Smart temperature control method and system for a home scene

Abstract

The application discloses a smart temperature control adjustment method and system for a home scene. The adjustment method comprises the following steps: collecting multi-source data, preprocessing the collected multi-source data; based on the preprocessed multi-source data, extracting a feature vector capable of uniquely representing different home scenes; weighting and fusing all the extracted features according to their contribution to scene recognition to generate a unified scene feature vector; using a random forest algorithm as a scene classifier to identify different home scenes in real time, and giving corresponding weight factors of comfort, energy saving and response speed according to the identified home scenes; according to the given weight factors, sending a temperature control strategy to a room temperature control device to adjust its operating state; based on real-time feedback and historical operation data of a user, updating the weight factors of the comfort, energy saving and response speed targets, thereby realizing scene adaptive control of the temperature control system and significantly improving the smart home temperature control experience.

Description

Technical Field

[0001] This invention relates to the field of smart home control technology, and in particular to a smart temperature control method and system for home use. Background Technology

[0002] With the rapid development of smart home technology, intelligent temperature control systems have become standard equipment in high-end residences, office areas and other places.

[0003] However, most existing intelligent temperature control systems remain at the static control stage, such as manual preset and timed control, lacking scene recognition capabilities and making it difficult to dynamically adjust to the differentiated needs of different scenarios. In addition, temperature control systems generally need to meet multiple objectives, such as user comfort, energy consumption, and system response speed, but existing systems usually use single-objective optimization or simple weighted summation methods for regulation, making it difficult to find the optimal balance point among multiple objectives.

[0004] Therefore, there is an urgent need for a home scene coordination and optimization method for intelligent temperature control systems, which can intelligently adjust and control the system by sensing and identifying the home scene to solve the problems existing in the current temperature control system. Summary of the Invention

[0005] This application provides an intelligent temperature control method and system for home scenarios, which can realize scene-adaptive control of the temperature control system, effectively balance user comfort, energy saving and system response speed, and significantly improve the smart home temperature control experience.

[0006] The first aspect of this application provides an intelligent temperature control method for home use, comprising the following steps: S10: Collect multi-source data and preprocess the collected multi-source data, including outlier detection, missing value imputation and data normalization. S20, based on preprocessed multi-source data, extracts feature vectors that can uniquely represent different home scenarios, including time features, environmental features, user behavior features, device status features, and historical habit features; S30, all extracted features are weighted and fused according to their contribution to scene recognition to generate a unified scene feature vector; The S40 uses the random forest algorithm as a scene classifier to identify different home scenes in real time and give corresponding weight factors for comfort, energy saving and response speed based on the identified home scenes. S50 sends temperature control strategies to the room's temperature control devices based on the given weighting factors, adjusting their operating status.

[0007] In one possible implementation, the intelligent temperature control method for home scenarios further includes step S60, which updates the weighting factors of objectives such as comfort, energy saving, and response speed based on real-time user feedback and historical operation data.

[0008] A second aspect of this application provides an intelligent temperature control system for home environments, used to implement the intelligent temperature control method for home environments as described above, wherein the intelligent temperature control system includes: The data acquisition unit is used to collect multi-source data and preprocess the collected multi-source data, including outlier detection, missing value imputation and data normalization. The feature extraction unit is used to extract feature vectors that can uniquely represent different home scenarios based on preprocessed multi-source data, including time features, environmental features, user behavior features, device status features, and historical habit features. The feature-level data fusion unit is used to weight and fuse all extracted features according to their contribution to scene recognition, and generate a unified scene feature vector. The scene classification and weighting unit is used to use the random forest algorithm as a scene classifier to identify different home scenes in real time, and to give corresponding weight factors for comfort, energy saving and response speed based on the identified home scenes. The strategy execution unit is used to send temperature control strategies to the room temperature control equipment according to the given weight factors, and adjust the operating status of the air conditioning / underfloor heating and fresh air system. The feedback optimization unit is used to update the weighting factors of objectives such as comfort, energy saving, and response speed based on real-time user feedback and historical operation data.

