Multi-intention control method and device based on large model, storage medium and electronic device

Abstract

The invention discloses a multi-intention control method and device based on a large model, a storage medium and an electronic device, and relates to the technical field of smart home, and the method comprises the steps: carrying out the intention recognition of a voice instruction corresponding to a target object, so as to determine a plurality of to-be-executed intentions indicated by the voice instruction; each to-be-executed intention is input into a preset large language model, so that the preset large language model extracts a keyword corresponding to each to-be-executed intention, the intention type corresponding to each to-be-executed intention is recognized according to the keyword corresponding to each to-be-executed intention, and the intention type comprises an equipment control type intention and a skill type intention; the conversion mode of each to-be-executed intention is determined according to the intention type, and each to-be-executed intention is converted into each target instruction according to the conversion mode, so that each target to-be-executed device executes each target instruction, and the target instructions at least comprise one of a control type instruction and a skill type instruction.

Description

Technical Field

[0001] The present application relates to the field of smart home technology, and more specifically, to a multi-intention control method and device based on a large model, a storage medium, and an electronic device. Background Art

[0002] In the smart home sector, with the rapid development and widespread adoption of IoT technology, the way users interact with smart home devices is undergoing a revolution, shifting from traditional manual operation to voice control. Voice control technology, particularly the application of natural language processing, has greatly improved the convenience and intelligence of interaction. Users can now synchronize and operate a complete set of devices with simple voice commands, such as "Good evening, prepare for bedtime."

[0003] However, existing technologies have many limitations when processing multi-device, multi-intention control commands. Most smart home systems can only recognize and respond to control commands for a single device when processing user voice commands. For example, when a user says "turn on the lights," the system can recognize and control the corresponding lights. However, for complex commands that include control intent for multiple devices, such as "turn on the lights, turn off the air conditioner, play light music," existing technologies often have difficulty converting each of them into corresponding control commands.

[0004] Regarding the problem in related technologies that it is impossible to accurately convert each of multiple intents into corresponding control instructions, no effective solution has been proposed yet. Summary of the Invention

[0005] The embodiments of the present application provide a multi-intention control method and device, a storage medium and an electronic device based on a large model, so as to at least solve the problem in the related art that each intention in multiple intentions cannot be accurately converted into a corresponding control instruction.

[0006] According to one embodiment of the embodiments of the present application, a multi-intention control method based on a large model is provided, including: performing intent recognition on the voice instruction corresponding to the target object to determine multiple intentions to be executed indicated by the voice instruction; inputting each intention to be executed into a preset large language model, so that the preset large language model extracts the keywords corresponding to each intention to be executed, and identifies the intention type corresponding to each intention to be executed according to the keywords corresponding to each intention to be executed, wherein the preset large language model is a model generated according to the generalized preset keywords and the preset intention types corresponding to the generalized preset keywords, and the intention types include: device control type intentions and skill type intentions; determining the conversion method of each intention to be executed according to the intention type, and converting each intention to be executed into each target instruction according to the conversion method, so that each target device to be executed executes each target instruction, wherein the target instruction includes at least one of the following: control type instruction, skill type instruction.

[0007] In an exemplary embodiment, a conversion method for each intention to be executed is determined according to the intention type, and each intention to be executed is converted into each target instruction according to the conversion method, so that each target device to be executed executes each target instruction, including: when the multiple intentions to be executed include device control class intentions, device information of multiple first devices to be executed is determined, and conversion methods corresponding to multiple device control class intentions are determined according to the device information, and a sending strategy for each control class instruction corresponding to each device control class intention is determined, wherein the target devices to be executed include: one or more first devices to be executed; each device control class intention is converted into each control class instruction according to the conversion method, and each control class instruction is sent to each first device to be executed according to the sending strategy, so that the first device to be executed executes the control class instruction.

[0008] In an exemplary embodiment, device information of multiple first devices to be executed is determined, and conversion methods corresponding to multiple device control class intentions are determined based on the device information, as well as a sending strategy for each control class instruction corresponding to each device control class intention is determined, including: when the device information indicates that the multiple first devices to be executed are not the same device to be executed, respectively determining the first control version supported by each first device to be executed, wherein the instruction format and device control protocol of the target instructions corresponding to different control versions are different; respectively converting each intention to be executed into each first control class instruction according to each first control version, wherein each first control class instruction is an instruction corresponding to each first control version, and the control class instructions include: each first control class instruction; and respectively sending each first control class instruction to each first device to be executed, so that each first device to be executed executes each first control class instruction.

