Equipment operation control method and device and storage medium

Abstract

The invention provides an equipment operation control method and device and a storage medium, and relates to the technical field of equipment intelligent control. The method comprises the steps of obtaining first instruction information input by a user; under the condition that the first instruction information belongs to the scene type instruction information, judging whether a scene matched with the first instruction information exists in a scene library or not; under the condition that a scene matched with the first instruction information does not exist in a scene library, constructing a first cue word according to a first cue word template corresponding to the scene type instruction information, the first instruction information and multi-source perception information; according to the first prompt word, generating a target scene and a first control instruction corresponding to the target scene by using a model; and controlling the target equipment to operate according to the first control instruction. Through the method, the scene construction function can be supported while the response real-time performance and the operation control accuracy of the equipment operation control are considered, the intelligence and the efficiency of the equipment operation control are improved, and the energy consumption of the equipment is reduced.

Description

Technical Field

[0001] This disclosure relates to the field of intelligent equipment control technology, and in particular to an equipment operation control method, apparatus and storage medium. Background Technology

[0002] Currently, the operation of common home appliances such as air conditioners mainly relies on manual control by users. For example, users adjust the temperature or humidity using the air conditioner remote control. In addition, some air conditioner remote controls are also equipped with function keys for preset modes (or preset scenes), such as "sleep mode" and "dehumidification mode." Users can achieve one-click control of the air conditioner by clicking the function key of the preset mode.

[0003] However, since manual adjustment relies heavily on the user's subjective experience, it can lead to unsatisfactory results, requiring repeated adjustments and impacting the user experience. Furthermore, preset modes such as sleep mode are often factory-set and may not meet individual user needs.

[0004] In recent years, with the development of artificial intelligence technology, some home appliances such as air conditioners have attempted to incorporate large-scale model technology. For example, related technologies propose the following equipment operation control scheme: collecting multimodal data such as human voice and image perception data and meteorological data, and inputting the multimodal data into a model (such as a generative large model) to obtain the equipment control strategy.

[0005] While the related equipment operation control solutions can improve the user experience to some extent, they still have the following problems: First, since the model needs to be called to generate control commands every time, the equipment may consume more energy and have poor real-time response. Second, they do not support automatic scene construction, making it difficult to meet the equipment operation control requirements based on scene libraries. Third, they adopt a relatively general large model architecture, and the model training and prediction process lacks vertical domain adaptation and adjustment, which makes it difficult for the accuracy of the control strategy output by the large model to meet the requirements, and the interpretability of the output control strategy is poor. Summary of the Invention

[0006] In order to solve at least one of the problems of high energy consumption and poor real-time response in the equipment operation control methods of related technologies, this disclosure provides an equipment operation control method, apparatus and storage medium.

[0007] According to a first aspect of this disclosure, a device operation control method is provided, comprising:

[0008] The system acquires first instruction information input by the user; if the first instruction information is a scene-type instruction, it determines whether a scene matching the first instruction information exists in the scene library; if no scene matching the first instruction information exists in the scene library, it constructs a first prompt word based on a first prompt word template corresponding to the scene-type instruction information, the first instruction information, and multi-source perception information, wherein the multi-source perception information includes environmental perception information and target device operating parameter information; based on the first prompt word, it uses a model to generate a target scene and a first control instruction corresponding to the target scene; and based on the first control instruction, it controls the target device to operate.

[0009] In some embodiments, the device operation control method further includes: if a scene matching the first instruction information exists in the scene library, controlling the target device to operate according to a second control instruction corresponding to the matching scene in the scene library.

[0010] In some embodiments, determining whether a scene in the scene library matches the first instruction information includes: calculating the similarity between the first instruction information and each scene in the scene library; determining the scene with the highest similarity from the scene library; if the similarity of the scene with the highest similarity is greater than a similarity threshold, taking the scene with the highest similarity as the scene matching the first instruction information; if the similarity of the scene with the highest similarity is less than or equal to the similarity threshold, determining that no scene matches the first instruction information.

[0011] In some embodiments, the device operation control method further includes: updating the scene library according to the target scene and the first control instruction; and / or,

[0012] The system acquires second instruction information input by the user, wherein the second instruction information is adjustment indication information in response to the first control instruction; constructs a second prompt word based on a second prompt word template corresponding to the adjustment indication information and the second instruction information, wherein the second prompt word template is different from the first prompt word template; generates an adjustment instruction for the first control instruction using the model based on the second prompt word; controls the operation of the target device according to the adjustment instruction; and modifies the first control instruction corresponding to the target scene in the scene library according to the adjustment instruction.

[0013] In some embodiments, the scene library also stores the number of corrections for each scene and the similarity threshold corresponding to each scene. The method further includes: updating the number of corrections for the target scene in the scene library when receiving the second instruction information from the user; and adjusting the similarity threshold corresponding to the target scene when the number of corrections for the target scene is greater than the number threshold.

[0014] In some embodiments, the device operation control method further includes: when the first instruction information is non-scene-type instruction information, constructing a third prompt word based on the third prompt word template corresponding to the non-scene-type instruction information, the first instruction information, and the multi-source perception information, wherein the third prompt word template is different from the first prompt word template; generating a third control instruction using the model based on the third prompt word; and controlling the operation of the target device based on the third control instruction.

[0015] In some embodiments, the target device is an air conditioner, and the model is a network model based on a Transformer, including a decoder and an instruction prediction layer, an energy consumption prediction layer, and a comfort prediction layer connected to the decoder. The method further includes: while outputting the first control instruction using the model, also outputting the energy consumption prediction value of the target device within a set time after executing the first control instruction, and the comfort prediction value of the user after executing the first control instruction.

[0016] In some embodiments, the decoder includes at least one attention layer based on a temporal attention mechanism; and / or, the model is trained according to a joint loss function, the joint loss function including: a first function term representing the loss calculated based on the difference between the instruction generated by the model and a pre-labeled instruction; a second function term representing the loss calculated based on the difference between the energy consumption prediction value obtained by the model and a pre-labeled minimum energy consumption; a third function term representing the reward given when the set temperature indicated by the instruction generated by the model is within the human comfort range; and a fourth function term representing the penalty given when the instruction generated by the model violates the domain physical rules or safe operating procedures of the target device.

[0017] In some embodiments, the environmental sensing information includes at least one of the following: whether the current date is a holiday, weather, outdoor temperature, indoor temperature, outdoor humidity, and indoor humidity; and / or, the target device's operating parameter information includes at least one of the following: current operating mode, current set temperature, and current set windshield; and / or, the multi-source sensing information further includes at least one of the following: user's historical preferences, user's health record, the geographical location of the target device, current season, current time period, and electricity price information corresponding to the current time period.

[0018] According to a second aspect of this disclosure, an equipment operation control apparatus is provided, comprising: a module for performing the equipment operation control method as described above.

[0019] According to a third aspect of this disclosure, a device operation control apparatus is provided, comprising: a memory; and a processor coupled to the memory, the processor being configured to execute the device operation control method as described above based on instructions stored in the memory.