[0009] Beneficial effects: Compared with existing technologies, the intelligent temperature control method and system for home scenarios provided in this application take scene perception as the core and constructs an intelligent, personalized and efficient temperature control system. It can accurately identify typical home scenarios such as waking up in the morning, returning home from outside, sleeping, leaving home, and emergency adjustment. It can also allocate weight factors of each optimization goal according to scene priority, generate intelligent temperature control strategies, and update the weight factors of each optimization goal according to real-time user feedback and historical operation data. This enables scene adaptive control of the temperature control system, effectively balancing user comfort, energy saving and system response speed, and significantly improving the temperature control experience of smart homes.

[0010] These and other objects, features and advantages of the present invention will become fully apparent from the following detailed description. Attached Figure Description

[0011] Figure 1 A flowchart illustrating the intelligent temperature control method for home use according to this application is shown. Detailed Implementation

[0012] The following description is intended to disclose the present invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art. The basic principles of the invention defined in the following description can be applied to other embodiments, modifications, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the invention.

[0013] Those skilled in the art should understand that, in the disclosure of this specification, the terms "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the above terms should not be construed as limiting the present invention.

[0014] Unless otherwise specified, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains. It will also be understood that terms, such as those defined in commonly used dictionaries, shall be interpreted as having a meaning consistent with their meaning in the context of the relevant art and shall not be interpreted as having an idealized or overly formal meaning unless expressly so defined herein.

[0015] It is understood that the term "a" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple, and the term "a" should not be understood as a limitation on the number.

[0016] refer to Figure 1 The first aspect of this application provides an intelligent temperature control method for home use, comprising the following steps: S10, collect multi-source data and preprocess the collected multi-source data, including outlier detection, missing value imputation, and data normalization. The multi-source data includes indoor environmental data, user behavior data, equipment status data, and external environmental data. The system uses temperature sensors, humidity sensors, air quality sensors, and light sensors to collect data on indoor temperature, indoor humidity, PM2.5 concentration, gas concentration (when a gas sensor is integrated), light intensity, outdoor temperature, outdoor humidity, and weather to distinguish between scenarios such as returning home and leaving home. By using human presence sensors, door and window status sensors, smart door locks, and smart wearable devices, user behavior data such as whether the user is at home, the room they are in, the intensity of their activity, the status of doors and windows, the status of smart door locks, their sleep status, and their heart rate can be determined to distinguish between scenarios such as returning home from outside and sleeping. The device status acquisition unit determines device status data such as air conditioner switch status, air conditioner mode, air conditioner fan speed, floor heating switch status, fresh air system status, and set temperature, in order to distinguish between scenarios such as leaving home and being at home. External environmental data such as weather forecasts, outdoor temperature and humidity, air quality, and electricity prices are obtained through external data interfaces.

[0017] Outlier detection uses 3 The principle is to identify abnormal data and remove invalid data. Principles are empirical rules in statistics used to determine whether data is normally distributed, often used to identify outliers; missing value imputation uses linear interpolation or the average of historical data; data normalization uses min-max normalization to map data of different dimensions to the [0, 1] interval, eliminating the influence of dimensional differences on multi-source data. Normalized data can be represented as: ; in, x It is the raw data. It is the maximum value of this feature. It is the minimum value of this feature; S20, based on preprocessed multi-source data, extracts feature vectors that uniquely represent different home scenarios, including time features, environmental features, user behavior features, device status features, and historical habit features. Preprocessing and feature extraction can be performed every 5 seconds. The time features extract six dimensions of data, including hour, minute, day of the week, whether it is a weekday, whether it is a holiday, and season, to distinguish scenarios with strong time regularity such as waking up in the morning and going to sleep; A total of eight dimensions of environmental features were extracted, including indoor temperature, indoor humidity, PM2.5 concentration, gas concentration, light intensity, outdoor temperature, outdoor humidity, and weather, to distinguish between scenarios such as returning home and leaving home. A total of 7 dimensions of user behavior characteristics were extracted, including whether the user was at home, the room they were in, the intensity of their activity, the status of doors and windows, the status of smart locks, their sleep status, and their heart rate, in order to distinguish between scenarios such as returning home from outside and sleeping. The device status features extract six dimensions of data, including air conditioner on / off status, air conditioner mode, air conditioner fan speed, floor heating on / off status, fresh air system status, and set temperature, to distinguish between scenarios such as leaving home and being at home. Historical habit features are extracted into two dimensions, including the probability of users being at home during that time period and the average set temperature, to address special situations such as users staying up late on weekends.

