment is controlled to operate at the preset minimum ventilation frequency; when the CO2 concentration is within the second CO2 concentration threshold range and the comprehensive thermal effect is greater than or equal to the third threshold, the ventilation equipment is controlled to operate at the preset medium ventilation frequency; when the CO2 concentration is within the third threshold range and the comprehensive thermal effect is greater than or equal to the third CO2 concentration threshold, the ventilation equipment is controlled to operate at the preset maximum ventilation frequency.
[0041] Specifically, the implementation of heating and ventilation strategies is accomplished by a central control system. This system receives thermal comfort prediction results and current environmental parameters in real time, and dynamically adjusts the operating status of heating and ventilation equipment through logic control algorithms to ensure a balance between comfort and energy consumption.
[0042] To ensure that the input data for the model is of high quality and has practical application value, the following steps are taken to process the collected data:
[0043] (1) Data collection: Key variables such as indoor and outdoor temperature and humidity, CO2 concentration, wind speed, student flow and activity type are collected in real time through high-precision sensors and monitoring equipment. In addition, students provide thermal comfort ratings using the ASHRAE seven-point scale after the course as subjective evaluation data.
[0044] (2) Data cleaning: The raw data is cleaned to remove outliers that may exist in the sensor (such as environmental noise, signal interference, etc.). Median filtering and 3σ rule are used to detect and remove outlier data points to ensure the reliability of the input data.
[0045] (3) Time series processing: The cleaned data is processed to generate time series samples containing multidimensional variables at a fixed time step (e.g., 5 minutes). Each sample includes data segments at multiple time points to capture dynamic changes.
[0046] (4) Data Standardization: To accelerate model training and improve prediction stability, all environmental variables are standardized to the [0,1] interval. The specific formula is:
[0047] ,
[0048] in, The original data, and These are the minimum and maximum values of the data, respectively.
[0049] (5) Feature selection: Using SHAP value analysis and XGBoost feature importance assessment methods, key variables (such as temperature and humidity, CO2 concentration and number of people) that contribute the most to thermal comfort are screened from multidimensional variables. This process not only optimizes the input feature set, but also reduces computational complexity.
[0050] (6) Data splitting: The preprocessed dataset is divided into a training set (80%), a validation set (10%), and a test set (10%) to ensure the generalization ability of the model.
[0051] The introduction of Temporal Fusion Transformer (TFT) provides a new technical approach for dynamic time series modeling. TFT uses a multi-head attention mechanism to acquire long-term dependencies and combines a feature selection module to dynamically identify key variables, significantly improving prediction accuracy and model interpretability in complex scenarios.
[0052] Compared with traditional methods, TFT can efficiently integrate static and dynamic characteristics, adapt to the needs of dynamic environments, and provide a reliable basis for optimizing heating and ventilation strategies.
[0053] By introducing the Temporal Fusion Transformer (TFT) model, a dynamic thermal comfort prediction model suitable for dynamic teaching scenarios in primary and secondary school classrooms in cold regions is constructed. The TFT model, by integrating static features, dynamic features, and target sequences, can accurately capture the complex temporal dependencies between indoor and outdoor environmental variables, behavioral variables, and subjective comfort evaluations, while possessing high interpretability and multi-step prediction capabilities. The application of this algorithm overcomes the limitations of traditional algorithms in long-term series modeling, not only improving prediction accuracy but also enhancing parallel computing capabilities and adaptability to complex scenarios, providing a reliable scientific basis for the optimization of heating and ventilation systems.
[0054] The core network architecture of TFT includes a multi-head attention mechanism, a feature selection module, and a time-series fusion module:
[0055] (1) Multi-head attention mechanism: used to capture the dependencies between time steps and dynamically adjust the weights to emphasize key time points.
[0056] (2) Feature selection module: Automatically selects the input features that contribute the most to thermal comfort prediction, reducing the interference of invalid information.
[0057] (3) Time series fusion module: integrates static and dynamic features to enable the model to dynamically adapt to complex scenarios.
[0058] To further define, static characteristics include interior area, floor height, and orientation.
[0059] Specifically, static features are used to describe fixed attributes of the built environment, such as classroom area, floor height, orientation, and regional climate conditions.
[0060] Further specifying, dynamic characteristics include indoor temperature and humidity, wind speed, number of students, and type of activity.
[0061] Specifically, dynamic characteristics are divided into two categories: known variables (such as class schedules, ventilation time and outdoor weather forecast) and unknown variables (such as temperature and humidity, CO2 concentration, wind speed, number of students and type of activity).
[0062] Further restrictions are set, with a preset lower limit for thermal comfort of -0.3, a first threshold for thermal comfort of -0.2, a second threshold for thermal comfort of -1.0, and a third threshold for thermal comfort of -0.5.
[0063] The maximum heating capacity is 0.75 times the maximum heating capacity.
