Air conditioner, control method of air conditioner and computer equipment

Abstract

The invention relates to an air conditioner, a control method of the air conditioner and computer equipment, and the air conditioner comprises an indoor unit used for adjusting indoor air; the outdoor unit is connected with the indoor unit; the controller is configured to respond to a voice control instruction and convert the voice control instruction into a control instruction text; current environment information and an air conditioner state code of the air conditioner are obtained, and an air conditioner control code and a target interaction text are obtained according to the control instruction text, the current environment information and the air conditioner state code; the operation state of the air conditioner is controlled according to the air conditioner control code, the loudspeaker module of the indoor unit is controlled to broadcast the target interaction text, the hardware equipment state of the air conditioner does not need to be represented by a natural language text, the problem of information loss or misunderstanding caused by the text is avoided, the accuracy of obtaining the hardware equipment state of the air conditioner is improved, and the user experience is improved. And the control efficiency of the air conditioner is improved, so that the intelligent level of the air conditioner is improved.

Description

Technical Field

[0001] This application relates to the field of air conditioner technology, and in particular to an air conditioner, an air conditioner control method, and a computer device. Background Technology

[0002] Air conditioners equipped with voice control allow users to control various functions of the air conditioner through voice commands, thus freeing up their hands and eliminating the need for a remote control.

[0003] However, in traditional technologies, air conditioners with voice control functions often rely on fixed command libraries or simple semantic parsing models. This means that air conditioners can only recognize structured and explicit voice control commands (such as "turn on the air conditioner" or "set the temperature to 26°C"). They have a weak ability to understand ambiguous commands, implicit needs, or commands for complex scenarios. Furthermore, the execution of voice control commands is mostly a direct mapping of "command-action" without fully adapting to the actual usage scenarios, resulting in insufficient intelligence of the air conditioner.

[0004] Therefore, air conditioners using traditional technology suffer from a low level of intelligence. Summary of the Invention

[0005] Therefore, it is necessary to provide an air conditioner, an air conditioner control method, and a computer device that can improve the intelligence level of air conditioners, in response to the above-mentioned technical problems.

[0006] In a first aspect, this application provides an air conditioner, comprising:

[0007] Indoor unit, used to regulate indoor air;

[0008] The outdoor unit is connected to the indoor unit;

[0009] The controller is configured as follows:

[0010] In response to voice control commands, convert the voice control commands into control command text;

[0011] Obtain the current environmental information and the air conditioner status code. Based on the control command text, the current environmental information, and the air conditioner status code, obtain the air conditioner control code and the target interaction text.

[0012] The operating status of the air conditioner is controlled according to the air conditioner control code, and the outdoor speaker module of the indoor unit is controlled to broadcast the target interactive text.

[0013] The above technical solution has the following advantages or effects: By acquiring the air conditioner's status code, which directly represents the hardware status of the air conditioner, there is no need to use natural language text to represent the hardware status, thus avoiding information loss or misunderstanding caused by text and improving the accuracy of acquiring the air conditioner's hardware status. The air conditioner control code can be directly sent to the air conditioner's hardware without converting the air conditioner control commands into commands that the hardware can recognize, which is more in line with the underlying hardware logic, improving the execution speed of air conditioner control. Furthermore, by broadcasting the target interactive text through the external speaker module, users can quickly understand the operating status of the air conditioner, thereby improving the intelligence level of the air conditioner.

[0014] In one embodiment, the controller is configured to:

[0015] The control command text and current environment information are input into the pre-trained first language model to generate the first implicit features;

[0016] The air conditioner status code is converted into the corresponding second implicit feature. The first and second implicit features are then input into a pre-trained second language model to generate the air conditioner control code and the target interactive text.

[0017] The air conditioning control code and the air conditioning status code use the same preset coding format.

[0018] The above technical solution has the following advantages or technical effects: The embodiments of this application are based on a two-stage model architecture. The pre-trained first language model serves as the first-stage model, focusing on processing control command text and current environmental information. It can capture implicit needs in user commands (for example, when a user says "it's a bit cold," the system can determine that the set temperature needs to be raised, rather than simply turning off the air conditioner), reducing misoperations caused by ambiguous commands and improving the accuracy and adaptability of voice control. The pre-trained second language model serves as the second-stage model, focusing on processing the first and second implicit features, enabling the generated air conditioner control code to adapt to the current device state of the air conditioner. Compared to directly inputting all data into a large language model, the two-stage model architecture can decompose complex tasks into two relatively simple sub-tasks. Each stage of the model focuses on a specific sub-task, improving the quality and accuracy of model processing at each stage.

[0019] In one embodiment, the air conditioner further includes a communicator, through which the controller establishes a communication connection with the cloud server; the controller is configured to:

[0020] Send control command text, current environmental information, and air conditioning status code to the cloud server;

[0021] Receive the air conditioning control code and target interactive text returned by the cloud server;

[0022] The cloud server is configured as follows:

[0023] The control command text and current environment information are input into the pre-trained first language model to generate the first implicit features;

[0024] The air conditioner status code is converted into the corresponding second implicit feature. The first and second implicit features are then input into a pre-trained second language model to generate the air conditioner control code and the target interactive text. The air conditioner control code and the air conditioner status code use the same preset coding form.