[0009] In an exemplary embodiment, device information of multiple first devices to be executed is determined, and conversion methods corresponding to multiple device control class intentions are determined based on the device information, as well as a sending strategy for each control class instruction corresponding to each device control class intention is determined, including: when the device information indicates that the multiple first devices to be executed are the same device to be executed, determining the second control version supported by the first device to be executed; converting each device control class intention into each second control class instruction based on the second control version, wherein each second control class instruction is a control instruction corresponding to the second control version, and the control class instruction includes: the second control class instruction; determining a first priority of each second control class instruction, and sending each second control class instruction to the first device to be executed in sequence based on the first priority, so that the first device to be executed executes each second control class instruction.

[0010] In an exemplary embodiment, a conversion method for each intention to be executed is determined according to the intention type, and each intention to be executed is converted into each target instruction according to the conversion method, so that each target device to be executed executes each target instruction, including: when the multiple intentions to be executed include skill-type intentions, determining the second device to be executed corresponding to the skill-type intention, and converting the skill-type intention into a skill-type instruction; sending the skill-type instruction to the second device to be executed, so that the second device to be executed displays the first execution result corresponding to the skill-type instruction on the target screen corresponding to the second device to be executed, and / or, voice broadcasting the first execution result.

[0011] In an exemplary embodiment, after each intention to be executed is input into a preset large language model, the method further includes: performing an intention conflict check on each intention to be executed based on the preset large language model to determine whether there is a conflicting intention among the multiple intentions to be executed, wherein the conflicting intention is an intention in which the execution results corresponding to the multiple intentions are conflicting, and / or an intention in which the execution result corresponding to the intention is consistent with the current execution state of the target device to be executed; in the case of determining that there are multiple first conflicting intentions in which the execution results corresponding to the multiple intentions are conflicting among the multiple intentions to be executed, obtaining the historical execution record of the target device to be executed through the preset large language model, and obtaining the historical intention set corresponding to the target device to be executed in the current time period based on the historical execution record, wherein the historical execution record at least includes: the historical intention corresponding to each historical instruction executed by the target device to be executed; comparing each first conflicting intention with each of the historical intention sets. Intentions are detected for similarity, and the multiple first conflicting intentions are modified according to the similarity detection results; a secondary intention conflict check is performed on the modified multiple first conflicting intentions based on the preset large language model to determine whether there are conflicting intentions among the modified multiple first conflicting intentions; in the case of determining that there are conflicting intentions among the modified multiple first conflicting intentions, the execution frequency of the historical instructions corresponding to each modified first conflicting intention is determined according to the historical execution record, and the first conflicting intention with an execution frequency less than the preset execution frequency among the modified multiple first conflicting intentions is deleted according to the execution frequency, so that there are no conflicting intentions among the multiple intentions to be executed; in the case that there is any conflicting intention among the multiple intentions to be executed, the execution result corresponding to the intention is consistent with the current execution state of the target device to be executed, the any conflicting intention is deleted through the preset large language model, so that there are no conflicting intentions among the multiple intentions to be executed.

[0012] In an exemplary embodiment, after converting each intention to be executed into each target instruction according to the conversion method, the method further includes: obtaining each second execution result corresponding to each target to-be-executed device, and determining the second priority corresponding to each second execution result; integrating multiple second execution results into a message list according to the second priority, and displaying the message list on a screen in the control device, wherein the second execution results in the message list are arranged based on the second priority.

[0013] According to another embodiment of the embodiments of the present application, a multi-intention control device based on a large model is also provided, including: an identification module, used to perform intent recognition on the voice instruction corresponding to the target object to determine multiple intentions to be executed indicated by the voice instruction; an input module, used to input each intention to be executed into a preset large language model, so that the preset large language model extracts the keywords corresponding to each intention to be executed, and identifies the intention type corresponding to each intention to be executed according to the keywords corresponding to each intention to be executed, wherein the preset large language model is a model generated based on the generalized preset keywords and the preset intention types corresponding to the generalized preset keywords, and the intention types include: device control type intentions and skill type intentions; a determination module, used to determine the conversion method of each intention to be executed according to the intention type, and convert each intention to be executed into each target instruction according to the conversion method, so that each target device to be executed executes each target instruction, wherein the target instruction includes at least one of the following: control type instruction, skill type instruction.

[0014] According to another aspect of an embodiment of the present application, a computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program is configured to execute the above-mentioned large model-based multi-intention control method when running.

[0015] According to another aspect of an embodiment of the present application, an electronic device is also provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the large model-based multi-intention control method through the computer program.

[0016] According to another aspect of the embodiments of the present application, a computer program product is provided, including a computer program, wherein the computer program is executed by a processor to perform the method.

