Self-adaptive control method and device for multiple air outlets

Abstract

The invention provides a multi-air-outlet self-adaptive control method based on user behaviors. The multi-air-outlet self-adaptive control method comprises the steps that historical setting parameters are obtained from a database through a control panel; controlling operation of an air outlet corresponding to the control panel based on the historical setting parameters; user real-time adjustment parameters are obtained through the control panel, and the user real-time adjustment parameters are stored in the database; processing the historical setting parameters and the user real-time adjustment parameters in the database by using a multi-air-outlet balance adaptive model so as to obtain operation parameters for a plurality of air outlets; and operation of the multiple air outlets is controlled based on the operation parameters of the multiple air outlets, and the operation parameters of the multiple air outlets are stored in a database. According to the method, the balance point which can best meet the user requirements is searched according to the historical user adjustment behaviors and the mutual influence of the users in the current space, the real requirements of numerous users or articles at different positions in the environment space can be met, the user experience is improved, and meanwhile energy is saved.

Description

Technical Field

[0001] The present application relates to the field of control of cooling and heating equipment, and in particular to a method and device for adaptively controlling an air outlet of an air conditioner. Background Art

[0002] In daily life, for example, in offices or some scenes that do not change often, there are usually multiple central air-conditioning outlets. Different people or objects are distributed around each outlet. Different people or different objects have different needs and feelings for temperature and air volume at different times. Usually, if the air conditioner is turned on in an office, some people may be afraid of the cold, and even if they wear a lot of clothes, they still feel that it is not warm enough. At the same time, some people feel hot and want to lower the air conditioner temperature as much as possible. Since the positions of people or objects in the office are generally fixed, it is impossible to divide different groups of people into multiple rooms according to individual needs. Therefore, people in the office will consciously adjust and control the switch panel of the air outlet, adjust the temperature and air volume of the corresponding air outlet, to adapt to the needs of their own body or objects. Summary of the invention

[0003] The inventor has found that there are some technologies for automatic adjustment of air outlets for human body temperature, posture, and body shape, which are mainly divided into two types. One is to collect the user's body temperature and the temperature around the air outlet through a temperature sensor at the air outlet of the air conditioner, collect the humidity around the air outlet through a humidity sensor, and collect the user's posture through a camera. The collected air outlet temperature, air outlet humidity, and user posture are uploaded to a central controller or a server, and the deviation between the air outlet temperature and humidity and the normal body temperature and humidity is calculated. The deviation value is calculated by comparing the relationship between the user's posture and the perceived cold and heat, and the deviation value is fed back to the controller. The controller adjusts the temperature and humidity of the air outlet of the air conditioner based on the calculated deviation value to meet the user's cold and hot needs. The other is to use a memory to automatically record the state parameters such as the air conditioner adjustment temperature, wind speed, wind direction, and the environmental parameters for executing the instruction according to the user's habits to form a database. Therefore, under the corresponding environmental parameters, the corresponding state parameters of the air outlet of the air conditioner in the database are called to achieve the adjustment of the parameters such as the temperature, wind speed, and wind direction of the air outlet of the air conditioner.

[0004] However, the inventors note that the two methods mentioned above do not solve the problem of balancing the impact of wind speed, wind direction, temperature and humidity on multiple users and multiple air outlets. They only consider adjusting and controlling the air-conditioning outlets based on the habits of individual users, and do not consider the mutual influence of wind speed, wind direction, temperature and humidity between multiple air outlet arrays that often appear in large spaces, which has a more obvious impact on multiple users. Multiple users will adjust the air outlet parameters of the user's area according to their own physique, body feeling, actual demand for air volume, etc., and users in other areas of the same space will alternately adjust the air outlet parameters of multiple air outlets. The user adjustment process is a dynamic and random process, and there will be repeated adjustments to the air outlet parameters. The user's temperature and air volume requirements may be contradictory, and there will be multiple rounds of game adjustment of the air outlet parameters, and finally a balanced steady state will be reached. This balance point is the result of the users' final compromise, but it is not necessarily the optimal solution. The process of users adjusting the air outlet parameters through multiple rounds of game adjustment is time-consuming and will affect the user experience. To this end, this application seeks the balance point that best meets user needs based on historical user adjustment behaviors and the mutual influence of users in the current space. This application can meet the real needs of many users or objects in different locations of the environment space by modifying the state of the adjustment outlet, improving the user experience, while saving energy, and meeting the current energy-saving indicators.

[0005] Specifically, according to one aspect of the present application, the present application provides a method for balanced adaptive control of multiple air outlets at an air terminal based on user behavior, including: obtaining historical setting parameters associated with at least one control panel from a database through at least one of a plurality of control panels, and one of the plurality of control panels corresponds to one of a plurality of air outlets; controlling the operation of at least one air outlet corresponding to the at least one control panel based on the historical setting parameters; obtaining user real-time adjustment parameters through at least one of the plurality of control panels, and storing the user real-time adjustment parameters in a database; using a multi-air outlet balanced adaptive model to process the historical setting parameters and the user real-time adjustment parameters in the database to obtain operating parameters for a plurality of air outlets; and controlling the operation of the plurality of air outlets based on the operating parameters of the plurality of air outlets, and storing the operating parameters of the plurality of air outlets in a database.