[0025] The above technical solution has the following advantages or technical effects: By deploying a pre-trained first language model and a pre-trained second language model on a cloud server, the high computing power of the cloud server can be utilized to improve the running efficiency of the model and speed up the response speed of the air conditioner to user control commands.

[0026] In one embodiment, the preset encoding format includes a multi-hot encoding format.

[0027] The above technical solution has the following advantages or effects: The embodiments of this application use a multi-thermal coding method to represent air conditioner status codes, which clearly defines different air conditioner status categories, avoids category confusion caused by numerical coding, and makes model recognition easier. Using a multi-thermal coding method to represent air conditioner control codes allows for direct recognition by the air conditioning equipment, improving the control efficiency of the air conditioner.

[0028] In one embodiment, the target interactive text includes at least execution instruction text; the instruction semantics of the execution instruction text matches the instruction semantics of the air conditioning control code.

[0029] The above technical solution has the following advantages or technical effects: the instruction semantics of the execution instruction text in this application embodiment match the instruction semantics of the air conditioner control code, which allows the user to accurately understand the current operating status of the air conditioner.

[0030] In one embodiment, the step of converting the air conditioner state code into a corresponding second implicit feature includes:

[0031] The air conditioner status code is normalized to obtain the normalized code.

[0032] Adjust the original dimension of the normalized encoding to the preset dimension to obtain the target dimension encoding;

[0033] The target dimension encoding is input into a pre-trained air conditioner state encoder to obtain the second implicit feature.

[0034] The above technical solution has the following advantages or technical effects: The embodiments of this application perform normalization processing, dimension adjustment and encoding processing on the air conditioner status code, which can convert the air conditioner status code into a high-dimensional feature vector, which is convenient for subsequent second language model processing.

[0035] In one embodiment, the first implicit feature and the second implicit feature are input into a pre-trained second language model to generate air conditioning control codes and target interactive text, including:

[0036] The first implicit feature and the second implicit feature are fused to obtain the fused implicit feature;

[0037] The fused implicit features are input into a pre-trained second language model to generate air conditioning control codes and target interactive text.

[0038] The above technical solution has the following advantages or technical effects: By fusing the first implicit feature and the second implicit feature, the second language model can fully understand the information carried by the low-frequency feature and the high-frequency feature, thereby improving the accuracy of generating air conditioning control codes and target interactive text.

[0039] In one embodiment, control command text and current environment information are input into a pre-trained first language model to generate first implicit features, including:

[0040] Combine the control command text, current environment information, and preset system prompts to obtain the target prompt;

[0041] The target prompt is input into the pre-trained first language model to generate the first implicit feature;

[0042] The system prompts are used to configure the response behavior patterns of the pre-trained first language model.

[0043] The above technical solution has the following advantages or technical effects: By constructing target prompt words, the embodiments of this application can guide the first language model to generate highly accurate first implicit features.

[0044] Secondly, this application also provides a control method for an air conditioner, applied to an air conditioner including an indoor unit, an outdoor unit, and a controller, the method including:

[0045] In response to voice control commands, convert the voice control commands into control command text;

[0046] Obtain the current environmental information and the air conditioner status code. Based on the control command text, the current environmental information, and the air conditioner status code, obtain the air conditioner control code and the target interaction text.

[0047] The operating status of the air conditioner is controlled according to the air conditioner control code, and the outdoor speaker module of the indoor unit is controlled to broadcast the target interactive text.

[0048] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method steps of the second aspect.

[0049] The aforementioned air conditioner, its control method, and computer equipment acquire the air conditioner's status code. Since this status code directly represents the air conditioner's hardware status, it eliminates the need for text-based representations, avoiding information loss or misunderstanding associated with text and improving the accuracy of hardware status acquisition. The air conditioner control code can be directly sent to the air conditioner's hardware without converting control commands into hardware-readable instructions, better aligning with underlying hardware logic and increasing control execution speed. Furthermore, by broadcasting interactive text through an external speaker module, users can quickly understand the air conditioner's operating status, enhancing its intelligence level. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is a structural block diagram of an air conditioner in one embodiment;

[0052] Figure 2 This is a schematic diagram of the control method for an air conditioner in one embodiment;

[0053] Figure 3 This is a model architecture diagram in one embodiment;

[0054] Figure 4 This is a schematic diagram illustrating the training of a second language model in one embodiment;

[0055] Figure 5 This is a schematic diagram of the control method for an air conditioner in another embodiment;

[0056] Figure 6 This is a system architecture diagram of an air conditioner and a cloud server in one embodiment;

[0057] Figure 7 This is a signal timing interaction diagram between the air conditioner and the cloud server in one embodiment;

[0058] Figure 8 This is a schematic diagram of the process for generating the first implicit feature in one embodiment;

[0059] Figure 9 This is a schematic diagram of the process for obtaining the second implicit feature in one embodiment;

