Split air conditioner self-adaptive control system and method based on human body perception

Abstract

The invention discloses a split air conditioner self-adaptive control system and method based on human body perception, and relates to the technical field of intelligent household electrical appliance control, and the method comprises the following steps: extracting time sequence characteristics from brain wave signals by using a graph convolutional network model, and carrying out weighted fusion on nose tip and forehead temperature difference data through a space attention mechanism to obtain a heat demand level; constructing a causal rule chain through a self-interpretation strategy engine, and visually displaying the causal rule chain through an interactive interface of the split air conditioner to obtain a weight adjustment instruction of the causal rule chain; the rule priority is dynamically modified through the slider, the dynamic weighting coefficient in the self-adaptive control strategy generator is updated, and the target temperature and the air deflector angle are calculated; by introducing a graph convolutional network and a space attention mechanism, efficient fusion modeling of brain wave signals and face multi-point temperature data is realized, and the accuracy and the stability of heat demand level identification are remarkably improved; and a more intelligent, accurate and interactive self-adaptive control scheme of the split air conditioner is provided.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent home appliance control, and in particular to a split air conditioner adaptive control system and method based on human body perception. Background Art

[0002] With the rapid development of intelligent air conditioning technology, users' demand for indoor thermal comfort is increasing. Traditional split-type air conditioners rely primarily on the difference between the set temperature and the room temperature for control, lacking a dynamic response mechanism to the human body's actual thermal perception. In recent years, human-computer interaction technologies based on physiological signals have been gradually applied to the field of thermal comfort control. For example, they use indicators such as galvanic skin response and heart rate variability to infer users' thermal perception, and combine them with machine learning models to implement personalized control strategies. The development of multimodal data fusion technology also makes it possible to more accurately identify thermal needs. For example, facial infrared thermal imaging can be combined with environmental parameters to enhance the system's perception capabilities.

[0003] Existing technologies still have limitations in efficiently integrating multi-source physiological signals into modeling and generating control strategies. Most methods utilize only a single modality input (such as brain waves or facial temperature), failing to fully exploit the spatiotemporal correlations between multi-channel sensory information. This results in limited accuracy in identifying thermal demand levels, restricting the practicality and user experience of adaptive control systems. In particular, it is difficult to achieve stable and personalized thermal comfort regulation in complex and dynamic environments. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a split air conditioner adaptive control method based on human perception to solve the problem of low accuracy in thermal demand recognition.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a split air conditioner adaptive control method based on human body perception, which includes collecting a user's brain wave signals, facial temperature gradient distribution data, and nose tip and forehead temperature difference data; A graph convolutional network model is used to extract temporal features from EEG signals, and the temperature difference data between the nose tip and forehead are weightedly fused through the spatial attention mechanism to obtain the thermal demand level. User historical operation data is collected, the thermal demand level is input into the neural feedback reinforcement learning model, and initial control parameters are generated. A causal rule chain is constructed through a self-explanatory strategy engine, and the causal rule chain is visualized through the interactive interface of the split air conditioner to obtain the weight adjustment instructions of the causal rule chain. The rule priority is dynamically modified through the slider, the dynamic weighting coefficient in the adaptive control strategy generator is updated, and the target temperature and air guide plate angle are calculated. The split air conditioner is adaptively controlled by controlling the compressor frequency, air guide plate angle and radiation panel power.

[0007] As a preferred solution of the split air conditioner adaptive control method based on human perception of the present invention, wherein: the user's brain wave signal, facial temperature gradient distribution data and nose tip and forehead temperature difference data are collected, the specific steps are as follows: The millimeter-wave radar scans the user's head area, collects the original electromagnetic reflection signal, and performs noise reduction processing to obtain the brain wave signal; The user's facial thermal image is captured using an infrared thermal imager. Based on the facial key point detection algorithm, three temperature monitoring points (nose tip, forehead, and face) are located. The temperature distribution matrix is ​​calculated and the facial temperature gradient distribution data is generated by taking the temperature differences between adjacent temperature monitoring points. The dynamic temperature difference index is calculated based on the temperature difference between the nose tip area and the forehead area to generate the nose tip and forehead temperature difference data.

[0008] As a preferred solution of the split air conditioning adaptive control method based on human perception described in the present invention, wherein: the graph convolutional network model is used to extract time series features from brain wave signals, and the nose tip and forehead temperature difference data are weightedly fused through the spatial attention mechanism to obtain the heat demand level. The specific steps are as follows: The EEG signal is divided into multiple time series segments by time window, the energy feature of each time window is used as a node, and the signal correlation of adjacent time windows is used as the edge weight to construct the EEG time series graph; Based on the facial temperature gradient distribution data, the temperature monitoring points in the nose tip area, forehead area and facial area are used as nodes, and the absolute value of the temperature difference between adjacent temperature monitoring points is used as the edge weight to construct the facial temperature space graph; Graph convolution operations are performed on the EEG time series graph and the facial temperature spatial graph respectively to extract the temporal dynamic features of the EEG signal and the global spatial correlation features of the temperature distribution; The global spatial correlation features of temperature distribution are dynamically weighted through the spatial attention mechanism, and are cross-modally spliced ​​and fused with the temporal dynamic features of EEG signals, and the thermal demand level is generated through the threshold segmentation algorithm.