[0006] According to one aspect of the present application, using a multi-outlet balance adaptive model to process historical setting parameters in a database and user real-time adjustment parameters includes: calculating the total amount of cold and heat at the location of at least one outlet according to the historical setting parameters. Where x is the horizontal coordinate of the air outlet in space, y is the vertical coordinate of the air outlet in space, θ is the angle of the air outlet blade, v is the wind speed of the air outlet, t is the air outlet time, T is the specific time, N is the number of times the air outlet is adjusted, m is the number of rows where the air outlet is located, and n is the number of columns where the air outlet is located; the user's cooling and heating demand u (x+Δx, y+Δy, θ, v, t, T) at the location of at least one air outlet is calculated according to the user's real-time adjustment parameters db , R H ), where x is the horizontal coordinate of the air outlet in space, Δx is the displacement from the horizontal coordinate of the air outlet, y is the vertical coordinate of the air outlet in space, Δy is the displacement from the vertical coordinate of the air outlet, θ is the angle of the air outlet blade, v is the wind speed at the air outlet, t is the real time, T db is the outdoor dry-bulb temperature of the day, R H is the outdoor relative humidity; and the difference between the total cooling and heating amount calculated based on the model and the cooling and heating amount demand of the user is obtained by the cooling (heating) amount demand Δε=G(x, y)-u(+Δx, y+Δy, θ, v, t, T db , R H ).

[0007] According to one aspect of the present application, using a multi-air outlet balance adaptive model to process historical setting parameters and user real-time adjustment parameters in a database includes using a Transformer model to process historical setting parameters and user real-time adjustment parameters, and the Transformer model includes: an encoding module, the encoding module is configured to encode parameter token feature sequence data of multiple air outlets and output an encoding vector, the encoding module includes a first encoder, a second encoder and a third encoder; a decoding module, the decoding module is configured to decode the encoding vector output by the encoding module and output a probability vector, the decoding module includes a first decoder, a second decoder and a third decoder; and a probability calculation module, the probability calculation module is configured to process the probability vector output by the decoding module to output the parameter token feature sequence data of the air outlet with the largest probability value.

[0008] According to one aspect of the present application, encoding the parameter token feature sequence data of multiple air outlets using the encoding module includes: using sine and cosine functions to position encode the parameter token feature sequences of multiple air outlets to obtain position encoding of the parameter token feature sequences of multiple air outlets. Among them, pos represents the position of the word in the token feature sequence, d represents the dimension of the position vector PE, 2i represents an even dimension, and 2i+1 represents an odd dimension (2i<d, 2i+1<d); and a d-dimensional encoding vector is generated based on the generated position encoding.

[0009] According to one aspect of the present application, decoding the encoded parameter token feature sequence data using a decoding module includes: generating a vector containing probability information based on the encoding vector generated by the encoding module and the position encoding vector output by the decoding module.

[0010] According to one aspect of the present application, using a probability calculation module to process the probability vector output by the decoding module includes: passing the vector containing probability information output by the decoding module through a linearization layer and a SoftMax layer to calculate the output probability, so as to output the parameter token feature sequence data of the air outlet with the largest probability value.

[0011] According to one aspect of the present application, each of the first encoder, the second encoder and the third encoder includes: a first encoding sublayer connection structure, the first encoding sublayer connection structure includes a multi-head attention layer and a residual connection and a normalization layer; and a second encoding sublayer connection structure, the second encoding sublayer connection structure includes a forward feedback layer and a residual connection and a normalization layer.

[0012] According to one aspect of the present application, the input of the multi-head attention layer is three matrices: Where X is the word vector, are three trained weight matrices; the output of the multi-head attention layer is: Among them, d k is the number of columns of the K matrix, that is, the dimension of the K vector.

[0013] According to one aspect of the present application, the input of the residual connection and normalization layer is: LayerNorm(X+MultiHeadAttetion(X)), LayerNorm(X+FeedForward(X)), where X represents the input of MultiHeadAttetion or FeedForward, MultiHeadAttetion(X) and FeedForward(X) represent the output, X+FeedForward(X) is a residual connection, and LayerNorm converts the input of each layer of neurons into mean variance to achieve layer normalization.

[0014] According to one aspect of the present application, each of the first decoder, the second decoder and the third decoder includes: a first decoding sublayer connection structure, the first decoding sublayer connection structure includes a masked multi-head attention layer, a residual connection and a normalization layer; a second decoding sublayer connection structure, the second decoding sublayer connection structure includes a multi-head self-attention layer, a residual connection and a normalization layer; and a third decoding sublayer connection structure, the third decoding sublayer connection structure includes a forward feedback layer, a residual connection and a normalization layer. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The drawings are not drawn to scale. In the drawings, each identical or nearly identical component represented in different figures is represented by the same reference numeral. For purposes of clarity, not every component may be labeled in every drawing. In the drawings:

[0016] Figure 1 An overall schematic diagram of multi-air outlet distribution adjustment according to an embodiment of the present application is shown;

[0017] Figure 2 A schematic diagram of a single air outlet air outlet direction adjustment device according to an embodiment of the present application is shown;

[0018] Figure 3 A flowchart of a method for adaptively controlling a balance of multiple air outlets according to an embodiment of the present application is shown;

[0019] Figure 4 A structural block diagram of an encoding module and a decoding module according to an embodiment of the present application is shown;

[0020] Figure 5A , 5B They are shown respectively Figure 4 The structural block diagram of the encoder in the encoding module and the decoder in the decoding module;

[0021] Figure 6 A schematic diagram showing an arrangement of token feature sequence data according to an embodiment of the present application is shown;

[0022] Figure 7 A block diagram of model training steps according to an embodiment of the present application is shown;

[0023] Figure 8 A block diagram showing an application of a model in a system according to an embodiment of the present application; and

[0024] Fig. 9 Shows Figure 1 The structural block diagram of the controller in the cloud server is shown. DETAILED DESCRIPTION

[0025] Various specific embodiments of the present application will be described below with reference to the accompanying drawings which constitute a part of this specification. It should be understood that, where possible, the same or similar reference numerals used in this application refer to the same components.