[0060] Figure 10 This is a schematic diagram of the architecture of an air conditioner status encoder in one embodiment;

[0061] Figure 11 This is a flowchart illustrating the process of generating air conditioning control codes and target interactive text in one embodiment;

[0062] Figure 12 This is a flowchart illustrating the control method of an air conditioner in one embodiment;

[0063] Figure 13 This is a structural block diagram of the control device of an air conditioner in one embodiment;

[0064] Figure 14 This is a structural block diagram of the first implicit feature generation module in one embodiment;

[0065] Figure 15 This is a structural block diagram of the air conditioning control code generation module in one embodiment;

[0066] Figure 16 This is a structural block diagram of the air conditioning control code generation module in another embodiment;

[0067] Figure 17 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0069] An air conditioner, short for room air conditioner, is a compact local air conditioning system assembled from air handling equipment, fans, refrigeration units, and automatic control instruments. The air handling equipment filters the air, regulates temperature, controls humidity, and improves air cleanliness. It includes filters, evaporators, condensers, humidification modules, and dehumidification modules. Fans drive air circulation, ensuring efficient indoor and outdoor air and heat exchange. Fans come in various types, including centrifugal fans, axial fans, and cross-flow fans. The refrigeration unit implements a vapor compression refrigeration cycle (compression → condensation → throttling → evaporation). It includes a compressor, refrigerant, and expansion valve. Automatic control instruments automatically regulate temperature, humidity, and fan speed, and provide fault protection. These instruments include temperature sensors, humidity sensors, pressure sensors, microcontrollers (MCUs) or programmable logic controllers (PLCs), electric dampers, electronic expansion valves, and compressor inverter modules.

[0070] Air conditioners can be divided into packaged air conditioners and split air conditioners. In a packaged air conditioner, all components are integrated into one casing (such as window air conditioners and portable air conditioners). In a split air conditioner, the indoor unit and the outdoor unit are separate and connected by pipes. The indoor unit is used for air handling and ventilation, while the outdoor unit is used for refrigeration cycle.

[0071] Large Language Models (LLMs) are deep learning models trained on massive amounts of text data, enabling them to generate natural language text or understand the meaning of language text. These models can provide in-depth knowledge and language production on a wide range of topics through training on large datasets. Their core idea is to learn patterns and structures of natural language through large-scale unsupervised training, mimicking human language cognition and generation processes to some extent.

[0072] The air conditioning vertical domain model is a dedicated large language model built using deep learning of air conditioning expertise, user behavior data, and environmental parameters, specifically for the needs of the air conditioning industry, particularly smart air conditioners, in terms of voice interaction, scene control, and energy efficiency optimization.

[0073] In one exemplary embodiment, such as Figure 1 As shown, an air conditioner is provided, the air conditioner comprising:

[0074] Indoor unit 110 is used to regulate indoor air;

[0075] Outdoor unit 120 is connected to indoor unit 110;

[0076] Controller 130 is configured as follows:

[0077] In response to voice control commands, convert the voice control commands into control command text;

[0078] Obtain the current environmental information and the air conditioner status code. Based on the control command text, the current environmental information, and the air conditioner status code, obtain the air conditioner control code and the target interaction text.

[0079] The operating status of the air conditioner is controlled according to the air conditioner control code, and the external speaker module of the indoor unit 110 is controlled to broadcast the target interactive text.

[0080] The outdoor unit 120 and the indoor unit 110 are connected in series via a refrigeration system communication bus, and the controller 130 and the indoor unit 110 are connected in series via a control system communication bus.

[0081] The current environmental information refers to environmental information that affects air conditioning, including but not limited to current weather, outdoor temperature, outdoor humidity, season, and current time.

[0082] The air conditioner status code is obtained by encoding the air conditioner status according to a preset coding format. The air conditioner status includes, but is not limited to, on / off status, operating mode, air conditioner set temperature, and air conditioner set humidity.

[0083] The air conditioning control code is obtained by encoding the air conditioning control commands according to a preset coding format. The air conditioning control commands include, but are not limited to, power on / off commands, operating mode switching commands, temperature adjustment commands, and humidity adjustment commands.

[0084] The external speaker module includes, but is not limited to, a display and a speaker.

[0085] The target interactive text is a text-based user response that describes the air conditioner's adjusted operating status. For example, the target interactive text could be: "Okay, you feel hot. I have turned on the air conditioner in cooling mode for you. The temperature is 26 degrees Celsius and the fan speed is medium. I believe it will bring the temperature down soon."

[0086] In this embodiment, the controller detects the user's voice control command and converts it into control command text using Automatic Speech Recognition (ASR) technology. It acquires current environmental information detected by sensors or input by the user. These sensors are connected to the controller via wired or wireless means. The controller obtains its own hardware status information and encodes it using a preset encoding method to obtain an air conditioner status code. It then calls the built-in air conditioner vertical domain model, inputting the control command text, current environmental information, and air conditioner status code to obtain the air conditioner control code and target interaction text. The air conditioner control code is then sent to the air conditioner's hardware to control its operating status, and the target interaction text is sent to the indoor unit's speaker module for playback.