[0009] As a preferred solution of the split air conditioner adaptive control method based on human perception described in the present invention, wherein: collecting user historical operation data, inputting the thermal demand level into the neural feedback reinforcement learning model, and generating initial control parameters, the specific steps are as follows: Collect historical user operation data and combine it with the current heat demand level to form a multi-dimensional state vector; The multi-dimensional state vector is input into the neural feedback reinforcement learning model. The mapping relationship between current thermal demand and environment is extracted through the real-time state analysis network, and the feature vector of the user's long-term operation habit is extracted through the historical preference learning network. Record historical preference learning results and integrate the user's long-term operating habit feature vector to generate initial control parameters.

[0010] As a preferred solution of the split air conditioner adaptive control method based on human perception described in the present invention, wherein: the causal rule chain is constructed by the self-explanatory strategy engine, and the causal rule chain is visualized through the interactive interface of the split air conditioner to obtain the weight adjustment instruction of the causal rule chain. The specific steps are as follows: Based on the generation logic of the target temperature, wind speed gear, and wind deflector angle in the initial control parameters, the heat demand level, historical temperature deviation, and environmental temperature and humidity influencing factors are extracted to construct a causal rule chain; The causal rule chain is visualized in the form of an editable flowchart through the interactive interface of the split air conditioner, and the rule weight priority is marked by color. The user's weight adjustment operation in the causal rule chain is received, and the weight adjustment instruction of the causal rule chain is generated.

[0011] As a preferred solution of the split air conditioner adaptive control method based on human perception described in the present invention, wherein: the rule priority is dynamically modified by the slider, the dynamic weighting coefficient in the adaptive control strategy generator is updated, and the target temperature and the air guide plate angle are calculated. The specific steps are as follows: Parse the weight adjustment instructions entered by the user through the interactive interface, extract the weight difference of the causal rule chain, adjust the thermal demand sensitivity coefficient and historical preference weight coefficient in the adaptive control strategy generator, and calculate the target temperature and air deflector angle.

[0012] As a preferred solution of the split air conditioner adaptive control method based on human perception of the present invention, wherein: the split air conditioner is adaptively controlled by controlling the compressor frequency, air guide plate angle and radiation plate power, the specific steps are as follows: Based on the deviation between the target temperature and the current room temperature, the compressor frequency is dynamically adjusted through a segmented control strategy. The air supply angle is calculated based on the thermal image of the user's real-time location and the facial temperature spatial map, combined with the inverse kinematics formula of the air deflector. Based on the temperature difference between the nose tip and forehead and the real-time temperature of the area covered by the radiation panel, the power of the radiation panel is adjusted through a fuzzy control algorithm; The adjustment parameters of the compressor frequency, air guide plate angle and radiation panel power are encoded into split air conditioner hardware executable instructions and sent to the split air conditioner main control panel through a preset communication protocol for temperature control.

[0013] In a second aspect, the present invention provides a split air conditioner adaptive control system based on human perception, comprising an acquisition module, a level module, a control parameter module, an instruction module, a strategy module and a control module; The acquisition module is used to collect the user's brain wave signals, facial temperature gradient distribution data and nose tip and forehead temperature difference data; the level module is used to use the graph convolutional network model to extract time series features from the brain wave signals, and to perform weighted fusion on the nose tip and forehead temperature difference data through the spatial attention mechanism to obtain the thermal demand level; the control parameter module is used to collect the user's historical operation data, input the thermal demand level into the neural feedback reinforcement learning model, and generate initial control parameters; the instruction module is used to construct a causal rule chain through a self-explanatory strategy engine, and visualize the causal rule chain through the interactive interface of the split air conditioner to obtain the weight adjustment instruction of the causal rule chain; the strategy module is used to dynamically modify the rule priority through the slider, update the dynamic weighting coefficient in the adaptive control strategy generator, and calculate the target temperature and the air guide plate angle; the control module is used to adaptively control the split air conditioner by controlling the compressor frequency, air guide plate angle and radiation panel power.

[0014] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the split air conditioner adaptive control method based on human perception as described in the first aspect of the present invention is implemented.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the split air conditioner adaptive control method based on human perception as described in the first aspect of the present invention.

[0016] The beneficial effects of the present invention are as follows: by introducing graph convolutional networks and spatial attention mechanisms, efficient fusion modeling of brain wave signals and multi-point facial temperature data is achieved, significantly improving the accuracy and stability of thermal demand level identification; by constructing an explainable causal rule chain and supporting users to dynamically adjust the priority of control rules, the transparency of the decision-making process and user participation are enhanced, providing a more intelligent, accurate and interactive split air-conditioning adaptive control solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 Flowchart of the adaptive control method for split air conditioners based on human perception.

[0019] Figure 2 Schematic diagram of the split air-conditioning adaptive control system based on human perception.

[0020] Figure 3 Flowchart for thermal demand level identification.

[0021] Figure 4 Flowchart of adaptive control. DETAILED DESCRIPTION

[0022] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0023] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0025] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides a split air conditioner adaptive control method based on human perception, comprising the following steps: S1. Collect the user's brain wave signals, facial temperature gradient distribution data, and nose tip and forehead temperature difference data.

[0026] The millimeter-wave radar scans the user's head area, collects the original electromagnetic reflection signal, and performs noise reduction processing to obtain the brain wave signal.