[0017] In an embodiment of the present application, the voice command corresponding to the target object is subjected to intent recognition to determine multiple intentions to be executed indicated by the voice command; each intention to be executed is input into a preset large language model generated based on the generalized preset keywords and the preset intention types corresponding to the generalized preset keywords, so that the preset large language model identifies the intention type corresponding to each intention to be executed based on the keywords corresponding to each intention to be executed, wherein the intention types include: device control type intentions and skill type intentions; the conversion method of each intention to be executed is determined based on the intention type, and each intention to be executed is converted into each target instruction based on the corresponding conversion method, so that each target device to be executed executes each target instruction, wherein the target instruction may be a control type instruction and / or a skill type instruction. That is, according to an embodiment of the present application, the keywords corresponding to each intention to be executed can be extracted based on the preset large language model, and then the intention type corresponding to each intention to be executed can be identified based on the keywords corresponding to each intention to be executed. Furthermore, different conversion methods corresponding to the intentions can be determined according to different intention types, and the intentions to be executed of different types can be converted into target instructions according to different conversion methods. Through the embodiments of the present application, the problem in the related art that each intention in multiple intentions cannot be accurately converted into a corresponding control instruction can be solved. Then, by determining the intention type of the intention to be executed and determining different conversion methods, each intention in the multiple intentions is accurately converted into a corresponding control instruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0019] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0020] Figure 1 Schematic diagram of the hardware environment of a multi-intention control method based on a large model according to an embodiment of the present application;

[0021] Figure 2 is a flow chart of a multi-intention control method based on a large model according to an embodiment of the present application;

[0022] Figure 3 This is a flowchart of a multi-device multi-intention control method in a smart home scenario according to an optional embodiment of the present application;

[0023] Figure 4This is a structural block diagram of a multi-intention control device based on a large model according to an embodiment of the present application. DETAILED DESCRIPTION

[0024] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0025] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in a sequence other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] According to one aspect of the embodiment of the present application, a multi-intention control method based on a large model is provided. The multi-intention control method based on a large model is widely used in smart home (Smart Home), smart home, smart home device ecology, smart residential (Intelligence House) ecology and other whole-house intelligent digital control application scenarios. Optionally, Figure 1 This is a hardware environment diagram of a multi-intention control method based on a large model according to an embodiment of the present application. In this embodiment, the multi-intention control method based on a large model can be applied to Figure 1 In the hardware environment shown in FIG. 1 , a central control device 102 and a server 104 are formed. Figure 1 As shown, the server 104 is connected to the central control device 102 via a network, and can be used to provide services (such as application services, etc.) for the terminal or the client installed on the terminal. A database can be set up on the server or independently of the server to provide data storage services for the server 104. Cloud computing and / or edge computing services can be configured on the server or independently of the server to provide data calculation services for the server 104.

[0027] The aforementioned network may include, but is not limited to, at least one of the following: a wired network and a wireless network. The aforementioned wired network may include, but is not limited to, at least one of the following: a wide area network, a metropolitan area network, and a local area network. The aforementioned wireless network may include, but is not limited to, at least one of the following: WIFI (Wireless Fidelity) and Bluetooth. The central control device 102 may be, but is not limited to, a PC, a mobile phone, a tablet computer, a smart air conditioner, a smart range hood, a smart refrigerator, a smart oven, a smart stove, a smart washing machine, a smart water heater, a smart washing machine, a smart dishwasher, a smart projection device, a smart TV, a smart clothes drying rack, smart curtains, smart audio and video, a smart socket, a smart speaker, a smart fresh air device, smart kitchen and bathroom equipment, smart bathroom equipment, a smart sweeping robot, a smart window cleaning robot, a smart mopping robot, a smart air purification device, a smart steamer, a smart microwave oven, a smart kitchen treasure, a smart purifier, a smart water dispenser, a smart door lock, etc.

[0028] In this embodiment, a multi-intention control method based on a large model is provided. Figure 2 : is a flowchart of a multi-intention control method based on a large model according to an embodiment of the present application, the process includes the following steps:

[0029] Step S202: performing intent recognition on the voice instruction corresponding to the target object to determine multiple intentions to be executed indicated by the voice instruction;

[0030] The voice command can be converted into text content through a speech recognition method (Automatic Speech Recognition, referred to as ASR), and then the text content is subjected to intent recognition to determine multiple intentions to be executed.

[0031] ASR refers to the process of converting human speech signals into text or commands that computers can understand. This process involves capturing, analyzing, extracting features from sound signals, and ultimately converting them into text.

[0032] Step S204: Input each intent to be executed into a preset large language model, so that the preset large language model extracts keywords corresponding to each intent to be executed, and identifies the intent type corresponding to each intent to be executed based on the keywords corresponding to each intent to be executed, wherein the preset large language model is a model generated based on generalized preset keywords and preset intent types corresponding to the generalized preset keywords, and the intent types include: device control intent and skill intent;

[0033] Device control intents refer to instructions directly related to physical device operations, intended to change the state or behavior of one or more devices. Device control intents generally include but are not limited to:

[0034] Switch operations: for example, "turn on the lights" and "close the curtains".