[0087] The aforementioned air conditioner, by acquiring its status code, directly represents the hardware status of the air conditioner, eliminating the need for text-based representations and avoiding information loss or misunderstanding associated with text-based representations. This improves the accuracy of acquiring the air conditioner's hardware status. The air conditioner control code can be directly sent to the air conditioner's hardware without converting control commands into hardware-readable instructions, better aligning with the underlying hardware logic and improving control execution speed. Furthermore, by broadcasting interactive text through an external speaker module, users can quickly understand the air conditioner's operating status, enhancing its intelligence level.

[0088] In one embodiment, please refer to Figure 2 A flowchart illustrating a control method for an air conditioner is provided. The controller is configured as follows:

[0089] Step S210: Input the control command text and current environment information into the pre-trained first language model to generate the first implicit feature;

[0090] Step S220: Convert the air conditioner status code into the corresponding second implicit feature, and input the first implicit feature and the second implicit feature into the pre-trained second language model to generate the air conditioner control code and the target interactive text; wherein, the air conditioner control code and the air conditioner status code adopt the same preset coding form.

[0091] Both the pre-trained first language model and the pre-trained second language model are air conditioning vertical domain models, employing a transformer architecture. The first language model has a moderate number of parameters, while the second language model has a smaller number of parameters. The first language model can be a large language model such as the Qianwen series, the Wudao series, or Lingyi Wanwu. In this embodiment, the first language model can be the qwen3-8b model, without any fine-tuning or training modifications, simply retaining its original information. The second language model can be a qwen2.5 series model.

[0092] In particular, the air conditioning vertical domain model is usually divided into multiple blocks, which are the number of layers of the model. Each block is relatively independent and contains modules such as self-attention mechanism, residual connection, and linearity to implement some functions of self-attention. A model can be configured with different numbers of blocks to adjust the amount of model parameters.

[0093] The first implicit feature is a low-frequency feature vector, which contains the user's reply and air conditioner adjustment information.

[0094] Among them, the second implicit feature is a high-frequency feature vector, which is obtained by converting the air conditioner state code through a pre-trained air conditioner state encoder, and contains the state information of the air conditioner.

[0095] The preset encoding format can be a multi-temperature encoding format. In the multi-temperature encoding format, each bit represents air conditioner status information with a different value range. For example, for the on / off status, 0 indicates off and 1 indicates on. For the fan speed, 0 indicates fan off, 1 indicates silent mode, 2 indicates low speed, 3 indicates medium speed, and 4 indicates high speed. For the set temperature, the number 16 indicates a set temperature of 16℃.

[0096] Current air conditioning vertical domain models are primarily trained based on text, including text fine-tuning and text pre-training + post-training methods. While text-based training can fully utilize the semantic information of the text, it cannot directly obtain the hardware status of the air conditioner. The hardware status usually needs to be converted into natural language before being input into the model for processing. Furthermore, the commands issued by the model are also in natural language form. Therefore, it cannot directly communicate with the air conditioner hardware and cannot achieve real-time control of the air conditioner's operating status.

[0097] In the embodiments of this application, please refer to Figure 3 This paper provides a model architecture diagram. Specifically, the pre-trained first language model LLM1 receives text input and outputs the first implicit feature feature1. The text input includes the user's control commands and current environmental information. The user's control commands can be expressed verbally and converted into control command text using ASR technology. The pre-trained second language model LLM2, as the second-stage model, receives the second implicit feature feature2 and the first implicit feature feature1 output by the first-stage model, and outputs the air conditioning control code and the target interactive text.

[0098] Please see Figure 4 This paper presents a schematic diagram of second language model training. The second-stage model is trained using a combination of supervised fine-tuning and reinforcement fine-tuning for the following reasons: Currently, the amount of sample data is relatively small. With limited sample data, reinforcement fine-tuning is slightly more effective than supervised fine-tuning, while also providing the model with better generalization and lower illusions.

[0099] The training process is as follows: Data preparation, consisting of two types: supervised fine-tuning datasets and reinforcement fine-tuning datasets. The supervised fine-tuning dataset comprises only positive examples, representing user input and expected output. A preference dataset is constructed based on these positive examples and their corresponding negative examples. Positive examples include sets of other air conditioner execution commands that meet user needs, while negative examples are broad, including incorrect user responses, incorrect sets of air conditioner execution commands, mismatches between text commands and multi-thermal encoded commands, etc. Supervised fine-tuning involves using the supervised fine-tuning dataset to fine-tune the initial model. Reinforcement fine-tuning uses the Direct Preference Optimization (DPO) method for fine-tuning; the model trained using DPO is the final model (second language model).

[0100] This application's embodiments are based on a two-stage model architecture. The pre-trained first language model serves as the first-stage model, focusing on processing control command text and current environmental information. It can capture implicit needs in user commands (for example, when a user says "it's a bit cold," the system can determine that the set temperature needs to be raised, rather than simply turning off the air conditioner), reducing misoperations caused by ambiguous commands and improving the accuracy and adaptability of voice control. The pre-trained second language model serves as the second-stage model, focusing on processing the first and second implicit features, enabling the generated air conditioner control code to adapt to the current device state of the air conditioner. Compared to directly inputting all data into a large language model, the two-stage model architecture can decompose complex tasks into two relatively simple sub-tasks. Each stage of the model focuses on a specific sub-task, improving the quality and accuracy of the model processing at each stage.