[0026] Figure 1 An overall schematic diagram of multi-air outlet distribution adjustment according to an embodiment of the present application is shown.

[0027] like Figure 1As shown in , several workstations are set in the space, and different users are located at the corresponding workstations. A plurality of air-conditioning outlets arranged in an array are arranged above these workstations to adjust the ambient temperature around the corresponding workstations. Each air outlet corresponds to a controller, and the users in the space can adjust the parameters of the corresponding air-conditioning outlet through the controller, such as air volume, wind direction, temperature, humidity, etc. In addition to directly operating the corresponding controller, the user can also adjust the parameters of the corresponding air outlet through smart control devices such as mobile phones. For example, the user can install the corresponding control software on his mobile phone, and directly send an infrared signal to the corresponding air-conditioning outlet through the infrared controller on the mobile phone, thereby adjusting the operating parameters of the air outlet. Alternatively, the user's mobile phone can be connected to the corresponding controller via the network, send a control signal to the controller through the mobile phone, and then adjust the operating parameters of the corresponding air-conditioning outlet through the controller. Those skilled in the art should understand that the operating parameters of the air-conditioning outlet can also be adjusted in other ways. Different users may adjust the controller of the air outlet in their area multiple times, and may also adjust the controller of the air outlet in adjacent areas or farther areas, so as to meet the user's physical needs. As Figure 1 As shown in , the real-time adjustment parameters of the air-conditioning outlet controller set by the user can be transmitted to the cloud through the user's smart controller (such as a mobile phone) for storage and processing. Those skilled in the art should understand that the real-time adjustment parameters can also be transmitted to the cloud server in other ways. For example, the air outlet controller can have a network connection function, so that the real-time adjustment parameters can be directly transmitted to the cloud server through the network. Figure 1 As shown in , since there are multiple air-conditioning outlets in a space, the air blown out of a certain outlet may affect the temperature and humidity of the space in the adjacent area while meeting the temperature and humidity requirements of the users in the corresponding area, thereby causing the temperature and humidity in the adjacent area to deviate from the actual requirements of the users in the area. Therefore, in order to meet the actual needs of users, it is necessary to consider the mutual influence between the air-conditioning outlets in the entire space, and adaptively control each air-conditioning outlet according to the user's adjustment behavior.

[0028] Figure 2 A schematic diagram of a single air outlet air outlet direction adjustment device according to an embodiment of the present application is shown.

[0029] like Figure 2As shown in , a single air outlet sealing direction adjustment device includes a transverse blade adjustment device and a longitudinal blade adjustment device. The transverse blade adjustment device is used to adjust the wind direction of a group of blades arranged transversely, and the longitudinal blade adjustment device is used to adjust the wind direction of a group of blades arranged longitudinally. The transverse blade adjustment device and the longitudinal blade adjustment device are installed on the air outlet support plate 6. Among them, the transverse blade adjustment device includes: a transverse blade controller 7, a transverse motor 8, a transverse motor rotor 9, a transverse blade linkage actuator 10 and a transverse steering wind blade 11. The transverse blade controller 7 receives a wind direction adjustment signal from a user, generates a transverse blade adjustment signal based on the adjustment signal, and transmits the transverse blade adjustment signal to the transverse motor 8 through a circuit connection. The rotor 9 of the transverse motor 8 is mechanically connected to the transverse blade linkage actuator 10, and a group of transverse steering wind blades 11 are mechanically mounted on the transverse blade linkage actuator 10. When the transverse motor 8 receives the transverse blade adjustment signal from the transverse blade controller 7, the rotor 9 of the transverse motor 8 rotates clockwise or counterclockwise by a certain angle, thereby driving the transverse blade linkage actuator 10 to move in the longitudinal direction, so that a group of transverse steering wind blades 11 installed on the transverse blade linkage actuator 10 rotates around their respective rotation axes in the longitudinal direction by a certain angle, thereby adjusting the longitudinal air outlet direction of the air outlet. Similarly, the longitudinal blade adjustment device includes: a longitudinal blade controller 1, a longitudinal motor 2, a longitudinal motor rotor 3, a longitudinal blade linkage actuator 4 and a longitudinal steering wind blade 5. The longitudinal blade controller 1 receives the user's wind direction adjustment signal, generates a longitudinal blade adjustment signal based on the adjustment signal, and transmits the longitudinal blade adjustment signal to the longitudinal motor 2 through a circuit connection. The rotor 3 of the longitudinal motor 2 is mechanically connected to the longitudinal blade linkage actuator 4, and a group of longitudinal steering wind blades 5 are mechanically installed on the longitudinal blade linkage actuator 4. When the longitudinal motor 2 receives the longitudinal blade adjustment signal from the longitudinal blade controller 1, the rotor 3 of the longitudinal motor 2 rotates clockwise or counterclockwise by a certain angle, thereby driving the longitudinal blade linkage actuator 4 to move in the lateral direction, so that a group of longitudinal steering wind blades 5 installed on the longitudinal blade linkage actuator 4 rotates in the lateral direction around their respective rotation axes by a certain angle, thereby adjusting the lateral air outlet direction of the air outlet. Those skilled in the art should understand that the above-mentioned lateral blade adjustment device and the longitudinal blade adjustment device can be controlled independently or in linkage, and "lateral" and "longitudinal" are only distinguishing names for distinguishing the two groups of blades that are interlaced with each other.