[0035] Mode adjustment: For example, "set the air conditioner to sleep mode."

[0036] Parameter settings: for example, "turn the TV volume to 50%", "preheat the oven to 200 degrees".

[0037] Scenario mode: for example, "activate away mode" (may include turning off all electrical appliances, locking the door, etc.).

[0038] The execution of device control intents usually involves direct communication with specific smart home devices, converting the user's intent into instructions that the device can understand, such as: turning on, turning off, adjusting parameters, etc.

[0039] Skill-based intents refer to instructions that call cloud services or application programming interfaces (APIs) to perform more advanced or complex tasks. These intents can cover information queries, entertainment services, health advice, and other aspects. For example:

[0040] Information query: for example, "What's the weather like in Beijing today?", "Check flights for tomorrow afternoon."

[0041] Entertainment services: for example, "play soothing music" or "tell a joke".

[0042] Calendar management: For example, "Remind me to have a meeting at 3 pm."

[0043] Smart suggestions: For example, "What exercises should I do to stay healthy?"

[0044] Skill intent processing typically doesn't involve directly controlling physical devices. Instead, it's accomplished by calling APIs of cloud services or applications. These services or applications might run on cloud servers and provide various value-added services and intelligent features. The execution of a skill intent might involve interacting with multiple services and obtaining data and information from the network.

[0045] The intent type of each intent to be executed can be determined based on a preset large language model, wherein the preset large language model is a model generated after adjusting the parameters in the preset model. The preset model can be a large model, such as: Generative Pre-Trained Transformer (GPT for short).

[0046] The process of training a large model may include: collecting a series of words or phrases related to device control and skill services as preset keywords, such as: open, close, mode, temperature, query, play, weather, recipe, etc.

[0047] The preset keywords are then input into the preset model, allowing the preset model to learn the semantics and contextual relationships of these preset keywords and then perform generalization training on the preset keywords. The purpose of generalization training is to enable the preset model to understand that even if it encounters new words or expressions that it has never seen before, it can infer the possible intent type based on its position and context in the sentence. For example, although "turn on" and "open" are expressed differently, the preset model can recognize through generalization training that they both may belong to the "turn on operation" in the "device control intent" category.

[0048] Finally, based on the generalized preset keywords and the preset intent types to which these generalized preset keywords belong (i.e., device control intent or skill intent), the preset model is fine-tuned and the parameters of the preset model are adjusted to generate a preset large language model, so that the preset large language model can more accurately identify and classify the user's intention to be executed when encountering new voice commands.

[0049] After generating the preset large language model, each intention to be executed can be input into the preset large language model respectively. The preset large language model can identify the keywords of each intention to be executed to determine the intention type corresponding to each intention to be executed.

[0050] Through the above technical solution, by inputting preset keywords into the preset large language model for generalization training, the model can understand a wider range of language expressions and contexts, thereby more accurately identifying device control intentions and skill intentions in voice commands, thereby improving the accuracy of intent recognition.

[0051] Step S206, determine the conversion method of each intention to be executed according to the intention type, and convert each intention to be executed into each target instruction according to the conversion method, so that each target device to be executed executes each target instruction, wherein the target instruction includes at least one of the following: control type instruction, skill type instruction.

[0052] Through the above steps, the voice command corresponding to the target object is subjected to intention recognition to determine multiple intentions to be executed indicated by the voice command; each intention to be executed is input into a preset large language model generated based on the generalized preset keywords and the preset intention types corresponding to the generalized preset keywords, so that the preset large language model identifies the intention type corresponding to each intention to be executed according to the keywords corresponding to each intention to be executed, wherein the intention types include: device control type intentions and skill type intentions; the conversion method of each intention to be executed is determined according to the intention type, and each intention to be executed is converted into each target instruction according to the corresponding conversion method, so that each target device to be executed executes each target instruction, wherein the target instruction may be a control type instruction and / or a skill type instruction. That is, according to the embodiment of the present application, the keywords corresponding to each intention to be executed can be extracted based on the preset large language model, and then the intention type corresponding to each intention to be executed can be identified according to the keywords corresponding to each intention to be executed. Furthermore, different conversion methods corresponding to the intentions can be determined according to different intention types, and the intentions to be executed of different types can be converted into target instructions according to different conversion methods. Through the embodiment of the present application, the problem in the related art that each intention in multiple intentions cannot be accurately converted into a corresponding control instruction can be solved. Then, by determining the intention type of the intention to be executed and determining different conversion methods, each intention in the multiple intentions is accurately converted into a corresponding control instruction.