[0101] In one embodiment, see Figure 5 A schematic flowchart of a control method for an air conditioner is provided. The air conditioner also includes a communicator, through which the controller establishes a communication connection with a cloud server; wherein...

[0102] The controller is configured as follows:

[0103] Step S510: Send the control command text, current environmental information, and air conditioning status code to the cloud server;

[0104] Step S520: Receive the air conditioning control code and target interactive text returned by the cloud server;

[0105] The cloud server is configured as follows:

[0106] Step S530: Input the control command text and current environment information into the pre-trained first language model to generate the first implicit feature;

[0107] Step S540: Convert the air conditioner status code into the corresponding second implicit feature, and input the first implicit feature and the second implicit feature into the pre-trained second language model to generate the air conditioner control code and the target interactive text; the air conditioner control code and the air conditioner status code adopt the same preset coding form.

[0108] In the embodiments of this application, please refer to Figure 6 This paper provides a system architecture diagram of an air conditioner and a cloud server. The air conditioner communicates with the cloud server 150 via a communicator 140, which can be a wireless module.

[0109] Please see Figure 7 This paper provides a signal timing interaction diagram between an air conditioner and a cloud server. The cloud server 150 is equipped with a pre-trained first language model and a pre-trained second language model. The cloud server 150 receives control command text, current environmental information, and air conditioner status codes sent by the controller 130 via the communicator 140. Using a two-stage model architecture of the pre-trained first and second language models, the cloud server 150 outputs the air conditioner control code and target interaction text to the controller 130. The controller 130 distributes the air conditioner control code to the air conditioner's hardware, controls the hardware's operating status, and sends the target interaction text to the speaker module of the indoor unit 110, which then broadcasts the target interaction text.

[0110] This application embodiment utilizes the high computing power of cloud servers by deploying pre-trained first language models and pre-trained second language models on cloud servers, thereby improving the running efficiency of the models and accelerating the response speed of the air conditioner to user control commands.

[0111] In one embodiment, see Figure 8 This document provides a flowchart illustrating the generation of the first implicit feature. The process involves inputting control command text and current environment information into a pre-trained first language model to generate the first implicit feature, including:

[0112] Step S810: Combine the control command text, current environment information, and preset system prompt words to obtain the target prompt word;

[0113] Step S820: Input the target prompt word into the pre-trained first language model to generate the first implicit feature;

[0114] The system prompts are used to configure the response behavior patterns of the pre-trained first language model.

[0115] In this embodiment, system prompts are used to locate the role, describe the task, and elucidate the control commands for the air conditioner, better guiding the first language model to complete the corresponding task and generate corresponding low-frequency feature vectors. The target prompt is obtained by combining the control command text, current environmental information, and preset system prompts. The target prompt is then input into the pre-trained first language model, which outputs the first implicit feature.

[0116] This application embodiment can guide the first language model to generate highly accurate first implicit features by constructing target prompt words.

[0117] In one embodiment, see Figure 9 This document provides a flowchart illustrating the process of obtaining the second implicit feature. Converting the air conditioner state code into the corresponding second implicit feature includes:

[0118] Step S910: Normalize the air conditioner status code to obtain the normalized code;

[0119] Step S920: Adjust the original dimension of the normalized encoding to the preset dimension to obtain the target dimension encoding;

[0120] Step S930: Input the target dimension encoding into the pre-trained air conditioner state encoder to obtain the second implicit feature.

[0121] The preset dimension can be set according to actual needs. Specifically, the preset dimension is the same as the vector dimension of the first implicit feature.

[0122] The air conditioner status encoder can be an embedding model. Specifically, it can be a bge-m3 model.

[0123] In the embodiments of this application, please refer to Figure 10 , Figure 10 A schematic diagram of the air conditioner status encoder architecture is provided. Specifically, the air conditioner status code is a 16-dimensional data, such as "1, 2, 0, 2, 26, 56, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0," which needs to be normalized (preprocessed) to obtain a normalized code. Specifically, the value of each dimension is divided by the maximum value corresponding to that dimension to obtain a decimal between 0 and 1. Through a linear layer, the 16 dimensions of the normalized code are adjusted to 4096 dimensions to obtain the target dimension code. The target dimension code is input into the pre-trained air conditioner status encoder (embedding model) for encoding processing to obtain the second implicit feature, feature2.

[0124] The pre-trained air conditioner state encoder is based on a transformer architecture, and its training is achieved through fine-tuning. Specifically, training data is prepared as follows: Since the air conditioner state encoding is a multi-hot encoding, each bit represents a different state, and the total number of states is fixed. Therefore, positive and negative samples in the training data are relatively easy to obtain. A negative sample is created by changing one key bit in the sample data, while a positive sample is created by changing a non-key bit. After obtaining the comparison data of positive and negative samples, a contrastive learning method can be used to fine-tune the air conditioner state encoder. After fine-tuning, the data is validated to determine whether the output of the air conditioner state encoder meets expectations. If it does not meet expectations, fine-tuning continues; if it does meet expectations, the fine-tuning is complete.