[0030] Figure 3 A flowchart of a method for adaptively controlling a plurality of air outlet balances according to an embodiment of the present application is shown.

[0031] According to one embodiment of the present application, a space includes several Figure 2In order to avoid the user adjusting the operating parameters of the air outlet in a certain area and affecting the physical sensation of users in adjacent areas, the present application considers the mutual influence between all the air outlets in the entire enclosed space and comprehensively controls all the air outlets based on the multi-air outlet balance adaptive control method. Figure 3 As shown in , the multi-outlet balanced adaptive control method of the present application includes the following steps:

[0032] 301: When different control panels are turned on, they send requests to the cloud server to obtain the historical setting data of the floor and room corresponding to the control panel, and the control panel controls the operation of the corresponding air-conditioning outlet according to the obtained historical setting data. For example, when a user turns on the air-conditioning outlet control panel of the conference room on the second floor of the office building, the control panel will send a request signal to the cloud server, and the request signal includes location information associated with the location of the control panel (such as the floor number and room number). Based on the location information, the cloud server retrieves the historical setting data associated with the location information from the database, and transmits the retrieved historical setting data to the control panel. The control panel controls the corresponding air-conditioning outlet to start operating according to the obtained historical setting data.

[0033] 302: The cloud server obtains user behavior data and unit operation parameters generated by the air outlet adjustment control in the corresponding floor and room, and calculates the air volume temperature and body sensation requirements at each air outlet position based on the above data obtained.

[0034] The calculation process of the air volume temperature G(x, y) at a certain air outlet position in the space is as follows:

[0035] The heat and cold of the air outlet in the space affects the temperature of the surrounding space. It diffuses around the air outlet. The heat and cold of the air outlet towards the center arrives first and the temperature is closest to the initial temperature. The coldness will weaken as the distance increases, and the speed and temperature will change, so it will also affect the parameter settings of other air outlets around it to varying degrees. i，j The calculation is as follows:

[0036] f i，j (x,y,θ,v,t,T db , R H )

[0037] Where x is the horizontal coordinate of the air outlet in space, y is the vertical coordinate of the air outlet in space, θ is the angle of the air outlet blade, v is the wind speed at the air outlet, t is the real time, T db is the outdoor dry-bulb temperature of the day, R H is the outdoor relative humidity for the day.

[0038] After multiple users open the air outlets at their locations, they adjust the air outlet parameters of the air conditioner through the controller to meet their own physical needs. After multiple multi-user adaptive adjustments, the air supply and user needs are finally balanced. At any position (x, y) in the room, the location is affected by the superposition of the cold and heat of many surrounding air outlets. This application takes the geometric superposition of the cold and heat corresponding to all air outlets in the indoor space:

[0039]

[0040] Among them, x is the horizontal coordinate of the air outlet in space, y is the vertical coordinate of the air outlet in space, θ is the angle of the air outlet blade, v is the air outlet wind speed, t is the air outlet time, T is the specific time, and N is the number of air outlet adjustments.

[0041] The calculation process of the user's body sensation demand (heat and cold speed demand) is as follows:

[0042] The user's demand for the amount of heat and cold in the space depends on the user's specific physical condition. Different physical needs determine the amount of heat and cold and the speed of heat and cold required in the user's area.

[0043] u(x+Δx,y+Δy,θ,v,t,T db , R H )

[0044] Among them, x is the horizontal coordinate of the air outlet in space, y is the vertical coordinate of the air outlet in space, θ is the angle of the air outlet blade, v is the outlet wind speed, t is the real time, and the outdoor dry bulb temperature of the day is T db , outdoor relative humidity R H .

[0045] 303: The cloud server generates a control instruction (for example, a control signal generated based on a PID algorithm, which may include the location information of the air outlet, the angle information of the air outlet blades, the wind speed information of the air outlet, etc.) according to the calculated air volume temperature and body sensory demand at each air outlet position, and sends the generated control instruction to the corresponding air conditioning outlet control panel.

[0046] The control instruction is generated based on the air volume temperature G(x, y) at a certain air outlet position calculated in step S2 and the user's body feeling demand (heat and cold speed demand). The difference between the sum of the heat and cold at the specified position and the heat and cold demand of the user at that position is as follows:

[0047] Δε=G(x,y)-u(x+Δx,y+Δy,θ,v,t,T db , R H )

[0048] The closer Δε is to 0, the more the cooling capacity provided by the multi-air outlet can meet the user's cooling demand. These cooling demand and supply are related to the parameters x, Δx, y, Δy, θ, v, t, T db , R H Among them, x is the horizontal coordinate of the air outlet in space, Δx is the displacement from the horizontal coordinate of the air outlet, y is the vertical coordinate of the air outlet in space, Δy is the displacement from the vertical coordinate of the air outlet, θ is the angle of the air outlet blade, v is the wind speed at the air outlet, t is the air outlet time, T is the specific time, T db is the outdoor dry-bulb temperature of the day, R H is the outdoor relative humidity of the day. The user adjusts the air volume at different times to meet the body's needs at different times by continuously adjusting these controller data. This demand can be static or dynamically change with time and space. The control command is generated based on the operating parameters (such as the outlet wind speed) when Δε is close to 0.