[0027] Specifically, the millimeter-wave radar is used to scan the user's head area. The millimeter-wave radar transmits a millimeter-wave signal (for example, a frequency of 77GHz is used as an example) to scan the surface of the user's head area. After scanning, the millimeter-wave radar receiver collects the original electromagnetic reflection signal, which comes from the reflection of the millimeter wave in the user's head area. The collected original electromagnetic reflection signal is subjected to noise reduction processing, and the noise reduction processing uses a digital low-pass filter to remove environmental noise interference. The signal after the noise reduction processing is completed directly obtains the brain wave signal.

[0028] The user's facial thermal image is captured by an infrared thermal imager, and the three temperature monitoring points in the nose tip area, forehead area and face area are located based on the facial key point detection algorithm. The temperature distribution matrix is ​​calculated, and the facial temperature gradient distribution data is generated by the temperature difference of adjacent temperature monitoring points.

[0029] Specifically, the infrared thermal imager captures the thermal image of the user's face at a rate of 9 frames per second and outputs thermal image data with a resolution of (example value 640×480) pixels, where each pixel contains the temperature value of the corresponding skin position; the face key point detection algorithm based on the MobileNetV3 architecture is used to process the thermal image, and the two-dimensional coordinates of (example value 68) individual facial key points are identified and output; the key point index example value (30) is selected as the center point of the nose tip area, and according to the general face annotation standard, the key point index 27 is used as the center point of the forehead area, and the key point indices 2 and 14 are used as the center points of the left face area and the right face area respectively; based on the coordinates of the center point of the nose tip, all temperature values ​​in the pixel area (example value 11×11) are extracted to calculate the arithmetic mean as the temperature value of the nose tip area; based on the coordinates of the center point of the forehead, the (example value 11×11) The arithmetic mean of all temperature values ​​in the pixel area is calculated as the temperature value of the forehead area; the temperature average of the pixel area is extracted based on the coordinates of the center point of the left face (for example, a value of 9×9) as the temperature value of the left face area, and the temperature average of the pixel area is extracted based on the coordinates of the center point of the right face (for example, a value of 9×9) as the temperature value of the right face area; the temperature distribution matrix is ​​the temperature data matrix originally output by the thermal imager; adjacent temperature monitoring points are selected to form five groups of adjacent relationship pairs: the nose tip area and the left face area, the nose tip area and the right face area, the forehead area and the left face area, the forehead area and the right face area, and the left face area and the right face area; the temperature difference is calculated for each group of adjacent relationship pairs, for example, the temperature difference between the temperature value of the left face area and the temperature value of the right face area is equal to the absolute value of the temperature value of the left face area minus the temperature value of the right face area; and the facial temperature gradient distribution data is generated.

[0030] The dynamic temperature difference index is calculated based on the temperature difference between the nose tip area and the forehead area to generate the nose tip and forehead temperature difference data.

[0031] Specifically, based on the real-time ambient temperature value collected by the ambient temperature sensor, the temperature compensation coefficient (for example, a temperature compensation coefficient of 0.12) and the human body comfort reference temperature (for example, 25°C) are applied to compensate the temperature value of the tip of the nose area and the temperature value of the forehead area: the compensated temperature value of the tip of the nose area is equal to the original temperature value of the tip of the nose area plus the difference between the value collected by the ambient temperature sensor using the temperature compensation coefficient and the human body comfort reference temperature; the compensated temperature value of the forehead area is equal to the original temperature value of the forehead area plus the temperature compensation coefficient combined with the difference between the value collected by the ambient temperature sensor and the human body comfort reference temperature; the difference between the compensated temperature value of the tip of the nose area and the compensated temperature value of the forehead area is calculated as the compensated temperature difference; the time change rate of the temperature value of the tip of the nose area is calculated by the central difference method (for example, the change in the temperature value of the tip of the nose area between adjacent 2-second time intervals divided by the time interval of 2 seconds); the compensated temperature difference and the time change rate of the temperature value of the tip of the nose area are added to generate a dynamic temperature difference index; the output includes the dynamic temperature difference index and the temperature difference data of the tip of the nose and forehead.

[0032] S2. Use the graph convolutional network model to extract temporal features from EEG signals, and perform weighted fusion of the nose tip and forehead temperature difference data through the spatial attention mechanism to obtain the thermal demand level.

[0033] The EEG signal is divided into multiple time series segments through time windows, the energy feature of each time window is used as a node, and the signal correlation of adjacent time windows is used as the edge weight to construct the EEG time series graph.

[0034] Specifically, the EEG signal (the example value is divided into 2000 millisecond time windows, and the adjacent windows slide with a step size of 500 milliseconds) to form a continuous time window sequence; the total energy value of the EEG signal in the β frequency band in each window is calculated by discrete Fourier transform as the time window energy feature; the linear correlation of the original EEG signal in the adjacent time windows is calculated based on the Pearson correlation coefficient formula as the edge weight; the energy eigenvalue set is used as the node set, the connection relationship between the nodes is used as the edge set, and the correlation coefficient value is used as the edge weight set to construct the adjacency matrix structure of the EEG time series graph, and the EEG time series graph in the form of a lower triangular matrix is ​​obtained, with the main diagonal as the energy eigenvalue and the first diagonal of the lower triangle as the correlation coefficient; Based on the facial temperature gradient distribution data, the temperature monitoring points in the nose tip area, forehead area and facial area are used as nodes, and the absolute value of the temperature difference between adjacent temperature monitoring points is used as the edge weight to construct a facial temperature space graph.