[0125] This application embodiment performs normalization, dimensionality adjustment, and encoding processing on the air conditioner status code, converting it into a high-dimensional feature vector for easier subsequent processing by a second language model. In one embodiment, please refer to... Figure 11 This paper provides a flowchart illustrating the generation of air conditioning control codes and target interactive text. The process involves inputting a first implicit feature and a second implicit feature into a pre-trained second language model to generate air conditioning control codes and target interactive text, including:

[0126] Step S1110: Fuse the first implicit feature and the second implicit feature to obtain the fused implicit feature;

[0127] Step S1120: Input the fused implicit features into the pre-trained second language model to generate air conditioning control codes and target interactive text.

[0128] In this embodiment, the first and second implicit features have the same vector dimension. The first and second implicit features are summed to obtain the fused implicit features. The fused implicit features are then input into a pre-trained second language model, which outputs air conditioning control codes and target interactive text.

[0129] The air conditioning control code uses a multi-heat code format, with each bit representing the following: 1. On / off status: 0-Off, 1-On; 2. Fan speed setting: 0-Fan off, 1-Silent, 2-Low, 3-Medium, 4-High; 3. Sleep mode setting: 0-Sleep off, 1-General, 2-Elderly, 3-Youth, 4-Child; 4. Operating mode: 0-Air supply, 1-Heating, 2-Cooling, 3-Dehumidification, 4-Auto; 5. Set temperature: 16-32°C; 6. Set humidity: %; 7. Humidification setting: 0-Off, 1-On; 8. Timer on / off: 0-Off, 1-On; 9. Cleaning setting: 0-Off, 1-On; 10. Up / down fan setting: 0-Off, 1-On; 11. Left / right fan setting: 0-Off, 1-On; 12. Energy saving setting. 0 - Off, 1 - On; 13. Set Indoor Cleaning: 0 - Off, 1 - On; 14. Set Outdoor Cleaning: 0 - Off, 1 - On; 15. Set Brightness: 0 - Off, 1 - Level 1 brightness, 2 - Level 2 brightness, 3 - Level 3 brightness, 4 - Level 4 brightness; 16. Set Direct Airflow: 0 - Off, 1 - On; 17. Set Powerful (Rapid Cooling / Heating): 0 - Off, 1 - On.

[0130] For example, the air conditioner control code "1, 2, 0, 2, 26, 56, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0" corresponds to "Power on, Low fan speed, Sleep mode off, Cooling, Set temperature 26℃, Set humidity 56%, Humidifier off, Timer off, Set cleaning off, Up and down fan off, Left and right fan off, Energy saving off, Indoor cleaning off, Outdoor cleaning off, Set brightness level 1, Direct airflow off, Power off".

[0131] This application embodiment integrates the first implicit feature and the second implicit feature, enabling the second language model to fully understand the information carried by the low-frequency and high-frequency features, thereby improving the accuracy of generating air conditioning control codes and target interactive text.

[0132] Experiments revealed that this application deployed the trained air conditioner state encoder and the second-stage large language model (second language model) on a graphics card. The air conditioner state encoder has 1.2B model parameters, and the second language model has approximately 3B model parameters, occupying a total of about 8G of existing storage.

[0133] We selected 100 test data points to compare the speed and accuracy of air conditioning control code distribution. Using only the qwen3-8b model for target interaction text generation and air conditioning control code distribution, the time taken was approximately 3.5 seconds, with an accuracy of approximately 84%. Using our proposed method for target interaction text generation and air conditioning control code distribution, the time taken was approximately 2.8 seconds, with an accuracy of 92%. This represents a 20% improvement in time and an 8% improvement in accuracy. The data shows that, under the same conditions, our proposed method improves both the speed and accuracy of air conditioning control code distribution.

[0134] In one embodiment, the preset encoding format includes a multi-hot encoding format.

[0135] This application embodiment uses a multi-thermal coding method to represent air conditioner status codes, which clearly defines different air conditioner status categories, avoids category confusion caused by numerical coding, and is easier for models to recognize. Using multi-thermal coding to represent air conditioner control codes allows them to be directly recognized by the air conditioning equipment, improving the control efficiency of the air conditioner.

[0136] In one embodiment, the target interactive text includes at least execution instruction text; the instruction semantics of the execution instruction text matches the instruction semantics of the air conditioning control code.

[0137] The execution instruction text includes, but is not limited to, instruction text for changing the air conditioner's operating mode, instruction text for adjusting the temperature, and instruction text for adjusting the humidity.

[0138] In this embodiment of the application, a bit correspondence of the air conditioning control code can be added to the system prompt words. In the bit correspondence, the meaning of each bit and its value range correspond to the corresponding execution instruction text.