[0020] According to a fourth aspect of this disclosure, an air conditioner is provided, comprising: an equipment operation control device as described above.

[0021] According to a fifth aspect of this disclosure, a computer-readable storage medium is provided having computer program instructions stored thereon, which, when executed by a processor, implement the device operation control method as described above.

[0022] According to a sixth aspect of this disclosure, a computer program product is provided having computer program instructions stored thereon, which, when executed by a processor, implement the device operation control method as described above.

[0023] Other features and advantages of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0024] The accompanying drawings, which form part of this specification, illustrate embodiments of this disclosure and, together with the specification, serve to explain the principles of this disclosure.

[0025] Figure 1 This is a schematic flowchart of a device operation control method according to some embodiments of the present disclosure;

[0026] Figure 2 This is a schematic flowchart of a device operation control method according to other embodiments of the present disclosure;

[0027] Figure 3 This is a schematic diagram of the network architecture of a large model according to some embodiments of this disclosure;

[0028] Figure 4 This is a schematic diagram of the structure of a device operation control apparatus according to some embodiments of the present disclosure;

[0029] Figure 5 This is a schematic diagram of the structure of a device operation control apparatus according to other embodiments of the present disclosure;

[0030] Figure 6 This is a schematic diagram of the structure of an air conditioner according to some embodiments of the present disclosure.

[0031] This disclosure can be more clearly understood with reference to the accompanying drawings and the following detailed description. Detailed Implementation

[0032] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present disclosure.

[0033] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0034] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit this disclosure or its application or use.

[0035] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0036] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0037] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0038] It should be noted that the collection, gathering, updating, analysis, processing, use, transmission, and storage of user personal information involved in this disclosed technical solution all comply with relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. Necessary measures are taken to prevent unauthorized access to user personal information data and to safeguard user personal information security, network security, and national security.

[0039] To make the objectives, technical solutions, and advantages of this disclosure clearer, the following detailed description is provided in conjunction with specific embodiments and accompanying drawings. To address at least one of the problems of high energy consumption and poor real-time response in related equipment operation control methods, this disclosure provides an equipment operation control method, apparatus, and storage medium.

[0040] Figure 1 This is a schematic flowchart illustrating a device operation control method according to some embodiments of the present disclosure. Figure 1 As shown, the equipment operation control method can be executed by the equipment operation control device, including steps 11 to 16.

[0041] Step 11: Obtain the first instruction information input by the user.

[0042] In some examples, during user interaction with the target device, the system acquires the user's initial command information input through various methods such as text input, gestures, or voice input. The target device can be an air conditioner, fan, or other smart device.

[0043] In practice, users can interact directly with the target device or through other devices. For example, assuming the target device is an air conditioner, users can interact with it through a mobile client, such as by inputting voice commands like "I'm getting ready to sleep" or "I'm getting ready to sleep, please set the temperature to 26℃." Alternatively, users can interact with the air conditioner through its display screen, such as by inputting text commands like "I'm a little cold, please adjust the temperature" or "Please switch to sleep mode." Furthermore, users can interact with the target device in other ways as well.

[0044] Step 12: If the first instruction information is a scene-type instruction information, determine whether there is a scene in the scene library that matches the first instruction information.

[0045] The scenario library is a module for storing scenario (data), which can be a database, document system, or other form of storage module. The scenario library can store data for multiple scenarios, and each scenario entry can include one or more of the following fields: scenario identifier, scenario description, triggering conditions, running strategy (or control instructions), similarity threshold, user feedback history, last update time, and associated user tags.

[0046] For example, each scene entry in the scene library includes fields as shown in Table 1.

[0047] Table 1

[0048] In some embodiments, in step 12, it can be determined whether a matching scene exists in the scene library according to the exemplary methods shown in steps B1 to B4.

[0049] Step B1: Calculate the similarity between the first instruction information and each scene in the scene library.

[0050] In step B1, the similarity between the instruction and the scene can be calculated in various ways. The following is an example of how to calculate the similarity.

[0051] In some examples, the user's initial instruction information can be preprocessed, such as by word segmentation and keyword extraction. Then, based on the preprocessed results of the initial instruction information, a semantic vector corresponding to the initial instruction information is constructed.

[0052] For example, natural language processing tools (such as Jieba segmentation tools) can be used for word segmentation and extraction of core keywords. Then, based on a pre-trained model, such as the Bidirectional Encoder Representations from Transformers (BERT) model, the preprocessed result of the first instruction information can be converted into a semantic vector corresponding to the first instruction information. Next, the similarity between the semantic vector corresponding to the first instruction information and the semantic vector of each scene in the scene library is calculated. The semantic vector of each scene in the scene library can be obtained in advance through vectorization, or it can be obtained in real time during B1 execution.

[0053] In other examples, keywords can be extracted first from the user's initial command information. Then, based on the keywords in the initial command information and the keywords of each scene in the scene library, the similarity between the command and the scene can be calculated.

[0054] For example, the similarity between an instruction and a scene can be determined by the ratio of the number of identical keywords in the first instruction information to the total number of keywords in that scene. Assuming the keywords in the first instruction information are the same as those in scene 1, the similarity between the first instruction information and scene 1 calculated using the above method is 1. Assuming the ratio of identical keywords in the first instruction information to the total number of keywords in scene 2 is 0.8, the similarity between the first instruction information and scene 2 calculated using the above method is 0.8.

[0055] In some further examples, the user's initial instruction information, along with other required information, can be treated as a whole to calculate the similarity between this whole and scenes in a scene library. For example, the other required information might be environmental awareness information, which could include at least one of the following: whether the current date is a holiday, the weather, outdoor temperature, indoor temperature, outdoor humidity, and indoor humidity. For instance, after receiving the user's initial instruction "I want to sleep," the initial instruction "I want to sleep," along with the perceived "current date is a holiday" and "rainy day," can be treated as a whole (e.g., denoted as whole A), and its similarity can be calculated against the whole described in the scene library as "sleep scene," "holiday," and "sunny day" (e.g., denoted as whole B).

[0056] Step B2: Identify the scene with the highest similarity from the scene library.

[0057] Step B3: If the similarity of the scene with the highest similarity is greater than the similarity threshold, then the scene with the highest similarity is selected as the scene to be matched with the first instruction information.

[0058] Step B4: If the similarity of the scene with the highest similarity is less than or equal to the similarity threshold, it is determined that there is no scene matching the first instruction information. The similarity threshold can be a fixed value or a dynamically changing value.

[0059] In some examples, the similarity threshold is dynamically adjusted as follows: the number of times a scene has been corrected in the scene library is counted; if the number of corrections for a scene exceeds the correction threshold, the similarity threshold for that scene is adjusted. If the number of corrections for a scene is less than or equal to the correction threshold, the similarity threshold for that scene remains unchanged. The number of corrections for a scene can be determined based on user feedback history regarding that scene.

[0060] For example, the initial value of the similarity threshold can be set to 0.7, and the number of corrections can be set to 3. When a user makes more than 3 corrections for a certain scenario, the similarity threshold for that scenario is adjusted to 0.75; when a user makes 3 or fewer corrections for a certain scenario, the similarity threshold for that scenario remains unchanged.