[0035] Specifically, based on the user's facial thermal map data captured by the infrared thermal imager, four temperature monitoring points, namely the center point of the nose tip area, the center point of the forehead area, the center point of the left face area and the center point of the right face area, are located; the four temperature monitoring points are defined as nodes in the graph structure; five connecting edges are established between the four nodes: the edge between the center point of the nose tip area and the center point of the left face area, the edge between the center point of the nose tip area and the center point of the right face area, the edge between the center point of the forehead area and the center point of the left face area, the edge between the center point of the forehead area and the center point of the right face area, and the edge between the center point of the left face area and the center point of the right face area; for each connecting edge, the absolute value of the temperature difference between the two nodes is calculated as the edge weight, for example, the edge weight between the center point of the nose tip area and the center point of the left face area is equal to the temperature value of the center point of the nose tip area minus the absolute value of the temperature value of the center point of the left face area; the attribute value of each node is the temperature value of the corresponding center point; and a facial temperature space graph containing four nodes and five weighted edges is constructed.

[0036] It should be noted that the formula for calculating the absolute value of the temperature difference between two nodes is: ; in, node and nodes The edge weights between Representation node The temperature value, Representation node The temperature value.

[0037] Graph convolution operations are performed on the EEG timing graph and the facial temperature space graph respectively to extract the temporal dynamic features of the EEG signal and the global spatial correlation features of the temperature distribution.

[0038] Specifically, based on the adjacency matrix structure of the EEG time-series graph (the first diagonal element below the main diagonal represents the edge weight), a symmetric normalized adjacency matrix is ​​calculated. Feature propagation is calculated using a graph convolutional network model, with the input features being the time window energy feature vector and the trainable weight matrix. Two layers of graph convolutional layers are stacked to output a temporal dynamic feature matrix, the row vectors of which are summed to obtain the 64-dimensional temporal dynamic features of the EEG signal. A 4×4 adjacency matrix is ​​constructed for the facial temperature spatial graph, with non-zero elements corresponding to the absolute value weights of the temperature differences along the five edges. Symmetric normalization is performed, and a single layer of graph convolution is performed to output a 4×16-dimensional temporal dynamic feature matrix. The row vectors of the temporal dynamic feature matrix are concatenated to obtain the 64-dimensional global spatial correlation features of the temperature distribution.

[0039] It should be noted that the graph convolutional network model is trained by preparing a training data set, which includes EEG time series graph samples (such as the energy feature matrix of 80 time windows), facial temperature space graph samples (such as four node temperature values ​​and five edge weight matrices), and corresponding real comfort labels (such as thermal comfort voting values); the EEG time series graph samples are input into a two-layer graph convolutional network (the first layer weight matrix dimension is 80×32, the second layer is 32×32) to output the EEG feature sequence; the facial temperature space graph samples are input into a single-layer graph convolutional network (the weight matrix dimension is 4×16) to output the temperature feature matrix; the EEG feature sequence is reduced to a 64-dimensional vector through a time dimension summation operation. The temperature feature matrix was reduced to a 64-dimensional vector by concatenating node features. Two 64-dimensional vectors were concatenated into a 128-dimensional fused feature vector. The 128-dimensional fused feature vector was input into a three-layer fully connected neural network (128×64 in the first layer, 64×32 in the second layer, and 32×1 in the third layer) to predict the comfort score. The mean squared error loss between the predicted score and the true comfort label was calculated. The Adam optimizer was used to train for 100 rounds with a learning rate of 0.001 and a batch size of 32 to update the graph convolutional network weight parameters and the fully connected neural network weight parameters. The training was terminated when the validation set loss change was less than 0.001 for 10 consecutive rounds, completing the training of the graph convolutional network model.

[0040] The global spatial correlation features of temperature distribution are dynamically weighted through the spatial attention mechanism, and are cross-modally spliced ​​and fused with the temporal dynamic features of EEG signals, and the thermal demand level is generated through the threshold segmentation algorithm.

[0041] Specifically, based on the global spatial correlation feature vector and three different trainable weight matrices, a 16-dimensional query vector, a 16-dimensional key vector and a 16-dimensional value vector are generated respectively; the dot product of the query vector and the key vector is calculated to obtain the initial attention score; the initial attention score is input into the softmax function to be converted into a standardized attention weight in the form of a probability distribution; the standardized attention weight is used to perform weighted summation on the 16-dimensional value vector to generate a 16-dimensional intermediate vector; the intermediate vector is transformed into a 64-dimensional dynamic weighted temperature feature vector through a fully connected layer, and a cascade splicing operation is performed on the 64-dimensional EEG signal time series dynamic feature vector and the 64-dimensional dynamic weighted temperature feature vector: the two feature vectors are connected end to end in the feature dimension to form a 128-dimensional fusion feature vector. The 128-dimensional fused feature vector is processed by a three-layer fully connected neural network: the first fully connected layer converts the 128-dimensional fused feature vector input into a 32-dimensional feature vector and applies the ReLU activation function; the second fully connected layer converts the 32-dimensional feature vector into a 16-dimensional feature vector and applies the ReLU activation function; the third fully connected layer converts the 16-dimensional feature vector into a one-dimensional raw prediction value, and the thermal demand level is determined based on the raw prediction value: based on physiological principles, if the raw prediction value is greater than 1.0, it is judged as Level 1 (obviously hot); if the raw prediction value is greater than 0 and does not exceed 1.0, it is judged as Level 2 (slightly hot); if the raw prediction value is equal to 0, it is judged as Level 3 (comfortable); if the raw prediction value is not less than -1.0 and less than 0, it is judged as Level 4 (slightly cold); if the raw prediction value is less than -1.0, it is judged as Level 5 (obviously cold).