[0049] 304: The air outlet control panel of the air conditioner receives the control instruction from the cloud server, and controls the operation of each actuator of the fan according to the received control instruction, such as setting the operating parameters of the fan such as temperature, air volume, wind direction, etc.;

[0050] 305: The user adjusts the operating parameters of the air outlet at the location through the control panel according to the actual needs of the user, and the control panel uploads the user's real-time adjustment parameters to the cloud server;

[0051] 306: The cloud server optimizes the operating parameters of other air outlets using a multi-air outlet balance adaptive model according to the historical setting parameters in the database and the real-time adjustment parameters received from the user, and obtains recalculated operating parameters;

[0052] 307: Confirm the recalculated operating parameters (for example, the sensor confirms whether the angle θ of the air outlet blades, the air outlet wind speed v and other parameters after the recalculated operating parameters are executed are consistent with the control operating parameters), and upload the recalculated operating parameters to the cloud database.

[0053] It should be noted that the above steps are shown only as an embodiment, and those skilled in the art may add, combine or omit certain steps according to actual needs, as long as the multi-outlet balanced adaptive control method of the present application can be implemented.

[0054] Figure 4 A structural block diagram of an encoding module and a decoding module according to an embodiment of the present application is shown.

[0055] Because multiple air vents influence each other, that is, they are correlated, the influence of the air vents around a person on the person gradually weakens from near to far. The specific cold and heat model is relatively complex and difficult to simulate using simple formulas and real-time simulation. This application uses the Transformer model to learn the multi-user operation rules.

[0056] The Transformer model can effectively represent and fuse multiple features such as user behavior, product parameters, and environmental parameters. Through the self-attention mechanism, the model can capture the spatial and temporal correlations between different features and generate richer feature representations, thereby improving the prediction effect. The user behavior parameters of the multi-air vent control system are constructed into a token feature sequence form, where x is the horizontal coordinate of the air vent in space, y is the vertical coordinate of the air vent in space, θ is the angle of the air vent blade, v is the air outlet wind speed, t is the air outlet time, T is the specific duration, N is the number of air vent adjustments, and T db is the outdoor dry-bulb temperature of the day, R H Token feature sequences are formed for many parameters such as outdoor relative humidity. The encoder of the Transformer model is used to identify the user's behavior of changing the current multi-vent controller, and the decoder is used to predict the upcoming multi-vent actuator state parameters to capture the interdependencies in the token feature sequence.

[0057] like Figure 4 As shown in , the internal structure of the Transformer model used in this application is divided into three parts: an encoding module, a decoding module, and a probability calculation module. The encoding module is used to encode the parameter token feature sequence data of multiple outlets and output the encoding vector, and the encoding module is composed of multiple encoders. The decoding module is used to decode the encoding vector output by the encoding module and output the probability vector, and the decoding module is composed of multiple decoders. The probability calculation module is used to process the probability vector output by the decoding module to output the parameter token feature sequence data of the outlet with the largest probability value. Figure 4 As shown in , the model of this application specifically uses 3 encoders and 3 decoders. In the encoding module part, the output of each encoder is used as the input of the next encoder; in the decoding module, the output of each decoder is combined with the output of the entire encoding module as the input of the next decoder. Among them, the input multi-vent parameter token feature sequence will first pass through the encoder, which encodes the token feature sequence data; the encoded token feature sequence data is then passed to the decoder, and the token feature sequence data is decoded to obtain the final result.

[0058] like Figure 4As shown in , the input multi-vent parameter token feature sequence passes through the input encoding layer, and then the sine and cosine function token feature sequence information is used for position encoding to obtain the position encoding vector of the encoded input data. In order to calculate the encoding corresponding to the position of different token feature sequence positions, the Transformer model of this application uses sine and cosine functions of different frequencies as shown below:

[0059] Positional encoding vector:

[0060] In the formula, pos represents the position of the word in the token feature sequence, d represents the dimension of the position vector PE (the same as the word vector dimension), 2i represents the even dimension, 2i+1 represents the odd dimension (2i<d, 2i+1<d), and the value of 10000 is selected based on the experience and experimental analysis of the length of the token feature sequence and the position encoding dimension. The position encoding vector is input into the encoder block to generate a d-dimensional vector for use in the decoder block. The encoding method of the output data of the Transformer model is the same as the encoding method of the input data of the Transformer model, thereby obtaining the position encoding vector of the output data. The position encoding vector of the decoder output data and the d-dimensional vector generated by the encoder block are used together to generate a vector containing probability information to calculate the output. When all the decoder blocks are executed, the output probability is calculated through the linearization layer and the SoftMax layer. The one with the largest probability value is the corresponding parameter of the next result of the output token feature sequence.

[0061] Figure 5A , 5B They are shown respectively Figure 4 Structural block diagram of the encoder in the encoding module and the decoder in the decoding module.

[0062] like Figure 5A As shown in , each encoder contains two layers. The first sub-layer connection structure includes a multi-head self-attention layer, a normalization layer, and a residual connection; the second sub-layer connection structure includes a feedforward fully connected layer, a normalization layer, and a residual connection. The multi-head self-attention mechanism and the feedforward neural network are connected through residual connections and layer normalization. The multi-head self-attention layer of the first sub-layer interacts each position in the input token feature sequence with all other positions to calculate the context representation vector of each position. The feedforward fully connected layer maps the context representation vector of each position to another vector space to capture higher-level features.