[0018] All features combined total 29 dimensions; S30 employs a weighted feature fusion method, which weights and fuses all features extracted from different sensors according to their contribution to scene recognition, generating a unified scene feature vector. Specifically: The information gain method is used to calculate the contribution of each feature to scene recognition. The formula is as follows: ; in, It is a feature For the target scenario T Information gain It is the entropy of the target scene. Given features Conditional entropy of the target scene at that time; Each feature is multiplied by its corresponding weight to generate the final fused feature vector. ,in, It is the first i The weights of each feature, , It is the first i The normalized values ​​of each feature.

[0019] This method requires less computation than data-level fusion, making it more suitable for edge devices. Compared to decision-level fusion, it retains more original information, has higher recognition accuracy, and can flexibly adjust the weights of different features, thus adapting to different users' home environments. S40 uses the random forest algorithm as a scene classifier to identify different home scenes in real time, and assigns corresponding weight factors for comfort, energy saving and response speed based on the identified home scenes. As a result, the overall recognition accuracy of this method on the test set can reach 96.8%. S50 sends temperature control strategies to the room's temperature control devices based on the given weighting factors, adjusts their operating status, and realizes scene-adaptive control of the temperature control system. This effectively balances user comfort, energy saving, and system response speed, significantly improving the temperature control experience of smart homes. The temperature control devices include air conditioners / underfloor heating, fresh air systems, etc.

[0020] In some embodiments, in step S40, the use of a random forest algorithm as a scene classifier to identify different home scenes in real time includes: First, fuse the feature vectors. It has 29 dimensions in total, including features such as time, environment, user behavior, device status, and historical habits, corresponding to the aforementioned 29 features; given a scene label, These correspond to five typical scenarios: waking up in the morning, returning home from an outing, sleeping, leaving home, and emergency adjustments; given a dataset, , N To represent the total number of samples, this application sets in one embodiment... N =2×10 6 Given the parameters of the random forest algorithm, the number of decision trees k =100, Random Feature Number of a Single Tree m =5, maximum depth h =12; During training, samples are first randomly drawn with replacement from the training set D to generate 100 subsets. S k Then, for each subset, select randomly. m =5 features, ,in Indicates the first t Tree candidate feature set, j 1~ j 5 represents the sequence number of the 5 features randomly selected from each subset. This represents the 5 selected features; then the information gain of each feature is calculated. ,in, Representation of features Information gain It is the prior entropy of Y. It is given Given the posterior entropy of Y, select the optimal splitting feature. The samples are split according to a threshold, and left and right subtrees are recursively generated until a depth is reached. h Stop when =12, and generate leaf node labels. ,in It outputs the probability. This means selecting the category with the highest probability within a node as the prediction output for that leaf node; Finally, the random forest is generated through ensemble: , Indicates the first t A trained independent decision tree, Input implementation features Each tree outputs a prediction. The final scenario is output through majority voting. If the highest probability If the scenario is correct, it is considered a transitional scenario, and the corresponding scenario is output; otherwise, it is determined to be a transitional scenario, and the current control strategy remains unchanged. The overall recognition accuracy of the technical solution in this application on the test set reaches 96.8%, and the specific recognition accuracy of each scenario is as follows: leaving home scenario—99.2%, emergency scenario—98.5%, returning home scenario—97.3%, sleep scenario—96.1%, and waking up in the morning scenario—94.7%.