[0061] In this embodiment, by calculating the similarity between the user's first instruction information and scenes in the scene library, and by determining the scene with the highest similarity and combining it with a similarity threshold, a comprehensive judgment is made as to whether a scene matches the first instruction information. This improves the accuracy and reliability of the judgment on whether a scene matches an instruction. Furthermore, by dynamically adjusting the similarity threshold corresponding to a scene based on the number of times the scene is corrected, compared to using a fixed similarity threshold, it helps to further improve the accuracy of judging whether an instruction matches a scene.

[0062] In some embodiments, the device operation control method further includes: before step S12, determining whether the first instruction information belongs to scene-type instruction information. Scene-type instruction information and non-scene-type instruction information are two types of instructions. For example, the instruction "I want to sleep, please adjust the air conditioner" issued by the user can be a scene-type instruction, while the instruction "I am a little hot" issued by the user can be a non-scene-type instruction.

[0063] In practice, various methods can be used to determine whether the first instruction information belongs to a scenario-based instruction information. The following two examples illustrate this.

[0064] In some examples, steps A1 to A4 determine whether the first instruction information belongs to scene-based instruction information. Step A1: Pre-set the rules for determining scene-based instruction information. Step A2: After obtaining the first instruction information, determine whether the first instruction information conforms to the rules for determining scene-based instruction information. Step A3: If the result of step A2 is yes, determine that the first instruction information belongs to scene-based instruction information. Step A4: If the result of step A2 is no, determine that the first instruction information belongs to non-scene-based instruction information. For example, the rules for determining scene-based instruction information may include pre-set scene keywords. If the first instruction information contains pre-set scene keywords, determine that the first instruction information belongs to scene-based instruction information. If the first instruction information does not contain pre-set scene keywords, determine that the first instruction information belongs to non-scene-based instruction information.

[0065] In other examples, it is determined whether the first instruction information belongs to the scene category instruction information as follows: A scene classification model is used to determine whether the first instruction information is scene category instruction information. Here, the scene classification model is a trained machine learning model that can be used for binary classification.

[0066] Step 13: If there is no scene in the scene library that matches the first instruction information, construct the first prompt word based on the first prompt word template corresponding to the scene-type instruction information, the first instruction information, and the multi-source perception information.

[0067] Multi-source sensing information can be acquired before executing step 13. This multi-source sensing information includes environmental sensing information and the target device's operating parameter information. Environmental sensing information can be environmental sensing data at the current moment or a sequence of environmental sensing data over a period of time. Environmental sensing information may include at least one of the following: whether the current date is a holiday, weather, outdoor temperature, indoor temperature, outdoor humidity, and indoor humidity. Weather may include data on sunny, cloudy, and rainy days. Furthermore, environmental sensing information may also include at least one of the following: detection data for volatile organic compounds, PM2.5, CO2, formaldehyde, and light intensity. The target device's operating parameter information may include at least one of the following: current operating mode, current set temperature, and current set fan speed. Additionally, the target device's operating parameter information may also include at least one of the following: voltage, current, compressor frequency, operating time, and pipeline temperature data.

[0068] For example, when the target device is an air conditioner, the multi-source sensing information may include whether the current date is a holiday, the weather, the current outdoor temperature, the current room temperature, the current indoor relative humidity, the current room temperature setting, and the current indoor unit fan speed setting. In some examples, the multi-source sensing information includes, in addition to environmental sensing information and the target device's operating parameters, at least one of the following: the user's historical preferences, the user's health record, the target device's geographical location, the current season, the current time of day, and the electricity price information corresponding to the current time of day.

[0069] For example, when the target device is an air conditioner, the multi-source sensing information includes: 1. Environmental sensing information, specifically the current indoor temperature, current outdoor temperature, current indoor humidity, current outdoor humidity, and PM2.5; 2. Air conditioner operating parameter information, specifically the current operating mode, set fan speed, and set temperature; 3. Spatiotemporal context information, specifically the geographical location of the air conditioner, geographical location, season, time of day, and electricity price cycle; 4. User profile information, specifically the user's historical preferences for operating the air conditioner and the user's health record.

[0070] In this embodiment, by including environmental sensing information such as whether the current date is a holiday, indoor and outdoor temperature, and humidity in the multi-source sensing information, it is easier to comprehensively understand the environmental factors related to equipment operation and control. Furthermore, by including operating parameters such as the current operating mode and current set temperature in the multi-source sensing information, it is easier to comprehensively understand the equipment operating status related to equipment operation and control. The combination of these two aspects helps the auxiliary model to make better equipment operation and control decisions, improving the accuracy of the control commands output by the model. In addition, by including not only environmental sensing information and equipment operating parameters in the multi-source sensing information, but also information such as user historical preferences, user health records, and the current season, it is helpful to generate more personalized and accurate equipment operation and control commands for real-world and complex scenarios.

[0071] In some examples, the first prompt word template, the first instruction information, and multi-source perception information can be concatenated to obtain the first prompt word.

[0072] Step 14: Based on the first prompt word, use the model to generate the target scene and the first control command corresponding to the target scene.

[0073] Step 14 can be implemented in several ways. Two examples are described below.

[0074] In the first implementation, in step 14, the first prompt word is input into the model to obtain the target scene output by the model and the first control command corresponding to the target scene. The model can be a large model or other machine learning model. A large model typically refers to an artificial intelligence model with a large number of parameters constructed from artificial neural networks.

[0075] In the second implementation, in step 14, the first prompt word is input into the model to obtain the target scene and control command output by the model this time; the control command output by the model this time is verified according to the set verification rules; if the verification passes, the control command output by the model this time is used as the first control command; if the verification fails, the model is called again to output a new target scene and control command, until the control command passes the verification or the number of model calls reaches the set value.

[0076] The verification rules can include reasonableness verification rules. For example, reasonableness verification rules verify whether the temperature is within the 16-30°C range and whether the fan speed is the preset value. Verification rules can also include conflict verification rules. For example, conflict verification rules verify whether the control command conflicts with the device's hardware capabilities. For instance, the control command requires the air conditioner to "swing," but the air conditioner does not support this function. Verification rules can also include invalidity verification rules. For instance, invalidity verification rules verify whether the output format of the control command is incorrect.

[0077] In this embodiment of the disclosure, by verifying the generated target scene and control commands based on multiple verification rules after generating the target scene and control commands using the model, the accuracy and reliability of the generated target scene and control commands can be further improved.

[0078] Step 15: Control the target device to operate according to the first control command.

[0079] For example, assuming the target device is an air conditioner, the first control command output by the model is "adjust the set temperature to 24℃, adjust the fan speed to medium speed, adopt cooling mode, and turn off the swing". Then, the operating parameters of the air conditioner are adjusted according to the above command.

[0080] For example, assuming the target device is a fan, and the first control command output by the model is "adjust the fan speed to medium speed", then the operating parameters of the fan will be adjusted according to the above command.