[0063] like Figure 5BAs shown in the figure, the structure of the decoding part is similar to that of the encoding part, which are the masked multi-head attention layer, the result is input to the residual and normalization layer, the result is input to the multi-head attention layer, the result is input to the residual and normalization layer, the result is input to the feedforward network layer, the result is input to the residual and normalization layer, the result is input to the linear layer, and the output probability is calculated through the SoftMax layer.

[0064] Figure 5A and Figure 5B The input of the self-attention layer in is the word vector X multiplied by three trained weight matrices available Three matrices:

[0065]

[0066] After getting the input word vector matrix X corresponding to After that, we can calculate the output of the self-attention layer.

[0067] The calculation formula is as follows:

[0068] Where, d k is the number of columns of the K matrix, that is, the dimension of the K vector.

[0069] Afterwards, the output matrix of the multi-head self-attention layer is the input of the residual and normalization layer. The calculation formula is as follows:

[0070] LayerNorm(X+MultiHeadAttetion(X)),

[0071] LayerNorm(X+FeedForward(X))

[0072] In the formula, X represents the input of MultiHeadAttetion or FeedForward, MultiHeadAttetion(X) and FeedForward(X) represent the output, X+FeedForward(X) is a residual connection, and LayerNorm converts the input of each layer of neurons into mean variance to achieve layer normalization and accelerate network convergence. Afterwards, the output of the residual and normalization layer is used as the input of the feedforward network layer. The feedforward network here uses a two-layer fully connected network and a nonlinear activation function. The two-layer fully connected network implements two linear transformations. The first layer uses the RELU activation function, and the second layer does not use an activation function. Afterwards, the output of the feedforward network layer is used as the input of the new residual and normalization layer.

[0073] Figure 6 A schematic diagram showing an arrangement of token feature sequence data according to an embodiment of the present application is shown.

[0074] Since the cooling capacity of the air outlets to the users is related, x is the horizontal coordinate of the air outlet in space, y is the vertical coordinate of the air outlet in space, θ is the angle of the air outlet blade, v is the outlet wind speed, t is the specific year, month, day, hour, minute, and second, and the outdoor dry bulb temperature T of the day is db , outdoor relative humidity R H , all have an impact on the user's experience of heat and cold. These 7 parameter influencing factors are used as input features of the transformer model. The output data is the key parameters of the vents with good user experience. Whether the output data is good depends on whether the users are satisfied after multiple users frequently operate the vent controller parameters in a short period of time. If the adjustment is always ongoing, it can be determined that the user experience of heat and cold is not good. If the number of parameter adjustments is limited and there is a stable time stagnation, it is considered that the user experience of heat and cold is good. These data are divided into long-term and short-term. Long-term data can be used to find users' stable experience habits when the space is in a long-term stable state. Short-term data is used to find users' stable experience habits in a recent stable state, and real-time adjustments are made due to changes in the number of people.

[0075] According to the above input and output data, a data set with a token feature sequence information pattern is established, specifically the historical data of the 7 parameters of the vent within M unit time, of which the model input data is selected according to the periodic time nodes set in the historical data, and the behavioral data uploaded when operating the vent controller is additionally added. The Transformer model captures the dependencies between elements in the token feature sequence through the self-attention mechanism. The self-attention mechanism allows the model to consider the information of the entire input token feature sequence when generating the output of each position. Therefore, even if the positions of two elements in the token feature sequence are far apart, if they have a strong correlation in semantics or content, the Transformer model can capture this relationship, sort these data in position order and time order, and establish a data set with a token feature sequence information pattern.

[0076] like Figure 6 As shown in , the arrangement of the near and far token feature sequence data of the input data in this application is sorted by row. In the same row, each vent parameter token feature sequence is followed by the next vent parameter token feature sequence in the row. The end of a row of vents is followed by the first vent in the second row, and so on, until the last vent parameter.

[0077] The data of each outlet is spliced ​​into a long token feature sequence. The input data single outlet parameter token feature sequence is [x, y, θ, v, t, T db , R H ], then the multi-vent parameter token feature sequence is:

[0078] Among them, the operator is the parameter token feature sequence arranged in ascending order according to the Euclidean distance, M is the total number of outlets, θ i is the tuyere angle of the tuyere controller, t is the specific time, is the outdoor dry bulb temperature, is the outdoor relative humidity.

[0079] The output data single vent parameter token feature sequence is [x, y, θ′, v′], then the multi-vent parameter token feature sequence is Among them, θ′ i is the actual output air outlet angle, v′ i is the actual output wind speed of the air outlet, x is the horizontal coordinate of the air outlet, and y is the vertical coordinate of the air outlet.

[0080] Figure 7 A block diagram of the model training steps according to an embodiment of the present application is shown.

[0081] like Figure 7 As shown in the example, in the prepared dataset, the dataset is dynamically split, with 80% of the data used as training data, 15% of the data used as test data, and the remaining 5% used as validation data. The order of the dataset is shuffled using a random number generator. This ensures that the distribution of the data is random each time it is split. By constantly re-splitting the dataset, it is ensured that the model faces a different data distribution each time it is trained. Then train the model as follows:

[0082] a) Forward propagation: pass the preprocessed input data to the Transformer model to obtain the output of the model.

[0083] b) Calculate loss: Compare the output of the model with the true label and calculate the loss function. This helps evaluate the performance of the model in the current iteration.