[0042] S3. Collect user historical operation data, input the thermal demand level into the neural feedback reinforcement learning model, and generate initial control parameters.

[0043] Collect user historical operation data and combine it with the current thermal demand level to form a multi-dimensional state vector.

[0044] Specifically, temperature setting value records (e.g., the set temperature value at the top of the hour) are extracted from the air conditioning control records to form a temperature setting time series; the number of wind speed gear switching times within the time window is calculated and divided by the total number of minutes to obtain the wind speed adjustment frequency; at the same time, the ambient temperature value and relative humidity value collected every minute by the ambient temperature and humidity sensor are obtained; the heat demand level generated at the current moment is combined with the wind speed adjustment frequency, ambient temperature value, and relative humidity value into a four-dimensional state vector to obtain a multi-dimensional state vector.

[0045] The multidimensional state vector is input into the reinforcement learning model, the mapping relationship between current thermal demand and environment is extracted through the real-time state analysis network, and the feature vector of the user's long-term operation habit is extracted through the historical preference learning network.

[0046] Specifically, the multidimensional state vector is input into the real-time state analysis network; the real-time state analysis network consists of a three-layer fully connected neural network, with 128 neurons in the first layer, 64 in the second layer, and 32 in the third layer, and the activation function adopts ReLU; the real-time state analysis network outputs a multidimensional state vector; the user's temperature setting time series for the past seven days is simultaneously input into the historical preference learning network; a bidirectional LSTM structure is adopted, the input sequence is the 24-hour hourly temperature setting values ​​for 7 consecutive days (sequence length 168), and the hidden layer dimension is 32; the LSTM time step hidden state is used as the feature vector of the user's long-term operating habits.

[0047] It should be noted that the multidimensional state vector containing the heat demand level, wind speed adjustment frequency, ambient temperature value, and relative humidity value is input into the three-layer fully connected structure of the real-time state analysis network; the 168 temperature setting values ​​of the user in the past seven days are simultaneously input into the bidirectional LSTM of the historical preference learning network (input 168-dimensional sequence → 32-dimensional hidden layer → output 32-dimensional features); the multidimensional state vector output by the real-time state analysis network and the 32-dimensional features output by the historical preference learning network are spliced ​​into a 36-dimensional joint feature vector; the A3C algorithm is used to update the weight parameters of the policy network with a learning rate of 0.0002; 256 sets of state-action-reward data are processed in each training batch; training is stopped when the average comprehensive reward value increase of 10 consecutive batches is less than 1%, and the trained real-time state analysis network is obtained.

[0048] Record historical preference learning results and integrate the user's long-term operating habit feature vector to generate initial control parameters.

[0049] Specifically, the first to fourth dimension eigenvalues ​​of the user's long-term operating habit feature vector are extracted as the basic temperature preference weight (e.g., the first dimension eigenvalue is 0.83), the wind speed preference coefficient (e.g., the second dimension eigenvalue is 1.2), the angle adjustment tendency (e.g., the third dimension eigenvalue is 0.6), and the environmental coupling strength (e.g., the fourth dimension eigenvalue is 0.9); the thermal demand level and environmental mapping relationship characteristics (the fourth dimension eigenvalue of the vector) are extracted from the four-dimensional feature vector output by the real-time state analysis network; four correction factors are obtained based on the fourth dimension eigenvalue of the multidimensional state vector and the fifth to eighth dimension eigenvalues ​​of the user's long-term operating habit feature vector; the target temperature value is calculated based on the basic temperature preference weight, the correction factor, and the current thermal demand level (e.g., the target temperature value is equal to the basic temperature preference weight multiplied by the thermal demand level coefficient plus the average of the correction factors); the wind speed gear (e.g., the wind speed gear value is equal to the wind speed preference coefficient multiplied by 3 and rounded up) and the air deflector angle (e.g., the air deflector angle value is equal to the angle adjustment tendency multiplied by 90 and rounded up) are generated based on the wind speed preference coefficient and the angle adjustment tendency; and the initial control parameters including the target temperature value, wind speed gear, and air deflector angle are output.

[0050] S4. Construct a causal rule chain through a self-explanatory strategy engine, and visualize the causal rule chain through the interactive interface of the split air conditioner to obtain a weight adjustment instruction for the causal rule chain.

[0051] Based on the generation logic of the target temperature, wind speed gear and air deflector angle in the initial control parameters, the heat demand level, historical temperature deviation and ambient temperature and humidity influencing factors are extracted to construct a causal rule chain.

[0052] Specifically, according to the thermal demand level, the historical temperature deviation value is calculated (by retrieving the user's temperature setting and ambient temperature data sequence within 60 minutes, calculating the difference between the air conditioner setting value and the ambient temperature minute by minute and then averaging it); based on the current ambient temperature value and relative humidity value, the temperature and humidity coupling model of the core standard of thermal comfort environment design is applied to calculate the ambient temperature and humidity influencing factor; the thermal demand level, historical temperature deviation and ambient temperature and humidity influencing factor are used as the attribute values ​​of the thermal demand level node, the historical temperature deviation node and the ambient temperature and humidity node respectively; a connection edge is established from the thermal demand level node to the target temperature node and a weight is set, a connection edge is established from the historical temperature deviation node to the target temperature node and a weight is set, a connection edge is established from the ambient temperature and humidity node to the wind speed node and a weight is set, and a connection edge is established from the ambient temperature and humidity node to the air deflector angle node and a weight is set; a causal rule chain is formed consisting of three input nodes (thermal demand level node, historical temperature deviation node, ambient temperature and humidity node), three output nodes (target temperature node, wind speed node, air deflector angle node) and four weighted edges.