[0064] Medium heating capacity is 0.5 times the maximum heating capacity;
[0065] The minimum heating power is 0.25 times the maximum heating power.
[0066] The lower limit for CO2 concentration is further specified as 450 ppm;
[0067] The first CO2 concentration threshold range is [450ppm, 600ppm].
[0068] The second CO2 concentration threshold range is [600ppm, 800ppm].
[0069] The third CO2 concentration threshold range is greater than or equal to 800 ppm.
[0070] While the invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed without departing from the spirit and scope of the invention as defined by the appended claims. It should be understood that different dependent claims and features described herein can be combined in ways different from those described in the original claims. It is also understood that features described in conjunction with individual embodiments can be used in other described embodiments.
Claims
1. A method for dynamically adjusting energy consumption based on deep learning, characterized in that, The method includes the following: Step 1: Collect historical indoor data, including indoor structural characteristics, indoor dynamic characteristics, comprehensive thermal effects, and CO2 concentration for each time period; Step 2, Data Preprocessing: The collected data is processed sequentially by removing outliers, time-series processing, data standardization, and selection of important features to obtain preprocessed data; Step 3: Use the preprocessed indoor structural features and indoor dynamic features of each time period as input data for one sample, and use the corresponding preprocessed comprehensive thermal effect and CO2 concentration as output data for one sample; use the input data and output data to train the TFT model to obtain the pretrained model. Step 4: Collect indoor structural features and indoor dynamic features during the current time period, preprocess the indoor structural features and indoor dynamic features, and input the preprocessed data into the pre-trained model to predict the corresponding comprehensive thermal effect and CO2 concentration. Step 5: Generate corresponding control signals based on the overall thermal effect and CO2 concentration to control the power of heating and ventilation equipment respectively.
2. The method for dynamically adjusting energy consumption based on deep learning according to claim 1, characterized in that, Based on the overall thermal effect and CO2 concentration, corresponding control signals are generated to control the power of heating and ventilation equipment respectively, as follows: When the overall thermal effect is less than the preset lower limit of the overall thermal effect and the CO2 concentration is less than the lower limit of the CO2 concentration, the heating equipment is controlled to operate at the preset maximum heating power; when the overall thermal effect is greater than or equal to the preset lower limit of the overall thermal effect and less than the first threshold of the overall thermal effect, and the CO2 concentration is less than the lower limit of the CO2 concentration, the heating equipment is controlled to operate at a higher heating power; when the overall thermal effect is greater than or equal to the first threshold of the overall thermal effect and less than the second threshold of the overall thermal effect, and the CO2 concentration is less than the lower limit of the CO2 concentration, the heating equipment is controlled to operate at a medium heating power; when the overall thermal effect is greater than or equal to the second threshold of the overall thermal effect and less than the third threshold of the overall thermal effect, and the CO2 concentration is less than the lower limit of the CO2 concentration, the heating equipment is controlled to operate at the preset minimum heating power. When the CO2 concentration is within the first CO2 concentration threshold range and the comprehensive thermal effect is greater than or equal to the third comprehensive thermal effect threshold, the ventilation equipment is controlled to operate at the preset minimum ventilation frequency; when the CO2 concentration is within the second CO2 concentration threshold range and the comprehensive thermal effect is greater than or equal to the third threshold, the ventilation equipment is controlled to operate at the preset medium ventilation frequency; when the CO2 concentration is within the third threshold range and the comprehensive thermal effect is greater than or equal to the third CO2 concentration threshold, the ventilation equipment is controlled to operate at the preset maximum ventilation frequency.
3. The method for dynamically adjusting energy consumption based on deep learning according to claim 1, characterized in that, Interior structural features include interior area, floor height, and orientation.
4. The method for adjusting energy consumption in a deep learning-based dynamic teaching model according to claim 1, characterized in that, Indoor dynamic characteristics include indoor temperature and humidity, wind speed, number of students, and type of activity.
5. The method for dynamically adjusting energy consumption based on deep learning according to claim 1, characterized in that, The preset lower limit of the comprehensive thermal effect is -0.3, the first threshold of the comprehensive thermal effect is -0.2, the second threshold of the comprehensive thermal effect is -1.0, and the third threshold of the comprehensive thermal effect is -0.
5. The maximum heating capacity is 0.75 times the maximum heating capacity. Medium heating capacity is 0.5 times the maximum heating capacity; The minimum heating power is 0.25 times the maximum heating power.
6. The method for regulating energy consumption for dynamic scene thermal comfort based on deep learning according to claim 1, characterized in that, The lower limit for CO2 concentration is 450 ppm; The first CO2 concentration threshold range is [450ppm, 600ppm]. The second CO2 concentration threshold range is [600ppm, 800ppm]. The third CO2 concentration threshold range is greater than or equal to 800 ppm.