[0084] c) Back propagation: Based on the calculated loss, the gradient is calculated and the parameters of the model are updated through the back propagation algorithm.

[0085] d) Parameter update: Use the optimizer to update the model's parameters according to the gradient. The choice of optimizer is crucial to the training speed and stability of the model.

[0086] e) Reset Gradients: After each iteration, the gradients are reset for the next iteration.

[0087] This application uses the data recorded in the database within a period of P days as training data. These data record the parameter data of all air outlets every Q minutes. When the parameters of the air outlet controller are found to have changed, the parameter data of all controllers are also recorded. The parameter data recorded when the air outlet controller parameters change are used as the input data of the Transformer model; the parameter data of the air outlet whose state parameters after adjustment are maintained for a time greater than or equal to the preset time R minutes are obtained from the database, and these air outlet parameter data are used as the output data of the Transformer model.

[0088] Figure 8 A block diagram of an application of a model in a system according to an embodiment of the present application is shown.

[0089] like Figure 8 As shown in , the user adjusts the operating parameters of the air outlet through the air outlet control device or the APP in the smart device, and these operating parameters are transmitted to the cloud server and the Transformer model. The Transformer model generates optimized operating parameters for multiple air outlets based on the user's historical setting parameters stored in the cloud server and the user's real-time adjustment parameters, and sends the optimized operating parameters to the control devices and actuators of the multiple air outlets. The actuator controls the operation of each air outlet according to these optimized parameters (for example, re-fine-tune the distribution of the wind speed of each air outlet and the angle of the air outlet plate), and feeds back the optimized operating parameters to the cloud server.

[0090] Fig. 9 Shows Figure 1 The structural block diagram of the controller 900 in the cloud server is shown.

[0091] like Fig. 9 As shown, the controller 900 includes a bus 902, a processor 904, a memory 906, an input interface 908, and an output interface 910. The processor 904, the memory 906, the input interface 908, and the output interface 910 are connected to the bus 902. The processor 904 can read a program (or instruction) from the memory 906 and execute the program (or instruction) to perform the processing of data. The processor 904 can also write data or a program (or instruction) into the memory 906. The memory 906 can store programs (instructions) or data. By executing the instructions in the memory 906, the processor 904 can control the memory 906, the input interface 908, and the output interface 910.

[0092] The input interface 908 is configured to receive historical setting parameter data from the cloud database and real-time adjustment parameter data from the user. The input interface 908 is also configured to convert the received data and signals into data and signals recognizable by the processor 904 and output the data and signals to the processor 904.

[0093] The processor 904 is configured to process the received data and signals to generate control signals. In one embodiment, the processor 904 generates control signals based on the memory 906. Figure 4 The Transformer model of the encoding and decoding module shown in the figure processes the historical setting parameter data received from the cloud database and the real-time adjustment parameter data from the user to generate recalculated operating parameters for each air outlet.

[0094] The output interface 910 is configured to receive the recalculated operating parameters from the processor 904, and send the recalculated operating parameters to the corresponding control panels of each air outlet to control the operating parameters of each air outlet.

[0095] The user behavior-based air terminal multi-outlet balance adaptive control method and device of the present application have the following advantages:

[0096] First, the adaptive control method of the present application takes into account the mutual influence between different air outlets in the entire space, so each air outlet can be well adjusted, avoiding the change of operating parameters of certain air outlets causing the user's perception in adjacent areas to deviate from the actual needs;

[0097] Second, the control method of the present application can meet the real needs of many users or objects in different positions of the environment space by modifying and adjusting the state of the air outlet, thereby improving the user experience and saving energy, which meets the current energy-saving indicators;

[0098] Third, the encoder and decoder in the Transformer model used in this application both have three sub-layers, which improves the computational efficiency while ensuring the accuracy of the algorithm and can obtain a new optimal solution more quickly.

[0099] Although the present application has been described in conjunction with the examples of the embodiments summarized above, various alternatives, modifications, changes, improvements and / or substantially equivalent solutions, whether known or currently or foreseeable in the near future, may be apparent to those skilled in the art. In addition, the technical effects and / or technical problems described in this specification are exemplary rather than restrictive; so the disclosure in this specification may be used to solve other technical problems and have other technical effects and / or may solve other technical problems. Therefore, the examples of the embodiments of the present application as stated above are intended to be illustrative rather than restrictive. Various changes may be made without departing from the spirit or scope of the present application. Therefore, the present application is intended to include all known or earlier developed alternatives, modifications, changes, improvements and / or substantially equivalent solutions.

Claims

1. A multi-air outlet adaptive control method (300), characterized in that: The method (300) comprises: S1: acquiring, from a database, a historical setting parameter associated with at least one control panel among a plurality of control panels, wherein one control panel among the plurality of control panels corresponds to one of the plurality of air outlets; S2: controlling the operation of at least one air outlet corresponding to the at least one control panel based on the historical setting parameters; S3: acquiring a user real-time adjustment parameter through at least one of the multiple control panels, and storing the user real-time adjustment parameter in the database; S4: using a multi-air outlet balance adaptive model to process the historical setting parameters and the user real-time adjustment parameters in the database to obtain operating parameters for the multiple air outlets; and S5: Controlling the operation of the plurality of air outlets based on the operation parameters of the plurality of air outlets, and storing the operation parameters of the plurality of air outlets in the database.