[0081] In some embodiments, the device operation control method further includes: updating the scene library according to the target scene and the first control command.

[0082] For example, assuming the model outputs the following target scene information and first control command: {"Scene Description":"Living room summer cooling","Trigger Condition":["Cooling", "Avoid direct airflow"],"First Control Command":{"Set Temperature":24,"Fan Speed":"Medium","Operating Mode":"Cooling","Swing":"Off"}, a new scene entry can be added to the scene library, and the generated target scene and first control command can be stored in that scene entry.

[0083] In this embodiment of the disclosure, by updating the scene library based on the above information after obtaining the target scene and the first control command, the scene library can be continuously enriched, which facilitates more efficient control of the target device based on the scenes and corresponding control commands in the scene library.

[0084] In some embodiments, the device operation control method further includes: if a scene matching the first instruction information exists in the scene library, controlling the target device to operate according to a second control instruction corresponding to the matching scene in the scene library.

[0085] For example, assuming the first instruction message received from the user is "I want to sleep", the matching scenario is found in the scenario library in step 12. The control instruction (i.e. the second control instruction) recorded in the entry of scenario 2 is "set temperature to 27°C, set fan speed to low". The set temperature and set fan speed of the target device can be adjusted according to the control instruction.

[0086] In this embodiment of the disclosure, by Figure 1 The process described above has two main aspects. First, by determining whether a scenario matching the user's command exists in the scenario library, if no matching scenario exists, control commands for the target device are generated based on the model. If a matching scenario exists, the target device is directly controlled based on the corresponding control commands. Compared to generating control commands based on the model each time, this approach balances the real-time responsiveness and precision of device operation control, reduces energy consumption, and improves the user experience. Second, by generating the target scenario based on the model after determining that the user's command is a scenario-based command and no matching scenario exists in the scenario library, this approach not only supports scenario building but also facilitates direct control of the target device based on the scenario and its corresponding control commands when needed, further improving the efficiency of device operation control.

[0087] Figure 2 This is a schematic flowchart illustrating a device operation control method according to other embodiments of this disclosure. For example... Figure 2 As shown, the equipment operation control method can be executed by the equipment operation control device, including steps 201 to 215.

[0088] Step 201: Obtain the first instruction information input by the user.

[0089] Step 202: Determine whether the first instruction information is a scene-based instruction information or a non-scene-based instruction information.

[0090] For details on how to implement steps 201 and 202, please refer to the relevant content of the embodiments described above.

[0091] If the first instruction information is a scenario-based instruction information, proceed to step 203; if the first instruction information is a non-scenario-based instruction information, proceed to step 213.

[0092] Step 203: Determine whether a scene matching the first instruction information exists in the scene library. For details on how step 203 is implemented, please refer to the relevant content of the previously described embodiments.

[0093] If no scene matching the first instruction information exists in the scene library, proceed to steps 204 to 211; if a scene matching the first instruction information exists in the scene library, proceed to step 212.

[0094] Step 204: Construct the first prompt word based on the first prompt word template, the first instruction information, and the multi-source perception information.

[0095] In some examples, the first prompt template includes: a model role prompt, a user instruction field, a context field, and a requirement field for the target scene and control instructions to be output. The user instruction field is used to concatenate the first instruction information, and the context field is used to concatenate multi-source perception information.

[0096] For example, the first prompt word template can be designed as follows:

[0097] Model character prompt: You are an air conditioning scene generation expert. Generate a scene based on the following information:

[0098] User command: {}

[0099] Context: {}

[0100] Requirements for the target scenario and control commands to be output:

[0101] 1. Scenario Description: Summarize the core requirement in 10 words or less (e.g., 'cooling down the living room in summer').

[0102] 2. Triggering conditions: Extract 2-3 keywords (e.g., ['cooling down', 'avoid direct airflow']);

[0103] 3. Control commands: Strictly output in JSON format, including temperature (16-30°C), fan speed (low / medium / high), and mode (cooling / heating);

[0104] 4. Control commands must be reasonable (e.g., if the room temperature is 28°C in autumn, set the temperature to 24°C).

[0105] Assuming the user's initial input is "I want to sleep," and the multi-source sensing information is "Holiday, rainy day, current outdoor temperature 30℃, current indoor temperature 25℃, current indoor humidity 60%, current device set temperature 25℃, current device set wind speed high," then step 204 will yield the following first prompt:

[0106] Model character prompt: You are an air conditioning scene generation expert. Generate a scene based on the following information:

[0107] User command: {I want to sleep}

[0108] Context: {Holiday, rainy day, current outdoor temperature 30℃, current indoor temperature 25℃, current indoor humidity 60%, current device set temperature 25℃, current device set windshield is high}

[0109] Requirements for the target scenario and control commands to be output:

[0110] 1. Scenario Description: Summarize the core requirement in 10 words or less (e.g., 'cooling down the living room in summer').

[0111] 2. Triggering conditions: Extract 2-3 keywords (e.g., ['cooling down', 'avoid direct airflow']);

[0112] 3. Control commands: Strictly output in JSON format, including temperature (16-30°C), fan speed (low / medium / high), and mode (cooling / heating);

[0113] 4. Control commands must be reasonable (e.g., if the room temperature is 28°C in autumn, set the temperature to 24°C).

[0114] It should be noted that the above content regarding the first prompt word template is merely an illustrative example. In actual implementation, the first prompt word template can also adopt other design methods. For example, in addition to including the aforementioned fields, the first prompt word template can also include knowledge from the relevant equipment field. For example, it can incorporate knowledge from the air conditioning field such as "prioritize dehumidification in summer and prevent frost formation in winter" and "increase dehumidification frequency in the south and focus on rapid heating in the north."

[0115] Step 205: Based on the first prompt word, use the model to generate the target scene and the first control command corresponding to the target scene.

[0116] In some embodiments, the target device is an air conditioner. In these embodiments, the model employed can be a transformer-based network model, the architecture of which is as follows: Figure 3 As shown, the system includes a decoder 31, and a command prediction layer 32 (or "command prediction head"), an energy consumption prediction layer 33, and a comfort prediction layer 34 connected to the decoder. In these embodiments, step 205 includes: inputting a first prompt word into the model to obtain a target scene and a first control command corresponding to the target scene, an energy consumption prediction value of the target device within a set time after executing the first control command, and a user comfort prediction value after executing the first control command. By adapting the model architecture based on the air conditioning field and outputting energy consumption and comfort prediction values, the interpretability of the control commands output by the model is improved, further enhancing the user experience.

[0117] In this disclosure, considering that at least a portion of the multi-source perceptual information can be a temporal data sequence, the model architecture can be further adapted by including at least one attention layer based on a temporal attention mechanism in the decoder 31. For example, in a model employing a transformer-based architecture, a portion of the standard attention layer (e.g., a multi-head self-attention layer) in the original model can be replaced by introducing an attention layer based on a temporal attention mechanism. Alternatively, one or more attention layers based on a temporal attention mechanism can be added after the standard attention layer in the original model decoder via residual connections.