[0053] Optionally, determining a conversion method for each of the intentions to be executed according to the intention type, and converting each of the intentions to be executed into each target instruction according to the conversion method so that each target device to be executed executes each target instruction, includes:

[0054] (1) For device control intent:

[0055] In the case where the multiple intentions to be executed include device control class intentions, the device information of multiple first devices to be executed is determined, and the conversion methods corresponding to the multiple device control class intentions are determined based on the device information, and the sending strategy for each control class instruction corresponding to each device control class intention is determined, wherein the target devices to be executed include: one or more first devices to be executed; each device control class intention is converted into each control class instruction according to the conversion method, and each control class instruction is sent to each first device to be executed according to the sending strategy, so that the first device to be executed executes the control class instruction.

[0056] Among them, when the device information indicates that the multiple first devices to be executed are not the same device to be executed, the first control version supported by each of the first devices to be executed is determined respectively, wherein the instruction format and device control protocol of the target instructions corresponding to different control versions are different; each intention to be executed is converted into each first control class instruction according to each first control version, wherein each first control class instruction is an instruction corresponding to each first control version, and the control class instructions include: each first control class instruction; each first control class instruction is sent to each first device to be executed respectively, so that each first device to be executed executes each first control class instruction.

[0057] Among them, when the device information indicates that the multiple first devices to be executed are the same device to be executed, the second control version supported by the first device to be executed is determined; each device control class intention is converted into each second control class instruction according to the second control version, wherein each second control class instruction is a control instruction corresponding to the second control version, and the control class instructions include: the second control class instruction; the first priority of each second control class instruction is determined, and each second control class instruction is sent to the first device to be executed in sequence according to the first priority, so that the first device to be executed executes each second control class instruction.

[0058] It is understandable that the conversion method for pending intents of different intent types is different, among which:

[0059] (1) The processing method for device control intent is as follows:

[0060] Determine the first device to be executed corresponding to each device control intent. For example, the device control intent of the target object is "turn on the light", "close the curtains", and "adjust the air conditioner to 25 degrees".

[0061] For "turn on the lights", the device to be executed may be the lighting device; for "close the curtains", the device to be executed may be the curtain device; for "adjust the air conditioner to 25 degrees", the device to be executed may be the air conditioner device.

[0062] After determining each first device to be executed, the device information of each first device to be executed can be determined, wherein the device information includes at least one of the following: a device identification, device status information, and device hardware configuration information of each first device to be executed; based on the device identification, it can be determined whether multiple first devices to be executed are the same device to be executed. Furthermore, based on whether the multiple devices to be executed are the same device to be executed, the conversion methods corresponding to the device control intent are divided into two categories, specifically:

[0063] 1) If multiple first pending devices are determined to be different based on device information, it is necessary to determine the first control version corresponding to each first pending device. Different control versions correspond to different instruction formats and control protocols. For example, a lighting device may support control version 1.0, a curtain device may support control version 2.0, and an air conditioner may support control version 3.0.

[0064] Each device control intent is converted into a corresponding first control instruction based on the different first control versions. For example, the "turn on the lights" instruction is converted into a version 1.0 control instruction if supported by the lighting device, "close the curtains" is converted into a version 2.0 control instruction, and "set the air conditioner to 25 degrees" is converted into a version 3.0 control instruction.

[0065] Then, each first control-type instruction is sent to each corresponding first device to be executed, so that each first device to be executed can execute the corresponding first control instruction.

[0066] 2) When it is determined based on device information that multiple first devices to be executed are the same device to be executed, the second control version supported by the target device to be executed can be determined, for example: "Turn on the air conditioner, adjust to 25 degrees, and enter sleep mode". These device control intentions are all aimed at air-conditioning equipment, and it is only necessary to determine whether the air-conditioning equipment supports control version 3.0.

[0067] Each device control intent can then be converted into a second control instruction according to the second control version. For example, "turn on the air conditioner," "adjust to 25 degrees," and "enter sleep mode" can all be converted into instructions of control version 3.0.

[0068] After converting the second control class instructions, the priority of each second control class instruction can be determined, and then the execution order of the second control class instructions by the second to-be-executed device can be determined based on the priority. For example, the "turn on the air conditioner" instruction may be executed first, followed by the "adjust to 25 degrees" instruction, and finally the "enter sleep mode" instruction.

[0069] (2) The processing method for skill-related intent is as follows:

[0070] When the multiple intentions to be executed include skill-type intentions, determine the second device to be executed corresponding to the skill-type intention, and convert the skill-type intention into a skill-type instruction; send the skill-type instruction to the second device to be executed, so that the second device to be executed displays the first execution result corresponding to the skill-type instruction on the target screen corresponding to the second device to be executed, and / or, voice broadcasts the first execution result.