2. The multi-air outlet adaptive control method (300) according to claim 1, characterized in that: In step S4, using the multi-outlet balance adaptive model to process the historical setting parameters and the user real-time adjustment parameters in the database includes: S4.1: Calculate the total amount of cold and heat at the location of the at least one air outlet according to the historical setting parameters Wherein, x is the horizontal coordinate of the air outlet in space, y is the vertical coordinate of the air outlet in space, θ is the angle of the air outlet blade, v is the wind speed of the air outlet, t is the air outlet time, T is the specific time, N is the number of times the air outlet is adjusted, m is the number of rows where the air outlet is located, and n is the number of columns where the air outlet is located; S4.2: Calculate the user's cooling and heating demand at the location of the at least one air outlet according to the user's real-time adjustment parameter u(x+Δx,y+Δy,θ,v,t,T db ,R H ), Where x is the horizontal coordinate of the air outlet in space, Δx is the displacement from the horizontal coordinate of the air outlet, y is the vertical coordinate of the air outlet in space, Δy is the displacement from the vertical coordinate of the air outlet, θ is the angle of the air outlet blade, v is the wind speed at the air outlet, t is the real time, T db is the outdoor dry-bulb temperature of the day, R H is the outdoor relative humidity; and S4.3: Calculate the difference between the total amount of cooling and heating and the cooling and heating demand of the user Δε=G(x,y)-u(+Δx,y+Δy,θ,v,t,T db ,R H )。 3. The multi-air outlet adaptive control method (300) according to claim 1, characterized in that: In step S4, using the multi-outlet balance adaptive model to process the historical setting parameters and the user real-time adjustment parameters in the database includes using a Transformer model to process the historical setting parameters and the user real-time adjustment parameters, and the Transformer model includes: An encoding module, the encoding module is configured to encode the parameter token feature sequence data of the plurality of air outlets and output an encoding vector, the encoding module comprising a first encoder, a second encoder and a third encoder; a decoding module, the decoding module being configured to decode the encoding vector output by the encoding module and output a probability vector, the decoding module comprising a first decoder, a second decoder and a third decoder; and A probability calculation module is configured to process the probability vector output by the decoding module to output parameter token feature sequence data of the air outlet with the maximum probability value.

4. The multi-air outlet adaptive control method (300) according to claim 3, characterized in that: Using the encoding module to encode the parameter token feature sequence data of the plurality of air outlets comprises: The parameter token feature sequences of the plurality of air outlets are position-encoded using sine and cosine functions to obtain position codes of the parameter token feature sequences of the plurality of air outlets. where pos represents the position of the word in the token feature sequence, d represents the dimension of the position vector PE, 2i represents an even dimension, and 2i + 1 represents an odd dimension (2i < d, 2i + 1 < d); and Generate an encoding vector of dimension d based on the generated position encoding.

5. The multi-air outlet adaptive control method (300) according to claim 4, characterized in that: Using the decoding module to decode the encoded parameter token feature sequence data includes: Generate a vector containing probability information based on the encoding vector generated by the encoding module and the position encoding vector output by the decoding module.

6. The multi-air outlet adaptive control method (300) according to claim 5, characterized in that: Using the probability calculation module to process the probability vector output by the decoding module includes: Calculating the output probability of the vector containing probability information output by the decoding module through a linearization layer and a SoftMax layer to output the parameter token feature sequence data of the air outlet with the maximum probability value.

7. The multi-air outlet adaptive control method (300) according to claim 3, characterized in that: Each of the first encoder, the second encoder, and the third encoder includes: A first encoding sub-layer connection structure, the first encoding sub-layer connection structure includes a multi-head attention layer and residual connection and normalization layers; and A second encoding sub-layer connection structure, the second encoding sub-layer connection structure includes a pre-feedforward layer and residual connection and normalization layers.

8. The multi-air outlet adaptive control method (300) according to claim 7, characterized in that: The input of the multi-head attention layer is three matrices: Where X is the word vector, are three trained weight matrices; The output of the multi-head attention layer is: Among them, d k is the number of columns of the K matrix, that is, the dimension of the K vector.

9. The multi-air outlet adaptive control method (300) according to claim 8, characterized in that: The input of the residual connection and normalization layer is: LayerNorm(X + MultiHeadAttetion(X)) Layerorm(X + FeedForward(X)) where X represents the input of MultiHeadAttetion or FeedForward, MultiHeadAttetion(X) and FeedForward(X) represent the outputs, X + FeedForward(X) is a residual connection, and LayerNorm converts the input of each layer of neurons into mean variance to achieve layer normalization.

10. The multi-air outlet adaptive control method (300) according to claim 3, characterized in that: Each of the first decoder, the second decoder, and the third decoder includes: A first decoding sub-layer connection structure, the first decoding sub-layer connection structure includes a masked multi-head attention layer and residual connection and normalization layers; A second decoding sub-layer connection structure, the second decoding sub-layer connection structure includes a multi-head self-attention layer and residual connection and normalization layers; and A third decoding sub-layer connection structure, the third decoding sub-layer connection structure includes a pre-feedforward layer and residual connection and normalization layers.

11. A multi-air outlet adaptive control device, characterized in that: The control device includes: One or more control panels, each of the one or more control panels is configured to control the operation of a corresponding one of the multiple air outlets according to the received operating parameters; A processing device, the processing device is communicatively connected to the one or more control panels and is configured to send the operating parameters to the one or more control panels; wherein, the processing device includes a processor, and the processor is configured to run the control method according to any one of claims 1-10.

Images