[0118] Compared to standard attention layers, attention layers based on temporal attention mechanisms introduce a time decay factor during computation. For example, the calculation formula for an attention layer based on temporal attention mechanisms is as follows:

[0119]

[0120] in, This represents the computation result of the attention layer based on the temporal attention mechanism, where Q represents the query matrix, K represents the key matrix, V represents the value matrix, and dk represents the dimension of the key vector. is a hyperparameter used to balance the importance of semantic similarity and temporal proximity, and f(t) is the decay function. This introduces a time decay factor. The decay function can be an exponential decay function, a Gaussian decay function, etc. For example, an exponential decay function can be expressed as: , It is a learnable decay coefficient. The larger the time interval Δt, the smaller the value of -α*Δt, which in turn makes the time decay factor calculated from it smaller. For example, the Gaussian decay function can be expressed as... . It is a learnable decay coefficient. By employing a Gaussian decay function, a smooth decay can be provided, overcoming the gradient explosion problem that may occur with exponential decay functions at very close time points.

[0121] In this embodiment of the disclosure, by introducing an attention layer based on a temporal attention mechanism into the model, the model is able to better understand and utilize the dependencies in the time-stamped data sequences in the input data (such as the environmental perception information in multi-source perception information, which is a temporal data sequence), so that the model can better understand the operating cycle and state changes of equipment such as air conditioners, thereby improving the accuracy of the predicted results such as the predicted operation control commands.

[0122] In this disclosure, considering that the model may include multiple prediction layers connected to the decoder, the model training process can be improved as follows: the model is trained using a joint loss function, wherein the joint loss function includes a first function term to a fourth function term. The first function term represents the loss calculated based on the difference between the instruction generated by the model and a pre-labeled instruction. The second function term represents the loss calculated based on the difference between the energy consumption prediction value obtained from the large model and the pre-labeled minimum energy consumption. The third function term represents the reward given when the set temperature indicated by the instruction generated by the model is within the human comfort range. The fourth function term represents the penalty given when the instruction generated by the model violates the domain physical rules or safe operating procedures of the target device. For example, when the target device is an air conditioner, if the model output during training contains content such as "close the safety valve" or "set temperature exceeds physical limits," a penalty can be applied through the fourth function term.

[0123] For example, the joint loss function is calculated using the following formula:

[0124]

[0125] Among them, L total The value of the loss function. For the first function term, For the second function term, L comfort For the third function term, It is the fourth function term. , , , These are the weighting coefficients.

[0126] In this embodiment of the disclosure, by training the model based on the above-mentioned joint loss function, the model can learn to find a balance between instruction generation, energy saving and comfort, thereby further improving the performance of the trained model.

[0127] In some embodiments, the target device is an air conditioner, and the device operation control method further includes: training a model based on training samples in the air conditioning field. The model training process may include the following steps: constructing training samples and training a large model based on the training samples.

[0128] In some examples, building training samples may include: collecting data related to the target device, including but not limited to: time-series data of device operation, user operation records, maintenance records, device status parameters, and environmental information; preprocessing and aligning the collected data to construct training samples containing user commands, multi-source sensing information, and labeled values ​​such as device operation control commands. Furthermore, based on domain knowledge, negative samples containing abnormal operation sequences and typical misoperation scenarios, as well as edge case samples incorporating seasonal and extreme operating conditions, can be proactively constructed to enhance the robustness and safety of the model. For example, assuming the target device is an air conditioner, historical work order texts, equipment manuals, and fault logs can be collected from the air conditioning management system, along with corresponding environmental sensing information such as temperature, humidity, and power consumption, as well as operating parameters of components such as the compressor and fan on the air conditioner. This data can then be time-aligned and feature extracted to form training samples such as "user commands - environmental sensing information - device operating parameters - labeled processing suggestions".

[0129] In some examples, training large models based on training samples involves: performing differentiated training on large models in groups according to the sensitivity of model parameters to different knowledge. This includes: fine-tuning the parameters at the bottom layer of the model with a smaller learning rate to preserve the model's general language capabilities; using a moderate learning rate for the scene adaptation parameters of the middle layer of the model (e.g., newly introduced temporal awareness layer) to enable it to quickly learn contextual patterns within the domain; and using a relatively large learning rate for the decision output parameters of the top layer of the model (e.g., parameters of multiple prediction layers) to focus on optimizing their performance on specific domain tasks. Simultaneously, a progressive learning rate adjustment strategy can be used during training, dynamically adjusting the global learning rate in the early, middle, and late stages of training to balance convergence speed and final performance.

[0130] For example, an optimizer (such as AdamW) is used to train a large model. The learning rate of the first 6 layers of the embedding layer and decoder of the large model is set to 1e. -6 Set the learning rate of the intermediate adaptation layer of the decoder to 5e. -5 Set the learning rate of the output layer of the top decoder and multiple prediction layers of the large model to 1e. -4 Meanwhile, assuming the training lasts for 5 epochs, the learning rate is kept constant for the first 3 epochs, then reduced to 1 / 10 of its original value in the 4th epoch, and then reduced to 1 / 10 of its original value again in the 5th epoch.

[0131] In this embodiment of the disclosure, by performing differentiated group training on the large model, the performance of the trained large model can be further improved, thereby enhancing the accuracy and reliability of the operation control of the target device based on the model.

[0132] Step 206: Control the target device to operate according to the first control command. For details on how steps 204 to 206 are implemented, please refer to the relevant descriptions in the previously described embodiments.

[0133] Step 207: Update the scene library based on the target scene and the first control command.

[0134] In some embodiments, step 207 includes: adding a new scene entry in the scene library and storing the generated target scene and the first control command into the scene entry.

[0135] For example, assuming the target device is an air conditioner, and the user's first input instruction is "I want to sleep," the generated target scene and first control instruction are: "Scenario Description: Sleep Scene; Trigger Conditions: Holiday, Rainy Day, Current Outdoor Temperature 30℃, Current Indoor Temperature 25℃, Current Indoor Humidity 60%, Current Device Set Temperature 25℃, Current Device Set Fan Speed ​​High; First Control Instruction: Set Temperature 27℃, Set Fan Speed ​​Low." Then, in step 206, the air conditioner's set temperature and fan speed can be adjusted according to the first control instruction. In step 207, a new air conditioner scene entry is added to the scene library, and the generated target scene and first control instruction are stored in this scene entry.

[0136] Step 208: Obtain the second instruction information input by the user. The second instruction information is an adjustment indication information in response to the first control instruction.

[0137] Step 208 can be implemented in several ways. Two examples are described below.

[0138] In the first embodiment, step 208 includes: after generating the first control command, outputting feedback prompt information to the user; receiving adjustment instructions from the user in response to the feedback prompt information, and using them as the second command information.