[0139] The semantics of the execution instruction text in this application embodiment match the semantics of the air conditioner control code, allowing users to accurately understand the current operating status of the air conditioner.

[0140] like Figure 12 As shown, this application also provides a control method for an air conditioner, applied to an air conditioner including an indoor unit, an outdoor unit, and a controller, the method including:

[0141] Step S1210: In response to the voice control command, the voice control command is converted into control command text;

[0142] Step S1220: Obtain the current environmental information and the air conditioner status code; based on the control command text, the current environmental information and the air conditioner status code, obtain the air conditioner control code and the target interaction text.

[0143] Step S1230: Control the operating status of the air conditioner according to the air conditioner control code, and control the outdoor speaker module of the indoor unit to broadcast the target interactive text.

[0144] It should be noted that the specific limitations of the above steps can be found in the specific limitations of an air conditioner mentioned above, and will not be repeated here.

[0145] In one embodiment, the step of obtaining the air conditioning control code and the target interaction text based on the control command text, current environmental information, and air conditioning status code includes:

[0146] The control command text and current environment information are input into the pre-trained first language model to generate the first implicit features;

[0147] The air conditioner status code is converted into the corresponding second implicit feature. The first and second implicit features are then input into a pre-trained second language model to generate the air conditioner control code and the target interactive text.

[0148] In one embodiment, the preset encoding format includes a multi-hot encoding format.

[0149] In one embodiment, the target interactive text includes at least execution instruction text; the instruction semantics of the execution instruction text matches the instruction semantics of the air conditioning control code.

[0150] In one embodiment, control command text and current environment information are input into a pre-trained first language model to generate first implicit features, including:

[0151] Combine the control command text, current environment information, and preset system prompts to obtain the target prompt;

[0152] The target prompt word is input into the pre-trained first language model to generate the first implicit feature; wherein, the system prompt word is used to configure the response behavior pattern of the pre-trained first language model.

[0153] In one embodiment, converting the air conditioner status code into a corresponding second implicit feature includes:

[0154] The air conditioner status code is normalized to obtain the normalized code.

[0155] Adjust the original dimension of the normalized encoding to the preset dimension to obtain the target dimension encoding;

[0156] The target dimension encoding is input into a pre-trained air conditioner state encoder to obtain the second implicit feature.

[0157] In one embodiment, the first implicit feature and the second implicit feature are input into a pre-trained second language model to generate air conditioning control codes and target interactive text, including:

[0158] The first implicit feature and the second implicit feature are fused to obtain the fused implicit feature;

[0159] The fused implicit features are input into a pre-trained second language model to generate air conditioning control codes and target interactive text.

[0160] The technical solution of this embodiment obtains the air conditioner's status code. Since the air conditioner status code directly represents the status of the air conditioner's hardware, it eliminates the need for using natural language text to represent the hardware status, avoiding information loss or misunderstanding caused by text and improving the accuracy of obtaining the air conditioner's hardware status. The air conditioner control code can be directly sent to the air conditioner's hardware without converting the air conditioner control commands into commands that the air conditioner's hardware can recognize, which is more in line with the underlying hardware logic and improves the execution speed of air conditioner control. By broadcasting the target interactive text through the external speaker module, users can quickly understand the operating status of the air conditioner, improving the air conditioner's intelligence level.

[0161] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0162] Based on the same inventive concept, this application also provides a control device for an air conditioner that implements the control method for the air conditioner described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more air conditioner control device embodiments provided below can be found in the limitations of the air conditioner control method described above, and will not be repeated here.

[0163] In one exemplary embodiment, such as Figure 13 As shown, a control device for an air conditioner is provided, comprising:

[0164] The instruction conversion module 1310 is used to convert voice control instructions into control instruction text in response to voice control instructions.

[0165] The first implicit feature generation module 1320 is used to obtain current environment information, input the control command text and current environment information into the pre-trained first language model, and generate the first implicit feature;

[0166] The air conditioning control code generation module 1330 is used to obtain the air conditioning status code of the air conditioner, convert the air conditioning status code into the corresponding second implicit feature, input the first implicit feature and the second implicit feature into the pre-trained second language model to generate the air conditioning control code and the target interactive text; the air conditioning control code and the air conditioning status code adopt the same preset coding form;

[0167] The operation status control module 1340 is used to control the operation status of the air conditioner according to the air conditioner control code and to broadcast the target interactive text.

[0168] In one embodiment, please refer to Figure 14 A structural block diagram of a first implicit feature generation module is provided. The first implicit feature generation module 1320 includes:

[0169] The target prompt word acquisition unit 1410 is used to combine the control instruction text, current environment information and preset system prompt words to obtain the target prompt word;

[0170] The first implicit feature generation unit 1420 is used to input the target prompt word into the pre-trained first language model to generate the first implicit feature;

[0171] The system prompts are used to configure the response behavior patterns of the pre-trained first language model.