[0071] It is understandable that for skill-type intents, it is first necessary to determine the skill-type intent corresponding to the second device to be executed. Among them, these skill-type intents may involve cloud service calls, information queries, entertainment services, etc. For example, "Query the weather in Beijing today" or "Play a soothing piece of music" are skill-type intents that belong to weather query and music playback respectively. They may be targeted at a smart speaker with voice assistant function (i.e., the second device to be executed).

[0072] Secondly, after determining the device to be executed, the skill intent can be converted into a skill instruction, and then the skill instruction is sent to the second device to be executed, for example:

[0073] After the second-to-be-executed device, the smart speaker, receives the skill-type instruction weather query instruction, it will obtain weather information from the cloud service; after receiving the music playback instruction, it will select a soothing music from the music library and play it.

[0074] Through the embodiments of the present application, it is possible to intelligently distinguish whether the user's voice command is intended to control the device (such as turning on / off the lights, adjusting the air conditioning temperature) or to request skill-related intentions (such as checking the weather, playing music), and adopt corresponding conversion methods according to different types to convert the intention into control commands that the device can recognize or API calls for cloud services, thereby improving the accuracy of command recognition and processing efficiency.

[0075] Optionally, after inputting each intention to be executed into the preset large language model in the above step S206, the method further includes: performing an intention conflict check on each intention to be executed based on the preset large language model to determine whether there are conflicting intentions among the multiple intentions to be executed, wherein the conflicting intention is an intention whose execution results corresponding to multiple intentions conflict, and / or an intention whose execution result is consistent with the current execution state of the target device to be executed;

[0076] In a case where it is determined that there are multiple first conflicting intentions among the multiple intentions to be executed, wherein the execution results corresponding to the multiple intentions conflict, obtaining a historical execution record of the target device to be executed through the preset large language model, and obtaining a historical intention set corresponding to the target device to be executed in the current time period based on the historical execution record, wherein the historical execution record at least includes: a historical intention corresponding to each historical instruction executed by the target device to be executed;

[0077] Performing a similarity check on each first conflicting intention with each intention in the historical intention set, and modifying the multiple first conflicting intentions based on the similarity test results; performing a secondary intention conflict check on the modified multiple first conflicting intentions based on the preset large language model to determine whether there is a conflicting intention among the modified multiple first conflicting intentions;

[0078] In the case where it is determined that there is a conflicting intention among the modified multiple first conflicting intentions, the execution frequency of the historical instructions corresponding to each modified first conflicting intention is determined according to the historical execution record, and the first conflicting intention among the modified multiple first conflicting intentions whose execution frequency is less than the preset execution frequency is deleted according to the execution frequency, so that there is no conflicting intention among the multiple intentions to be executed; in the case where there is any conflicting intention among the multiple intentions to be executed, the execution result corresponding to the intention is consistent with the current execution state of the target device to be executed, the any conflicting intention is deleted through the preset large language model, so that there is no conflicting intention among the multiple intentions to be executed.

[0079] It is understandable that after the preset large language model receives multiple intentions to be executed, it is necessary to perform an intention conflict check on the multiple intentions to be executed based on the preset large language model to determine whether there are conflicting intentions among the multiple intentions to be executed. The embodiment of the present application also defines the processing methods corresponding to the presence of conflicting intentions and the absence of conflicting intentions, specifically:

[0080] When it is determined that there is no conflict intention, the above steps S204 and S206 can be directly executed.

[0081] In cases where it is determined that there are conflicting intentions, the type of conflicting intention needs to be determined:

[0082] In the case where the conflicting intention is determined to be multiple first conflicting intentions in which the execution results corresponding to multiple intentions conflict, the historical execution records of the target device to be executed can be obtained based on the preset large language model, wherein the historical execution records may include: the historical instructions received by the target device to be executed, the historical execution results corresponding to each historical instruction, and each historical intention corresponding to each historical instruction; then, each first conflicting intention is modified according to the historical execution records, and a secondary intention check is performed on the modified first conflicting intention; in the case where multiple modified first conflicting intentions do not pass the secondary intention check, it is necessary to determine whether to delete each modified first conflicting intention based on the execution frequency corresponding to each modified first conflicting intention, and determine the specific conflicting intention to be deleted. After deleting the multiple first conflicting intentions, the multiple first conflicting intentions can be replaced with multiple deleted first conflicting intentions, and the multiple intentions to be executed after the replacement intentions are subjected to three conflict processings, and then it is determined whether the multiple intentions to be executed after the replacement processing pass the conflict check.