[0021] The priority of optimization targets varies depending on the scenario. For example, the sleep scenario requires the highest level of comfort, the emergency scenario requires the highest response speed, and the away-from-home scenario requires the highest level of energy saving.

[0022] Therefore, in some embodiments, in step S40, the home scenarios include a morning waking-up scenario, a homecoming scenario, a sleep scenario, a leaving-home scenario, and an emergency scenario, wherein: The weighting factors for the morning wake-up scenario are: comfort weight 0.6, energy saving weight 0.3, and response speed weight 0.1. The weighting factors for the outing and returning home scenarios are as follows: comfort weight 0.4, energy saving weight 0.1, and response speed weight 0.5. The weighting factors for sleep scenarios are as follows: comfort weight 0.7, energy saving weight 0.2, and response speed weight 0.1. The weighting factors for the "away from home" scenario are: comfort weight 0.1, energy saving weight 0.8, and response speed weight 0.1. The weighting factors for emergency scenarios are: comfort weight 0, energy saving weight 0.1, and response speed weight 0.9.

[0023] In some embodiments, the intelligent temperature control method for home scenarios further includes step S60, which updates the weighting factors of objectives such as comfort, energy saving, and response speed based on real-time user feedback and historical operation data.

[0024] Specifically, the Q-learning algorithm is used to learn user preferences, and user operation feedback is used as a reward signal to continuously update the user's preference weights for each optimization goal; First, given the state space ,in These correspond to the initial weighting factors for comfort, energy efficiency, and response speed, respectively. The step size changes by 0.05, and there are a total of 100 state variables; Secondly, given the action space ,in , There are a total of 9 action variables; Then the reward function: ; Given the reward value of the selected action, update the Q value based on the reward value:

[0025] in The initial value is 0. It's a state. It is the learning rate. In one embodiment, the learning rate... It is 0.1. It is a discount factor. In one embodiment, the discount factor... It is 0.9. Indicates the greatest future Q value. It is to perform an action The new state after that. It's a new state All possible actions below; Finally passed Q Value feedback updates the weight factor.

[0026] A second aspect of this application provides an intelligent temperature control system for home environments, used to implement the intelligent temperature control method for home environments as described above, wherein the intelligent temperature control system includes: The data acquisition unit is used to collect multi-source data and preprocess the collected multi-source data, including outlier detection, missing value imputation and data normalization. The feature extraction unit is used to extract feature vectors that can uniquely represent different home scenarios based on preprocessed multi-source data, including time features, environmental features, user behavior features, device status features, and historical habit features. The feature-level data fusion unit is used to weight and fuse all extracted features according to their contribution to scene recognition, and generate a unified scene feature vector. The scene classification and weighting unit is used to use the random forest algorithm as a scene classifier to identify different home scenes in real time, and to give corresponding weight factors for comfort, energy saving and response speed based on the identified home scenes. The strategy execution unit is used to send temperature control strategies to the room temperature control equipment according to the given weight factors, and adjust the operating status of the air conditioning / underfloor heating and fresh air system. The feedback optimization unit is used to update the weighting factors of objectives such as comfort, energy saving, and response speed based on real-time user feedback and historical operation data.

[0027] More specifically: Take the scenario of waking up in the morning as an example: 1. The smart wearable device detected that the user has transitioned from deep sleep to wakefulness, and the human infrared sensor detected that the user is active in the bedroom. At this time, the initial indoor temperature is 21℃ and the outdoor temperature is 15℃. 2. Preprocess the data from smart devices, then extract features such as time and user sleep status to identify the current scenario as a morning wake-up. 3. Apply the multi-objective weighting factors corresponding to the morning wake-up scenario: comfort 0.6, energy saving 0.3, response speed 0.1; 4. Generate temperature control execution strategy: Set the bedroom temperature to 24℃, air conditioner heating mode, low fan speed, and fresh air volume to 50m³ / h; the system sends control commands to the bedroom air conditioner, and the air conditioner starts running; 5. The feedback optimization unit records user operations. If the user does not modify the temperature control execution strategy, a +10 reward is given, and the current weight factor remains unchanged.