[0139] For example, the equipment operation control device displays a feedback message to the user: "The air conditioner is operating as instructed. Are you satisfied?", and shows the user two buttons: "Satisfied" and "Needs Adjustment". The user can input feedback information for the first control command by clicking the "Satisfied" button or the "Needs Adjustment" button. After the user clicks the "Satisfied" button, the subsequent process ends. After the user clicks the "Needs Adjustment" button, the "Needs Adjustment" information clicked by the user is directly used as the second command information. Alternatively, after the user clicks the "Needs Adjustment" button and inputs specific adjustment instructions (such as "Lower the temperature" or "Higher the temperature") via voice or text input, the specific adjustment instructions input by the user are used as the second command information.

[0140] In the second embodiment, step 208 includes: after generating the first control command, receiving the second command information actively input by the user.

[0141] For example, after the equipment operation control device controls the target equipment to operate according to the first control command, if the user actively inputs the adjustment instruction information "temperature too high" through voice or text input, then the adjustment instruction information input by the user will be used as the second instruction information.

[0142] Step 209: Construct the second prompt word based on the second prompt word template and the second instruction information.

[0143] In some embodiments, the second prompt word template differs from the first prompt word template. For example, the second prompt word template may include a user feedback field, a current scene strategy field, and a requirement field for the output adjustment instructions. The user feedback field is used to concatenate the second instruction information input by the user, the current scene strategy field is used to concatenate the previously output scene and first instruction information, and the requirement field is used to constrain the format of the output second instruction information.

[0144] For example, the second prompt template is designed as follows: "User feedback: {}; Current scenario strategy: {}; Requirement: Generate the modified strategy, only adjust the relevant parameters (such as temperature), keep other parameters unchanged, and output in JSON format." Further, assuming the user's first input instruction is "I want to sleep," and using the model's output of the target scenario for the air conditioner and the first control instruction as "Sleep scenario, set temperature 27℃, set fan speed to low," and the second instruction as "The temperature is a bit high," then step 209 yields the following second prompt: User feedback: {The temperature is a bit high}; Current scenario strategy: {Sleep scenario, set temperature 27℃, set fan speed to low}; Requirement: Generate the modified strategy, only adjust the relevant parameters (such as temperature), keep other parameters unchanged, and output in JSON format.

[0145] It should be noted that the above content regarding the second prompt word template is merely an illustrative example. In actual implementation, the second prompt word template may also employ other design methods.

[0146] Step 210: Based on the second prompt word, use the model to generate adjustment instructions for the first control command.

[0147] Step 210 can be implemented in various ways. Two implementation methods are described below as examples.

[0148] In a first implementation, in step 210, the second prompt word is input into the model to obtain an adjustment instruction output by the model in response to the first control command. The model can be a large model or other machine learning model. Furthermore, in some embodiments, in addition to outputting the adjustment instruction, the model can also output corresponding energy consumption prediction values ​​and comfort prediction values.

[0149] In the second implementation, in step 210, the second prompt word is input into the model to obtain the adjustment instruction output by the model for the first control command; the adjustment instruction output by the model is verified according to the set verification rules; if the verification passes, step 211 is executed; if the verification fails, the model is called again to output a new adjustment instruction, until the adjustment instruction passes the verification or the number of model calls reaches a set value. For details on how to set the verification rules, please refer to the relevant content of the foregoing embodiments. In this embodiment, by verifying the generated adjustment instruction based on the verification rules after generating it using the model, the accuracy and reliability of the generated adjustment instruction can be further improved.

[0150] Step 211: Control the target device to operate according to the adjustment instructions, and modify the first control instructions corresponding to the target scene in the scene library according to the adjustment instructions.

[0151] For example, suppose a scene entry with a unique identifier "SCN_20251124_456" was added to the scene library in step 207. This scene entry records the following information: scene description "sleep scene", trigger conditions "holiday, rainy day, current outdoor temperature 30℃, current indoor temperature 25℃, current indoor humidity 60%, current device set temperature 25℃, current device set fan speed high", and operating strategy "set temperature 27℃, set fan speed low". Furthermore, the adjustment instruction generated in step 210 is "adjust the air conditioner's set temperature to 22℃". In step 211, the air conditioner's set temperature is adjusted according to the adjustment instruction, and the operating strategy in the above scene entry in the scene library is modified to: set temperature 22℃, set fan speed low.

[0152] In this embodiment, on the one hand, by executing steps 208 to 211, adjustment instructions for the first control command can be generated using the model based on user feedback information, and the scene library can be corrected. This not only better meets the user's personalized device usage needs and improves the user's experience using the target device, but also enables dynamic updates of the scenes in the scene library, facilitating more efficient and convenient fulfillment of the user's personalized needs for the target device based on the updated scenes. On the other hand, by using different prompt word templates for the first and second command information issued by the user, and generating prompt words in different formats accordingly, the model can generate control commands more efficiently and accurately, improving the real-time performance and accuracy of the device operation control method, further enhancing the user's experience using the target device. In some embodiments, the scene library also stores the number of corrections for each scene and the similarity threshold corresponding to each scene. In these embodiments, the device operation control method further includes: updating the number of corrections for the target scene in the scene library when the user's second command information is received; adjusting the similarity threshold corresponding to the target scene when the number of corrections for the target scene is greater than the number of corrections threshold; and keeping the similarity threshold corresponding to the target scene unchanged when the number of corrections for the target scene is not greater than the number of corrections threshold.

[0153] For example, the initial similarity threshold can be set to 0.7, and the number of corrections threshold can be set to 3. Suppose that after controlling the air conditioner for 10 minutes according to the first control command corresponding to the sleep scenario, "Set the temperature to 27℃, set the fan speed to low," and then receiving the user's first second command, "The temperature is a bit high," the correction count for that scenario is set to 1, while keeping the similarity threshold unchanged. Suppose that after controlling the air conditioner for 20 minutes according to the first control command corresponding to the sleep scenario, "Set the temperature to 27℃, set the fan speed to low," and then receiving the user's fourth second command, "The temperature is a bit low," the correction count for that scenario is set to 4, and the similarity threshold for that scenario is adjusted to 0.75.

[0154] In this embodiment of the disclosure, by dynamically updating the number of corrections for the target scene in the scene library based on user feedback, and dynamically adjusting the similarity threshold corresponding to the scene based on the number of corrections, compared with using a fixed similarity threshold, it helps to more accurately determine whether the instruction matches the scene, thereby improving the accuracy of device operation control.

[0155] Step 212: Control the target device to run according to the second control command corresponding to the matching scene in the scene library.

[0156] If the judgment result in step 203 is that there is a scene in the scene library that matches the first instruction information, then step 212 is executed.

[0157] For example, assuming the target device is an air conditioner, the user's first instruction is "I want to sleep". The scene in the scene library that matches the first instruction is a sleep scene, and the second control instruction corresponding to this scene is "set the temperature to 27°C and set the fan speed to low". Then, the air conditioner's set temperature and set fan speed are adjusted according to the second control instruction.

[0158] Step 213: Construct the third prompt word based on the third prompt word template, the first instruction information, and the multi-source perception information.

[0159] If the judgment result in step 202 is that the first instruction information is a non-scenario instruction information, then step 213 is executed.