[0083] When the multiple intentions to be executed after the replacement process pass the conflict check, it is determined that there are no conflicting intentions among the multiple intentions to be executed after the replacement process.

[0084] Among them, "delete the first conflicting intention whose execution frequency is less than the preset execution frequency among the modified multiple first conflicting intentions according to the execution frequency", the preset execution frequency can be set according to the multiple execution frequencies corresponding to the multiple first conflicting intentions, for example: the execution frequency of the first conflicting intention is 3 times, the execution frequency of the second conflicting intention is 5 times, and the execution frequency of the third conflicting intention is 8 times. The execution frequencies corresponding to the multiple first conflicting intentions are sorted from small to large, and the first conflicting intentions are deleted in order from small to large according to the arrangement order of the execution frequencies. When a conflicting intention is deleted, the deleted conflicting intention is used to replace the multiple first intentions to be executed, and it is determined whether there is a conflicting intention in the multiple intentions to be executed after replacement. When it is determined that there is still a conflicting intention, the conflicting intention corresponding to the second execution frequency after sorting from small to large is deleted, and the above deletion, replacement and conflict checking operations are repeated until it is determined that there is no conflicting intention in the multiple intentions to be executed after replacement.

[0085] When it is determined that the conflicting intention is any conflicting intention whose execution result corresponding to the intention is consistent with the current execution state of the target device to be executed, any conflicting intention can be deleted based on the preset large language model.

[0086] For example, when the conflicting command is "turn on the air conditioner, turn off the air conditioner," the conflicting command can be fed back to the target object through voice feedback: "Sending the commands to turn on and off the air conditioner at the same time is a conflict, please select one." The user can then confirm to delete one of the commands through voice or the interactive interface to resolve the conflict.

[0087] If a conflicting instruction is determined to be "an instruction whose second execution result corresponds to any control instruction that is consistent with the current execution state of the target device to be executed," the instruction can be automatically ignored to avoid repeated execution. For example, if a user says "Turn on the air conditioner," but the air conditioner is already on, the "Turn on the air conditioner" instruction can be automatically identified and ignored to avoid meaningless operations.

[0088] Through the embodiments of this application, in scenarios where multiple devices are controlled, conflict detection is not limited to different commands within the same device, but can also include command conflicts between different devices. For example, executing "open window" and "turn on air conditioning" at the same time may require additional energy consumption or affect device performance. By avoiding such conflicts, device operation can be managed more coordinatedly, achieving efficient collaboration between devices.

[0089] Optionally, after converting each intention to be executed into each target instruction according to the conversion method in the above step S206, the method also includes: obtaining each second execution result corresponding to each target device to be executed, and determining the second priority corresponding to each second execution result; integrating multiple second execution results into a message list according to the second priority, and displaying the message list on the screen of the control device, wherein the second execution results in the message list are arranged based on the second priority.

[0090] It is understandable that the priority of the execution result corresponding to the device control intent is greater than the priority of the execution result corresponding to the skill intent. After converting each intent to be executed into each control instruction, it is also necessary to obtain the second execution result of each target device, integrate multiple second execution results according to the second priority of the second execution result, and integrate multiple second execution results into a message list. In other words, the device control services, skill services and the language in the intent parsing context can be integrated. The final broadcast content is sorted and optimized and organized into a message list. The execution results corresponding to the device control intent are placed first in the message list, followed by text content such as weather and recipes, and finally audio resources such as music. The integrated execution results can be displayed on the screen of the control device, or the integrated results can be read out through voice broadcast.

[0091] In order to better understand the process of the above-mentioned multi-intention control method based on a large model, the implementation method flow of the above-mentioned multi-intention control based on a large model is described below in combination with an optional embodiment, but it is not used to limit the technical solution of the embodiment of this application.

[0092] The optional embodiment of this application defines a multi-device multi-intention control method in a smart home scenario. Figure 3 This is a flow chart of a multi-device multi-intention control method in a smart home scenario according to an optional embodiment of the present application. Figure 3 As shown:

[0093] Step S301: The user (i.e., the target object) inputs voice content;

[0094] Step S302, converting the recognized voice (i.e., voice command) into text content;

[0095] The system pre-processes the input speech, extracts key sound features, and builds a recognition template based on these features. During the recognition process, the system analyzes and recognizes the input speech signal based on these templates.

[0096] Modern speech recognition technology typically relies on machine learning and deep learning algorithms, particularly neural network models. These models are capable of learning and understanding large amounts of speech data, accurately converting speech to text. Neural networks built using deep learning frameworks such as TensorFlow or PyTorch can be trained to produce efficient speech-to-text conversion models.