[0053] The causal rule chain is visualized in the form of an editable flowchart through the interactive interface of the split air conditioner, and the rule weight priority is marked by color. The user's weight adjustment operation in the causal rule chain is received, and the weight adjustment instruction of the causal rule chain is generated.

[0054] Specifically, the six nodes in the causal rule chain (thermal demand level node, historical temperature deviation node, ambient temperature and humidity node, target temperature node, wind speed node, and air guide plate angle node) and the four connecting edges are rendered as geometric graphics in the flowchart editing area of ​​the air conditioning interaction interface (for example, the thermal demand level node is rendered as a red circle with a diameter of 25 mm, and the ambient temperature and humidity node is rendered as a blue diamond with a side length of 20 mm); the line color gradient is set according to the connecting edge weight value; the connecting edge weight value editing control is opened (for example, a slider with a range of ±30% is displayed on the right side of each edge); the user drags the slider operation (for example, the weight of the thermal demand level node to the target temperature node is adjusted from 0.65 to 0.75); the slider displacement is captured in real time to calculate the weight difference (for example, a displacement of +15% corresponds to a weight difference of +0.1); the target connecting edge identifier and the weight difference are packaged into a triple data structure to generate a weight adjustment instruction.

[0055] S5. Dynamically modify the rule priority through the slider, update the dynamic weighting coefficient in the adaptive control strategy generator, and calculate the target temperature and the wind deflector angle.

[0056] Parse the weight adjustment instructions entered by the user through the interactive interface, extract the weight difference of the causal rule chain, adjust the thermal demand sensitivity coefficient and historical preference weight coefficient in the adaptive control strategy generator, and calculate the target temperature and air deflector angle.

[0057] Specifically, the target connection edge identifier string and the slider displacement value are extracted from the weight adjustment instruction; the displacement weight difference is calculated according to the displacement calculation formula; the target connection edge identifier string is mapped to the coefficient type (if it is the heat demand level-target temperature, it corresponds to the heat demand sensitivity coefficient; if it is the historical temperature deviation-target temperature, it corresponds to the historical preference weight coefficient); according to the coefficient type, the heat demand sensitivity coefficient is updated (the new value is equal to the original value plus 0.15 multiplied by the weight difference) or the historical preference weight coefficient is updated; the updated heat demand sensitivity coefficient and the dynamic temperature difference index are used to input the Gaussian error function to calculate the temperature compensation amount, and the user preset temperature value and the updated historical preference weight coefficient are superimposed with the result of the historical temperature deviation to generate the target temperature value; the environmental response gain and the time change rate of the environmental temperature and humidity influencing factor are input into the hyperbolic tangent function to obtain the angle base value, and then combined with (1+environmental relative humidity RH / 100) and truncated to 0 degrees to 90 degrees to generate the air guide plate angle value.

[0058] S6. Adaptively control the split air conditioner by controlling the compressor frequency, air guide plate angle, and radiation panel power.

[0059] Based on the deviation between the target temperature and the current room temperature, the compressor frequency is dynamically adjusted through a segmented control strategy. The air supply angle is calculated based on the thermal imaging of the user's real-time location and the facial temperature spatial map, combined with the inverse kinematics formula of the air deflector.

[0060] Specifically, the target temperature is calculated by an adaptive control strategy generator based on the heat demand level and historical user operation data. The current room temperature, collected in real time by the air conditioner's built-in ambient temperature sensor, is then subtracted from the target temperature to obtain a temperature deviation. A step-by-step control strategy is used to adjust the compressor frequency based on the temperature deviation. When the temperature deviation is high (e.g., greater than 3°C), the compressor frequency is set to a high frequency (e.g., 80Hz in this example) to quickly approach the target temperature. When the temperature deviation is medium (e.g., between 1 and 3°C), the compressor frequency is set to a medium frequency (e.g., 60Hz in this example) to achieve smooth cooling or heating. When the temperature deviation is low (e.g., less than 1°C), the compressor frequency is set to a low frequency (e.g., 45Hz in this example) to obtain the compressor frequency. An infrared thermal imager captures the user's real-time location. The user's head position is determined based on this thermal image, and the user's current air supply zone is analyzed using a facial temperature spatial map. Substitute the user's head position coordinates into the inverse kinematics formula of the air deflector to calculate the angle parameters that need to be adjusted for the air deflector.

[0061] Based on the temperature difference data between the nose tip and forehead and the real-time temperature of the area covered by the radiation panel, the power of the radiation panel is adjusted through a fuzzy control algorithm.

[0062] Specifically, the temperature difference data between the nose tip and forehead (i.e., the value of the dynamic temperature difference index (DTI)) and the real-time temperature value of the area covered by the radiation panel are obtained; the dynamic temperature difference index (DTI) is divided into seven fuzzy sets and a triangular membership function is defined; the real-time temperature value of the radiation panel is divided into five fuzzy sets (too low, too low, appropriate, too high, and too high) and a trapezoidal membership function is defined; the output membership is calculated using the maximum-minimum reasoning method (calculating the antecedent truth value for each rule and trimming the consequent output set, and then aggregating all rule outputs); the center of gravity method is used to defuzzify and calculate the precise power adjustment amount; and the power adjustment amount is increased or decreased based on the current power value of the radiation panel to obtain the radiation panel power.