[0028] Results: The bedroom temperature reached 24°C within 18 minutes, with a PMV value of 0.12 (very comfortable), and the energy consumption was 0.32 kWh per hour, meeting the user's comfort needs after waking up in the morning while also being energy-efficient.

[0029] Taking the scenario of returning home from a trip as an example: 1. When a user enters the home through the smart door lock, the human infrared sensor detects that the user has entered the living room. Initial indoor temperature: 16℃, outdoor temperature: 12℃. 2. Preprocess the data from smart devices, then extract features such as smart door locks and human presence signals to identify the current scenario as returning home from outside; 3. Apply the following multi-objective weighting factors for the outing and returning home scenario: comfort 0.5, energy saving 0.4, response speed 0.1; 4. Generate temperature control execution strategy: Set the bedroom temperature to 23℃, set the air conditioner to heating mode, high fan speed, and turn on the underfloor heating auxiliary heating; the system sends control commands to the living room air conditioner and underfloor heating system; when the indoor temperature reaches 23℃, the system automatically switches to energy-saving mode, the air conditioner switches to low fan speed, and the underfloor heating maintains low power operation. 5. The feedback optimization unit records user actions. If the user makes a minor change to the set temperature, a +5 reward is given, and the action is executed. The weighting factors were updated to comfort 0.55, energy saving 0.35, and response speed 0.1.

[0030] Results: The living room temperature reached 23°C within 12 minutes, with a PMV value of 0.08 (very comfortable). The total energy consumption in 30 minutes was 0.45 kWh, achieving the goal of rapid heating while avoiding unnecessary energy waste.

[0031] Take an emergency scenario as an example: 1. The kitchen gas sensor detected that the gas concentration had reached the alarm threshold, and the human infrared sensor detected that the user was in the living room; 2. When the gas concentration in the kitchen is detected to have reached a threshold by the gas sensor, an emergency adjustment scenario is identified, and the response speed is given the highest priority. 3. Apply the weighting factors corresponding to the emergency adjustment scenario: comfort 0, energy saving 0.1, response speed 0.9; 4. Generate temperature control execution strategy: Turn off all heating / cooling equipment, turn on the fresh air system at maximum airflow, and turn on the kitchen exhaust fan; The system sends control commands to all temperature control equipment, fresh air system and exhaust fan, and at the same time sends gas leak alarm information to users through mobile APP and dials the preset emergency contact number. When the gas concentration drops below the safe value, the system automatically switches to ventilation mode and ventilates continuously for 30 minutes. 5. The feedback optimization unit records user operations. If the user does not modify the strategy, a +10 reward is given, and the current weight factor remains unchanged.

[0032] Operation results: The system completed the control operation of all equipment within 2 seconds, and the gas concentration dropped to a safe level within 8 minutes, effectively preventing gas poisoning and explosion accidents and protecting the life and property safety of users.

[0033] Taking sleep scenarios as an example: 1. The smart wearable device detects that the user has transitioned from deep sleep to wakefulness, and the light sensor detects that the lights are off. The initial indoor temperature at this time is 20℃. 2. Preprocess the data from smart devices, then extract features such as time and user sleep status to identify the current scenario as sleep. 3. Apply the multi-objective weighting factors corresponding to the sleep scenario: comfort 0.7, energy saving 0.2, response speed 0.1; 4. Generate temperature control execution strategy: Set the bedroom temperature to 24℃, air conditioner heating mode, medium fan speed, and fresh air volume to 40m³ / h; the system sends control commands to the bedroom air conditioner, and the air conditioner starts running; when the indoor temperature reaches 24℃, the system automatically switches to energy-saving mode, and the air conditioner switches to low fan speed. 5. The feedback optimization unit records user operations. If the user does not modify the temperature control execution strategy, a +10 reward is given, and the current weight factor remains unchanged.

[0034] Results: The bedroom temperature reached 24°C within 20 minutes, the PMV value was 0.08 (very comfortable), and the energy consumption was 0.32 kWh per hour, meeting the user's comfort needs after sleep while also being energy-efficient.