[0160] In some embodiments, the third prompt word template differs from the first prompt word template. For example, the third prompt word template may include: a model role prompt, a user instruction field, a context field, and a requirement field for control instructions regarding the output. The user instruction field is used to concatenate the first instruction information, and the context field is used to concatenate multi-source perception information.

[0161] For example, the third prompt word template can be designed as follows:

[0162] Model character prompt: You are an air conditioning scene generation expert. Generate a scene based on the following information:

[0163] User command: {}

[0164] Context: {}

[0165] Requirements for output control commands:

[0166] 1. Control commands: Strictly output in JSON format, including temperature (16-30°C), fan speed (low / medium / high), and mode (cooling / heating);

[0167] 2. Control commands must be reasonable (e.g., if the room temperature is 28°C in autumn, set the temperature to 24°C).

[0168] Assuming the user's first input command is "temperature is too cold", and the multi-source sensing information is "holiday, rainy day, current outdoor temperature 34℃, current indoor temperature 25℃, current indoor humidity 60%, current device set temperature 25℃, current device set wind speed high", then step 213 will yield the following third prompt:

[0169] Model character prompt: You are an air conditioning scene generation expert. Generate control commands based on the following information:

[0170] User command: {Temperature is too cold}

[0171] Context: {Holiday, rainy day, current outdoor temperature 34℃, current indoor temperature 25℃, current indoor humidity 60%, current device set temperature 25℃, current device set windshield is high}

[0172] Requirements for output control commands:

[0173] 1. Control commands: Strictly output in JSON format, including temperature (16-30°C), fan speed (low / medium / high), and mode (cooling / heating);

[0174] 2. Control commands must be reasonable (e.g., if the room temperature is 28°C in autumn, set the temperature to 24°C).

[0175] It should be noted that the above content regarding the third prompt word template is merely illustrative. In actual implementation, the third prompt word template can also adopt other design methods. For example, in addition to including the aforementioned fields, the third prompt word template can also include knowledge from the relevant equipment field. For example, it can incorporate knowledge from the air conditioning field such as "prioritize dehumidification in summer and prevent frost formation in winter" and "increase dehumidification frequency in the south and focus on rapid heating in the north."

[0176] Step 214: Based on the third prompt word, use the model to generate the third control command.

[0177] Step 214 can be implemented in various ways. Two examples are described below.

[0178] In the first implementation, in step 214, a third prompt word is input into the model to obtain a third control command output by the model. This model can be a large model or other machine learning models. Furthermore, in some embodiments, in addition to outputting the third control command, the model can also output corresponding energy consumption prediction values ​​and comfort prediction values. By adapting and adjusting the model architecture based on knowledge of the air conditioning domain and outputting adjusted energy consumption and comfort prediction values, the interpretability of the control commands output by the model is improved, further enhancing the user experience.

[0179] In the second implementation, in step 214, a third prompt word is input into the model to obtain the third control command output by the model. The adjustment command output by the model is then verified according to the set verification rules. If the verification passes, step 215 is executed. If the verification fails, the model is called again to output a new third control command, until the third control command passes the verification or the number of model calls reaches a set value. For details on how to set the verification rules, please refer to the relevant content of the foregoing embodiments.

[0180] In this embodiment, by distinguishing whether the user-input command information belongs to a scene-based command, it is convenient to adopt two different processing methods in different situations: generating only control commands or generating both control commands and target scenes. This improves the compatibility of the device operation control method with multiple output modes, taking into account both low energy consumption and support for scene construction in device operation control. Furthermore, by distinguishing whether the user-input command information belongs to a scene-based command and using different prompt word templates for different situations, and generating prompt words in different formats accordingly, the model can generate control commands more efficiently and accurately, improving the real-time performance and accuracy of the device operation control method, and further enhancing the user experience of using the target device.

[0181] Step 215: Control the target device to operate according to the third control command.

[0182] For example, assuming the target device is an air conditioner, and the third control command generated by the model is "set temperature to 26°C, set fan speed to low", then in step 215, the set temperature and set fan speed of the air conditioner can be adjusted according to the third control command.

[0183] In this embodiment, the above process enables the simultaneous consideration of real-time response and precision in equipment operation control, while also supporting scenario building functions. This improves the intelligence and efficiency of equipment operation control and reduces equipment energy consumption. Furthermore, by adapting and adjusting the model training and prediction process based on knowledge of the target device's domain, the accuracy and interpretability of the control strategy based on the model output can be improved. Therefore, the above equipment operation control method solves the problems existing in related art equipment operation control schemes.

[0184] Figure 4 This is a schematic diagram of the structure of a device operation control apparatus according to some embodiments of the present disclosure. For example... Figure 4 As shown, the equipment operation control device 40 includes an acquisition module 41, a judgment module 42, a construction module 43, a generation module 44, and a control module 45.

[0185] The acquisition module 41 is used to acquire the first instruction information input by the user.

[0186] The judgment module 42 is used to determine whether there is a scene in the scene library that matches the first instruction information when the first instruction information belongs to the scene type instruction information.

[0187] Module 43 is used to construct a first prompt word based on the first prompt word template corresponding to the scene-type instruction information, the first instruction information, and multi-source perception information when no scene matching the first instruction information exists in the scene library. The multi-source perception information includes environmental perception information and the target device's operating parameter information.

[0188] The generation module 44 is used to generate a target scene and a first control command corresponding to the target scene based on the first prompt word using the model.

[0189] The control module 45 is used to control the operation of the target device according to the first control command.

[0190] In some embodiments, the equipment operation control apparatus may further include modules required for performing other steps in the equipment operation control method as described above.

[0191] In this embodiment of the disclosure, the above device can support scene construction function while taking into account the real-time response and accuracy of equipment operation control, thereby improving the intelligence and efficiency of equipment operation control and reducing equipment energy consumption.

[0192] Figure 5 This is a schematic diagram of the structure of a device operation control apparatus according to other embodiments of this disclosure. For example... Figure 5 As shown, the device operation control apparatus 50 includes a memory 51 and a processor 52 coupled to the memory 51. The memory 51 is used to store instructions corresponding to embodiments of the device operation control method. The processor 52 is configured to execute the device operation control method in any of the embodiments of this disclosure based on the instructions stored in the memory 51.

[0193] Figure 6 This is a structural schematic diagram of an air conditioner according to some embodiments of this disclosure. For example... Figure 6 As shown, the air conditioner 60 includes an equipment operation control device 40.

[0194] The equipment operation control device 40 is used to execute the equipment operation control method as described above to control the operation of the air conditioner.

[0195] In some embodiments, in addition to outputting control commands through a model, the equipment operation control device 40 can also output predicted energy consumption and comfort levels through a model. Furthermore, the equipment operation control device can also feed back the predicted energy consumption and comfort levels of the air conditioner to the user, thereby making the energy-saving performance of the air conditioner and the satisfaction of human comfort indicators explicit.

[0196] In some embodiments, the air conditioner 60 can meet the following human comfort indicators: comfort scores are calculated based on various dimensions such as temperature fluctuation, temperature uniformity, vertical air temperature difference, wind feel index, PMV, and warm body dummy, and the average value of the comfort scores of each dimension is weighted and taken as the human comfort indicator, and the value of the indicator is greater than or equal to 4.5.