[0063] The adjustment parameters of the compressor frequency, air guide plate angle and radiation panel power are encoded into split air conditioner hardware executable instructions and sent to the split air conditioner main control panel through the communication protocol for temperature control.

[0064] Specifically, the adjustment parameters include the compressor frequency, the required air deflector angle, and the radiant panel power. The compressor frequency adjustment parameters are converted into binary instructions executable by the split air conditioner hardware. The hardware-defined instruction coefficients are calculated using a Hertz value. The air deflector angle adjustment parameters are converted into hardware binary instructions by directly mapping the degree value to the instruction value table (preset in the split air conditioner hardware manual). The radiant panel power adjustment parameters are converted into hardware binary instructions by dividing the watt value by the hardware baseline power value. These binary instructions are assembled into a predefined communication protocol data structure (for example, in Modbus-RTU format, the data assembly order is compressor frequency instruction first, air deflector angle instruction second, and radiant panel power instruction last). The assembly process strictly matches the protocol field length (for example, the compressor frequency instruction occupies 1 byte, the air deflector angle instruction occupies 1 byte, and the radiant panel power instruction occupies 2 bytes). After the instructions are assembled, they are transmitted to the split air conditioner's main control panel via the communication protocol. The data frames are sent using the predefined serial port. During the transmission process, a checksum algorithm (such as a CRC checksum algorithm to calculate the data frame redundancy value) is applied to verify the data correctness. After receiving the instructions, the main control panel of the split air conditioner interprets them as corresponding hardware operations, driving the compressor frequency adjustment component, the air guide plate angle adjustment component and the radiation plate power adjustment component to execute parameter changes, thereby realizing temperature control of the split air conditioner.

[0065] This embodiment also provides a split air conditioner adaptive control system based on human perception, including: an acquisition module, a level module, a control parameter module, an instruction module, a strategy module, and a control module; the acquisition module is used to collect the user's brain wave signals, facial temperature gradient distribution data, and nose tip and forehead temperature difference data; the level module is used to extract time series features from the brain wave signals using a graph convolutional network model, and perform weighted fusion on the nose tip and forehead temperature difference data through a spatial attention mechanism to obtain a thermal demand level; the control parameter module is used to collect the user's historical operation data, input the thermal demand level into a neural feedback reinforcement learning model, and generate initial control parameters; the instruction module is used to construct a causal rule chain through a self-explanatory strategy engine, and visualize the causal rule chain through the split air conditioner's interactive interface to obtain a weight adjustment instruction for the causal rule chain; the strategy module is used to dynamically modify the rule priority through a slider, update the dynamic weighting coefficient in the adaptive control strategy generator, and calculate the target temperature and air deflector angle; the control module is used to adaptively control the split air conditioner by controlling the compressor frequency, air deflector angle, and radiation panel power.

[0066] This embodiment also provides a computer device, which is suitable for the case of a split air conditioner adaptive control method based on human perception, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the split air conditioner adaptive control method based on human perception proposed in the above embodiment.

[0067] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0068] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for implementing the adaptive control of a split air conditioner based on human perception as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0069] In summary, the present invention achieves efficient fusion modeling of EEG signals and multi-point facial temperature data by introducing graph convolutional networks and spatial attention mechanisms, significantly improving the accuracy and stability of thermal demand level identification; by constructing an interpretable causal rule chain and supporting users to dynamically adjust the priority of control rules, it enhances the transparency of the decision-making process and user participation, and provides a more intelligent, accurate and interactive split air-conditioning adaptive control solution.

[0070] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A split air conditioner adaptive control method based on human body perception, characterized by: include, Collect the user's brain wave signals, facial temperature gradient distribution data, and nose tip and forehead temperature difference data; A graph convolutional network model is used to extract temporal features from EEG signals, and a spatial attention mechanism is used to perform weighted fusion of the nose tip and forehead temperature difference data to obtain the thermal demand level. Collect historical user operation data, input the thermal demand level into the neural feedback reinforcement learning model, and generate initial control parameters; A causal rule chain is constructed through a self-explanatory strategy engine, and the causal rule chain is visualized through the interactive interface of the split air conditioner to obtain the weight adjustment instruction of the causal rule chain; Dynamically modify the rule priority through the slider, update the dynamic weighting coefficient in the adaptive control strategy generator, and calculate the target temperature and air deflector angle; The split air conditioner is adaptively controlled by controlling the compressor frequency, air guide plate angle and radiation panel power.

2. The method for adaptive control of split air conditioners based on human perception according to claim 1, characterized in that: The specific steps of collecting the user's brain wave signals, facial temperature gradient distribution data and nose tip and forehead temperature difference data are as follows: Collect the original electromagnetic reflection signal and perform noise reduction processing to obtain the brain wave signal; The user's facial thermal image is captured using an infrared thermal imager. Based on the facial key point detection algorithm, three temperature monitoring points (nose tip, forehead, and face) are located. The temperature distribution matrix is ​​calculated and the facial temperature gradient distribution data is generated by taking the temperature differences between adjacent temperature monitoring points. The dynamic temperature difference index is calculated based on the temperature difference between the nose tip area and the forehead area to generate the nose tip and forehead temperature difference data.