[0035] Take the scenario of leaving home as an example: 1. The user leaves home by closing the smart door lock. The human infrared sensor does not detect the user in the home. The initial indoor temperature at this time is 20℃. 2. Preprocess the data from smart devices, then extract features such as time and smart lock status to identify the current scenario as being away from home; 3. Apply multi-objective weighting factors corresponding to the away-from-home scenario: comfort 0.1, energy saving 0.8, response speed 0.1; 4. Generate temperature control execution strategy: Turn off equipment such as air conditioners / underfloor heating, fresh air systems, and televisions; keep only equipment such as refrigerators running; 5. The feedback optimization unit records user operations. If the user does not modify the temperature control execution strategy, a +10 reward is given, and the current weight factor remains unchanged.

[0036] Results: The relevant smart devices stopped operating 10 minutes after the user left home, with only the refrigerator and other equipment remaining running, which greatly improved energy efficiency.

[0037] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the invention. The advantages of the present invention have been fully and effectively realized. The functional and structural principles of the present invention have been demonstrated and explained in the embodiments; any variations or modifications can be made to the implementation of the present invention without departing from these principles.

Claims

1. A smart temperature control method for home use, characterized in that, Includes the following steps: S10: Collect multi-source data and preprocess the collected multi-source data, including outlier detection, missing value imputation and data normalization. S20, based on preprocessed multi-source data, extracts feature vectors that can uniquely represent different home scenarios, including time features, environmental features, user behavior features, device status features, and historical habit features; S30, all extracted features are weighted and fused according to their contribution to scene recognition to generate a unified scene feature vector; The S40 uses the random forest algorithm as a scene classifier to identify different home scenes in real time and give corresponding weight factors for comfort, energy saving and response speed based on the identified home scenes. S50 sends temperature control strategies to the room's temperature control devices based on the given weighting factors, adjusting their operating status.

2. The intelligent temperature control method for home use as described in claim 1, characterized in that, In step S10, outlier detection uses 3 The principle is to identify abnormal data, remove invalid data, fill missing values ​​using linear interpolation or the average value of historical data, and normalize data using the minimum-maximum normalization method to map data of different dimensions to the [0, 1] interval, thereby eliminating the impact of dimension differences on multi-source data.

3. The intelligent temperature control method for home use as described in claim 1, characterized in that, In step S20, a total of 6 dimensions of time features are extracted, including hour, minute, day of the week, whether it is a weekday, whether it is a holiday, and season; A total of eight dimensions of environmental characteristics were extracted, including indoor temperature, indoor humidity, PM2.5 concentration, gas concentration, light intensity, outdoor temperature, outdoor humidity, and weather. A total of 7 dimensions of user behavior characteristics were extracted, including whether the user was at home, the room they were in, the intensity of their activity, the status of doors and windows, the status of smart locks, their sleep status, and their heart rate. The equipment status features are extracted in six dimensions, including air conditioner on / off status, air conditioner mode, air conditioner fan speed, floor heating on / off status, fresh air system status, and set temperature. Historical habit features were extracted into two dimensions: the probability of users being at home during that time period and the average set temperature.

4. The intelligent temperature control method for home use as described in claim 3, characterized in that, In step S30, the information gain method is used to calculate the contribution of each feature to scene recognition. The calculation formula is as follows: ； in, It is a feature For the target scenario T Information gain It is the entropy of the target scene. Given features Conditional entropy of the target scene at that time; Each feature is multiplied by its corresponding weight to generate the final fused feature vector. , It is the first i The weights of each feature , It is the first i The normalized values ​​of each feature.