[0197] In some embodiments, the device operation control commands generated based on the model include one or more of the following: start operation command, stop operation command, operation mode adjustment command, temperature adjustment command, fan speed adjustment command, air direction adjustment command, compressor operating frequency adjustment command, humidity adjustment command, air quality adjustment command, and display adjustment command. In the above embodiments, the air conditioner 60 adjusts at least one of the following according to the control commands: start or stop operation at the correct time, operate in the corresponding mode, adjust the room temperature to the desired temperature, achieve the desired airflow direction and air volume, adjust the compressor operating frequency, adjust the room humidity to the desired humidity, adjust the room air quality (excluding temperature and humidity) to an excellent state, and adaptively adjust the display brightness.

[0198] In some embodiments, the air conditioner 60 can be connected to an industry public service platform, enabling it to be operated by the interactive terminals of other smart home manufacturers and to interact with other smart home appliances and products. The functions of the air conditioner that can be operated by other interactive terminals include one or more of the following: power on / off, temperature adjustment, mode adjustment, and fan speed adjustment.

[0199] In this embodiment of the disclosure, the above-mentioned air conditioner can support scene construction function while taking into account the real-time response and accuracy of air conditioner operation control, thereby improving the intelligence and efficiency of air conditioner operation control and reducing equipment energy consumption.

[0200] Hereinafter, various aspects of the present disclosure are described with reference to flowchart illustrations and / or block diagrams of a device operation control method, apparatus, storage medium, and computer program product according to embodiments of the present disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks, can be implemented by computer-readable program instructions.

[0201] These computer-readable program instructions are provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable device to produce a machine, such that execution of the instructions by the processor produces means for implementing the functions specified in one or more boxes of the flowchart and / or block diagram.

[0202] These computer-readable program instructions may also be stored in a computer-readable storage medium. These instructions cause a computer to work in a particular manner to produce an article of manufacture, including instructions that implement the functions specified in one or more boxes in a flowchart and / or block diagram.

[0203] This disclosure may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects.

[0204] The device operation control method, apparatus, and storage medium according to this disclosure have been described in detail above. To avoid obscuring the concept of this disclosure, some details known in the art have not been described. Those skilled in the art can fully understand how to implement the technical solutions disclosed herein based on the above description.

Claims

1. A method for controlling equipment operation, comprising: Obtain the first instruction information input by the user; If the first instruction information is a scene-type instruction information, determine whether there is a scene in the scene library that matches the first instruction information; If no scene matching the first instruction information exists in the scene library, a first prompt word is constructed based on the first prompt word template corresponding to the scene-type instruction information, the first instruction information, and multi-source perception information, wherein the multi-source perception information includes environmental perception information and target device operating parameter information; Based on the first prompt word, a target scene and a first control command corresponding to the target scene are generated using the model; The target device is controlled to operate according to the first control command.

2. The method according to claim 1, further comprising: If a scene matching the first instruction information exists in the scene library, the target device is controlled to operate according to the second control instruction corresponding to the matching scene in the scene library.

3. The method according to claim 1, wherein, Determining whether a scene matching the first instruction information exists in the scene library includes: Calculate the similarity between the first instruction information and each scene in the scene library; The scene with the highest similarity is determined from the scene library; If the similarity of the scene with the highest similarity is greater than the similarity threshold, then the scene with the highest similarity is taken as the scene that matches the first instruction information. If the similarity of the scene with the highest similarity is less than or equal to the similarity threshold, it is determined that there is no scene that matches the first instruction information.

4. The method according to any one of claims 1 to 3, further comprising: Update the scene library according to the target scene and the first control command; And / or, The system acquires second instruction information input by the user, wherein the second instruction information is adjustment indication information feedback to the first control instruction; constructs a second prompt word based on a second prompt word template corresponding to the adjustment indication information and the second instruction information, wherein the second prompt word template is different from the first prompt word template; generates an adjustment instruction for the first control instruction using the model based on the second prompt word; controls the operation of the target device according to the adjustment instruction; and modifies the first control instruction corresponding to the target scene in the scene library according to the adjustment instruction.

5. The method according to claim 4, wherein, The scene library also stores the number of corrections for each scene and the similarity threshold corresponding to each scene. The method further includes: Upon receiving the second instruction from the user, update the number of corrections for the target scene in the scene library; If the number of corrections for the target scene exceeds the number threshold, the similarity threshold corresponding to the target scene is adjusted.

6. The method according to claim 1, further comprising: When the first instruction information is a non-scene-type instruction information, a third prompt word is constructed based on the third prompt word template corresponding to the non-scene-type instruction information, the first instruction information, and the multi-source perception information, wherein the third prompt word template is different from the first prompt word template; Based on the third prompt word, the model is used to generate a third control command; The target device is controlled to operate according to the third control command.

7. The method according to any one of claims 1 to 6, wherein, The target device is an air conditioner, and the model is a Transformer-based network model, including a decoder, and an instruction prediction layer, an energy consumption prediction layer, and a comfort prediction layer connected to the decoder. The method further includes: While using the model to output the first control command, the system also outputs the energy consumption prediction value of the target device within a set time period after executing the first control command, as well as the user's comfort prediction value after executing the first control command.

8. The method according to claim 7, wherein: The decoder includes at least one attention layer based on a temporal attention mechanism; and / or The model is trained using a joint loss function, which includes: a first function term representing the loss calculated based on the difference between the instruction generated by the model and a pre-labeled instruction; a second function term representing the loss calculated based on the difference between the energy consumption prediction value obtained from the large model and the pre-labeled minimum energy consumption; a third function term representing the reward given when the set temperature indicated by the instruction generated by the model is within the human comfort range; and a fourth function term representing the penalty given when the instruction generated by the model violates the domain physical rules or safe operating procedures of the target device.

9. The method according to any one of claims 1 to 6, wherein: The environmental sensing information includes at least one of the following: whether the current date is a holiday, weather, outdoor temperature, indoor temperature, outdoor humidity, and indoor humidity; and / or The target device's operating parameter information includes at least one of the following: current operating mode, current set temperature, and current set windshield; and / or The multi-source sensing information also includes at least one of the following: the user's historical preferences, the user's health record, the geographical location of the target device, the current season, the current time period, and the electricity price information corresponding to the current time period.

10. A device for controlling equipment operation, comprising: A module for performing the equipment operation control method as described in any one of claims 1 to 9.

11. A device for controlling equipment operation, comprising: Memory; as well as A processor coupled to the memory, the processor being configured to execute the device operation control method as described in any one of claims 1 to 9 based on instructions stored in the memory.

12. An air conditioner, comprising: The equipment operation control device as described in claim 10 or 11.

13. A computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the device operation control method as described in any one of claims 1 to 9.

14. A computer program product having stored computer program instructions thereon, which, when executed by a processor, implement the device operation control method as described in any one of claims 1 to 9.

Images