3. The method for adaptive control of split air conditioners based on human perception according to claim 2, characterized in that: The graph convolutional network model is used to extract time series features from brain wave signals, and the temperature difference data between the nose tip and forehead are weighted and fused through the spatial attention mechanism to obtain the thermal demand level. The specific steps are as follows: The EEG signal is divided into multiple time series segments by time window, the energy feature of each time window is used as a node, and the signal correlation of adjacent time windows is used as the edge weight to construct the EEG time series graph; Based on the facial temperature gradient distribution data, the temperature monitoring points in the nose tip area, forehead area and facial area are used as nodes, and the absolute value of the temperature difference between adjacent temperature monitoring points is used as the edge weight to construct the facial temperature space graph; Graph convolution operations are performed on the EEG time series graph and the facial temperature spatial graph respectively to extract the temporal dynamic features of the EEG signal and the global spatial correlation features of the temperature distribution; The global spatial correlation features of temperature distribution are dynamically weighted through the spatial attention mechanism, and are cross-modally spliced ​​and fused with the temporal dynamic features of EEG signals, and the thermal demand level is generated through the threshold segmentation algorithm.

4. The method for adaptive control of split air conditioners based on human perception according to claim 3, characterized in that: The user's historical operation data is collected, the thermal demand level is input into the neural feedback reinforcement learning model, and the initial control parameters are generated. The specific steps are as follows: Collect historical user operation data and combine it with the current heat demand level to form a multi-dimensional state vector; The multi-dimensional state vector is input into the neural feedback reinforcement learning model. The mapping relationship between current thermal demand and environment is extracted through the real-time state analysis network, and the feature vector of the user's long-term operation habit is extracted through the historical preference learning network. Record historical preference learning results and integrate the user's long-term operating habit feature vector to generate initial control parameters.

5. The method for adaptive control of split air conditioners based on human perception according to claim 4, characterized in that: The causal rule chain is constructed through the self-explanatory strategy engine, and the causal rule chain is visualized through the interactive interface of the split air conditioner to obtain the weight adjustment instruction of the causal rule chain. The specific steps are as follows: Based on the generation logic of the target temperature, wind speed gear, and wind deflector angle in the initial control parameters, the heat demand level, historical temperature deviation, and environmental temperature and humidity influencing factors are extracted to construct a causal rule chain; The causal rule chain is visualized in the form of an editable flowchart through the interactive interface of the split air conditioner, and the rule weight priority is marked by color. The user's weight adjustment operation in the causal rule chain is received, and the weight adjustment instruction of the causal rule chain is generated.

6. The method for adaptive control of split air conditioners based on human perception according to claim 5, characterized in that: The specific steps of dynamically modifying the rule priority through the slider, updating the dynamic weighting coefficient in the adaptive control strategy generator, and calculating the target temperature and the wind deflector angle are as follows: Parse the weight adjustment instructions entered by the user through the interactive interface, extract the weight difference of the causal rule chain, adjust the thermal demand sensitivity coefficient and historical preference weight coefficient in the adaptive control strategy generator, and calculate the target temperature and air deflector angle.

7. The method for adaptive control of split air conditioners based on human perception according to claim 6, characterized in that: The split air conditioner is adaptively controlled by controlling the compressor frequency, air guide plate angle and radiation plate power. The specific steps are as follows: Based on the deviation between the target temperature and the current room temperature, the compressor frequency is dynamically adjusted through a segmented control strategy. The air supply angle is calculated based on the thermal image of the user's real-time location and the facial temperature spatial map, combined with the inverse kinematics formula of the air deflector. Based on the temperature difference between the nose tip and forehead and the real-time temperature of the area covered by the radiation panel, the power of the radiation panel is adjusted through a fuzzy control algorithm; The adjustment parameters of the compressor frequency, air guide plate angle and radiation panel power are encoded into split air conditioner hardware executable instructions and sent to the split air conditioner main control panel through a preset communication protocol for temperature control.

8. A split air conditioner adaptive control system based on human perception, based on the split air conditioner adaptive control method based on human perception according to any one of claims 1 to 7, characterized in that: Including, acquisition module, level module, control parameter module, instruction module, strategy module and control module; The acquisition module is used to collect the user's brain wave signals, facial temperature gradient distribution data, and nose tip and forehead temperature difference data; The level module is used to extract time series features from EEG signals using a graph convolutional network model and perform weighted fusion of the nose tip and forehead temperature difference data through a spatial attention mechanism to obtain the thermal demand level; The control parameter module is used to collect historical user operation data, input the thermal demand level into the neural feedback reinforcement learning model, and generate initial control parameters; The instruction module is used to construct a causal rule chain through a self-explanatory strategy engine, and visualize the causal rule chain through the interactive interface of the split air conditioner to obtain the weight adjustment instruction of the causal rule chain; The strategy module is used to dynamically modify the rule priority through the slider, update the dynamic weighting coefficient in the adaptive control strategy generator, and calculate the target temperature and the deflector angle; The control module is used to adaptively control the split air conditioner by controlling the compressor frequency, air guide plate angle and radiation panel power.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the split air conditioner adaptive control method based on human perception according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the split air conditioner adaptive control method based on human perception according to any one of claims 1 to 7 are implemented.

Images