5. The intelligent temperature control method for home use as described in claim 4, characterized in that, In step S40, the use of the random forest algorithm as a scene classifier to identify different home scenes in real time includes: given a dataset, , N Given the total number of samples; given the parameters of the random forest algorithm, the number of decision trees. k =100, Random Feature Number of a Single Tree m =5, maximum depth h =12; During training, samples are first randomly drawn with replacement from the training set D to generate 100 subsets. S k Then, for each subset, select randomly. m =5 features, ,in Indicates the first t Tree candidate feature set, j 1~ j 5 represents the sequence numbers of the 5 features randomly selected from each subset. This represents the 5 selected features; then the information gain of each feature is calculated. ,in, Representation of features Information gain It is the prior entropy of Y. It is given Given the posterior entropy of Y, select the optimal splitting feature. The samples are split according to a threshold, and left and right subtrees are recursively generated until a depth is reached. h Stop when =12, and generate leaf node labels. ,in It outputs the probability. This means selecting the category with the highest probability within a node as the prediction output for that leaf node; Finally, the random forest is generated through ensemble: , Indicates the first t A trained independent decision tree, Input implementation features Each tree outputs a prediction. The final scenario is output through majority voting. If the highest probability If the condition is met, the corresponding scene will be output; otherwise, it will be considered a transitional scene.

6. The intelligent temperature control method for home scenarios as described in claim 5, characterized in that, In step S40, the home scenarios include morning wake-up scenarios, returning home scenarios, sleep scenarios, leaving home scenarios, and emergency scenarios, wherein: The weighting factors for the morning wake-up scenario are: comfort weight 0.6, energy saving weight 0.3, and response speed weight 0.

1. The weighting factors for the outing and returning home scenarios are as follows: comfort weight 0.4, energy saving weight 0.1, and response speed weight 0.

5. The weighting factors for sleep scenarios are as follows: comfort weight 0.7, energy saving weight 0.2, and response speed weight 0.

1. The weighting factors for the "away from home" scenario are: comfort weight 0.1, energy saving weight 0.8, and response speed weight 0.

1. The weighting factors for emergency scenarios are: comfort weight 0, energy saving weight 0.1, and response speed weight 0.

9.

7. The intelligent temperature control method for home use as described in any one of claims 1 to 6, characterized in that, It also includes step S60, which updates the weighting factors of objectives such as comfort, energy saving and response speed based on real-time user feedback and historical operation data.

8. The intelligent temperature control method for home use as described in claim 7, characterized in that, In step S60, the Q-learning algorithm is used to learn user preferences, and the user's operation feedback is used as a reward signal to continuously update the user's preference weights for each optimization goal; First, a state space is given. ,in These correspond to the initial weighting factors for comfort, energy efficiency, and response speed, respectively. The step size changes by 0.05, and there are a total of 100 state variables; Secondly, given the action space ,in , There are a total of 9 action variables; Then the reward function: ； Given the reward value of the selected action, update the Q value based on the reward value: in The initial value is 0. It's a state. It's the learning rate. It is a discount factor. Indicates the greatest future Q value, It is to perform an action The new state after that, It's a new state All possible actions below; Finally passed Q Value feedback updates the weighting factor.

9. The intelligent temperature control method for home scenarios as described in claim 8, characterized in that, Learning rate The discount factor is 0.

1. It is 0.

9.

10. An intelligent temperature control system for home use, used to implement the intelligent temperature control method for home use as described in any one of claims 7 to 9, characterized in that, include: The data acquisition unit is used to collect multi-source data and preprocess the collected multi-source data, including outlier detection, missing value imputation and data normalization. The feature extraction unit is used to extract feature vectors that can uniquely represent different home scenarios based on preprocessed multi-source data, including time features, environmental features, user behavior features, device status features, and historical habit features. The feature-level data fusion unit is used to weight and fuse all extracted features according to their contribution to scene recognition, and generate a unified scene feature vector. The scene classification and weighting unit is used to identify different home scenes in real time using the random forest algorithm as a scene classifier, and to give corresponding weight factors for comfort, energy saving and response speed based on the identified home scenes. The strategy execution unit is used to send temperature control strategies to the room temperature control equipment according to the given weight factors, and adjust the operating status of the air conditioning / underfloor heating and fresh air system. The feedback optimization unit is used to update the weighting factors of objectives such as comfort, energy saving, and response speed based on real-time user feedback and historical operation data.

Images