[0172] In one embodiment, please refer to Figure 15 A structural block diagram of an air conditioning control code generation module is provided. The air conditioning control code generation module 1330 includes:

[0173] The normalization processing unit 1510 is used to normalize the air conditioner status code to obtain the normalized code;

[0174] The dimension adjustment unit 1520 is used to adjust the original dimension of the normalized encoding to the preset dimension to obtain the target dimension encoding;

[0175] The second implicit feature acquisition unit 1530 is used to input the target dimension encoding into the pre-trained air conditioner state encoder to obtain the second implicit feature.

[0176] In one embodiment, please refer to Figure 16A structural block diagram of an air conditioning control code generation module is provided. The air conditioning control code generation module 1330 includes:

[0177] The feature fusion unit 1610 is used to fuse the first implicit feature and the second implicit feature to obtain the fused implicit feature;

[0178] The air conditioning control coding generation unit 1620 is used to input the fused implicit features into a pre-trained second language model to generate air conditioning control codes and target interactive text.

[0179] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 17 As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a method for controlling an air conditioner. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0180] Those skilled in the art will understand that Figure 17The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, which, when executed by the processor, causes the processor to perform the steps of the aforementioned air conditioner control method. The steps of the air conditioner control method described here can be steps from the air conditioner control methods of the various embodiments described above.

[0181] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, causes the processor to perform the steps of the air conditioner control method described above. The steps of the air conditioner control method described here can be the steps in the air conditioner control method of the various embodiments described above.

[0182] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, causes the processor to perform the steps of the air conditioner control method described above. The steps of the air conditioner control method described here can be the steps in the air conditioner control method of the various embodiments described above.

[0183] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0184] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic resistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0185] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0186] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. An air conditioner, characterized in that, include: Indoor unit, used to regulate indoor air; The outdoor unit is connected to the indoor unit; The controller is configured as follows: In response to a voice control command, the voice control command is converted into control command text; Obtain the current environmental information and the air conditioner status code of the air conditioner; and obtain the air conditioner control code and target interaction text based on the control command text, the current environmental information, and the air conditioner status code. The operating status of the air conditioner is controlled according to the air conditioner control code, and the speaker module of the indoor unit is controlled to broadcast the target interactive text.

2. The air conditioner according to claim 1, characterized in that, The controller is configured as follows: The control command text and the current environment information are input into a pre-trained first language model to generate a first implicit feature; The air conditioner status code is converted into a corresponding second implicit feature. The first implicit feature and the second implicit feature are input into a pre-trained second language model to generate the air conditioner control code and the target interactive text. The air conditioning control code and the air conditioning status code use the same preset coding format.

3. The air conditioner according to claim 1, characterized in that, The air conditioner also includes a communicator, through which the controller establishes a communication connection with the cloud server; wherein... The controller is configured as follows: Send the control command text, the current environment information, and the air conditioner status code to the cloud server; Receive the air conditioning control code and target interactive text returned by the cloud server; The cloud server is configured as follows: The control command text and the current environment information are input into a pre-trained first language model to generate a first implicit feature; The air conditioner status code is converted into a corresponding second implicit feature. The first implicit feature and the second implicit feature are input into a pre-trained second language model to generate the air conditioner control code and the target interactive text. The air conditioning control code and the air conditioning status code use the same preset coding format.

4. The air conditioner according to any one of claims 2 to 3, characterized in that, The preset encoding format includes multi-hot encoding.

5. The air conditioner according to any one of claims 1 to 3, characterized in that, The target interactive text includes at least execution instruction text; the instruction semantics of the execution instruction text matches the instruction semantics of the air conditioning control code.

6. The air conditioner according to any one of claims 2 to 3, characterized in that, The step of converting the air conditioner status code into a corresponding second implicit feature includes: The air conditioner status code is normalized to obtain the normalized code; Adjust the original dimension of the normalized encoding to a preset dimension to obtain the target dimension encoding; The target dimension encoding is input into a pre-trained air conditioner state encoder to obtain the second implicit feature.

7. The air conditioner according to any one of claims 2 to 3, characterized in that, The step of inputting the first implicit feature and the second implicit feature into a pre-trained second language model to generate the air conditioning control code and the target interactive text includes: The first implicit feature and the second implicit feature are fused to obtain the fused implicit feature; The fused implicit features are input into the pre-trained second language model to generate the air conditioning control code and the target interactive text.

8. The air conditioner according to any one of claims 2 to 3, characterized in that, The step of inputting the control command text and the current environment information into a pre-trained first language model to generate a first implicit feature includes: The target prompt word is obtained by combining the control command text, the current environment information, and the preset system prompt word. The target prompt word is input into the pre-trained first language model to generate the first implicit feature; The system prompts are used to configure the response behavior pattern of the pre-trained first language model.

9. A control method for an air conditioner, characterized in that, Applied to an air conditioner, the air conditioner including an indoor unit, an outdoor unit, and a controller, the method includes: In response to a voice control command, the voice control command is converted into control command text; Obtain the current environmental information and the air conditioner status code of the air conditioner; and obtain the air conditioner control code and target interaction text based on the control command text, the current environmental information, and the air conditioner status code. The operating status of the air conditioner is controlled according to the air conditioner control code, and the speaker module of the indoor unit is controlled to broadcast the target interactive text.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method of claim 9.

Images