[0097] Speech recognition-to-text technology achieves the conversion from sound to text through complex signal processing and pattern recognition processes, greatly promoting the naturalness and convenience of human-computer interaction.

[0098] For example, after waking up the interactive device, the user can directly say "turn on the lights, set the air conditioner to comfort mode, 26 degrees, and play music". After the device completes the voice pickup, it converts it into corresponding text through the voice recognition service for the next processing.

[0099] Step S303: The scheduling service performs intent recognition on the text content;

[0100] Step S304, generating an intent to be processed;

[0101] That is, intent parsing is performed to generate intents to be processed (i.e., multiple intents to be executed). The intent parsing service is a technical service for natural language processing. In order to improve the accuracy and efficiency of intent recognition, the large model GPT is introduced here.

[0102] This model uses a self-attention mechanism to process sequential data in parallel, improving computational efficiency. During the pre-training phase, GPT uses unsupervised learning to learn a language model by predicting the next word. During the fine-tuning phase, GPT can be adjusted to specific task requirements, such as text classification and named entity recognition.

[0103] GPT is typically pre-trained using large datasets, such as text data from the internet. After cleaning and pre-processing, this data is used to train the GPT model. During training, GPT learns a wealth of linguistic knowledge and contextual information. To adapt to specific application scenarios, developers can fine-tune GPT to better perform specific tasks.

[0104] For example, "Turn on lights, set the air conditioner to comfort mode, 26 degrees, and play music" will be identified as two main intent categories: device control intent and music skill intent. Device control intent is parsed into three intents: turning on lights, setting the air conditioner mode and temperature, and each intent must include the Media Access Control address (MAC) of the device being controlled. Music skill intent is parsed into the play music intent, which also includes the MAC address of the device playing music.

[0105] For device control intent, execute steps S305 - 1 to S305 - 3 .

[0106] Step S305-1, sending the intent to be processed to the device control service, for example, the intent to be processed may include turning on the lights, air conditioning comfort mode, 26 degrees;

[0107] Step S305-2, the device control service controls the light and the air conditioner to execute the pending intention;

[0108] Step S305-3, returning the multi-device multi-intent control result.

[0109] In other words, the biggest changes in the device control service between multi-device multi-intent control and single-device single-intent control are first the format and data of NLP parsing, and secondly the execution process in the device control service.

[0110] In the control of a single device and a single intent, the result of the intent parsing service technology is that the device category and intent can be directly converted into device instructions for execution by the device control service. In this solution, the intent parsing service technology parses a fixed intent for multi-device and multi-intent device control, namely action:multiIntent. When the device control service recognizes that the value of action is equal to multiIntent during execution, it directly enters the multi-device and multi-intent process.

[0111] The multi-device, multi-intent control process includes three iterations of device control (i.e., control versions), which are executed serially, waiting for the synchronization result of the final control result. In contrast, a single-device, single-intent control process only selects the corresponding version from the three iterations of device control and outputs the asynchronous result of the device execution, that is, whether the issued device command was actually executed successfully.

[0112] For example, to turn on a light, a baseboard command with onOffStatus=true is issued to the light with device ID VM2024070414264794 based on data analyzed using natural language processing (NLP). Because the light category has been upgraded to support Device Control 3.0, this version is selected for execution. However, after issuing this device command, the asynchronous execution result is not processed. Instead, the next step, an air conditioner with device ID 84267A7EC2D9, is executed in a loop. This air conditioner uses Device Control 2.0, setting the mode and temperature, respectively. Two baseboard commands, {operationMode=0, temperature=26}, are issued to the baseboard to control the air conditioner. This version of the device control process does not wait for asynchronous control result messages, but instead outputs a synchronous result indicating the success of the network request for the command. When there are no pending intents or commands, the synchronization results of all previous device controls are combined, ultimately returning a single, multi-device, multi-intent synchronization result to the scheduling service.

[0113] When issuing commands to control devices, regardless of the device control process version, a status check is performed before issuing or executing the command. This can also be understood as conflict resolution. For example, when setting the air conditioner to cooling mode, the system verifies whether the electric auxiliary heat is on before executing the command. If it is on, the electric auxiliary heat is turned off first, then the cooling is turned on. For example, when turning on the air conditioner, the system verifies whether the air conditioner is turned off. If not, the power-on command will not be issued, and a prompt will be output saying "the air conditioner is already on."

[0114] For skill-based intent, execute steps S306-1 to S306-2.

[0115] Step S306-1: The skill-related intent may be playing music or checking the weather, and the intent to be processed is sent to the corresponding execution device;

[0116] Step S306-2: The execution device broadcasts the result corresponding to the skill intent.

[0117] For example, voice intents such as playing music, checking the weather, and looking up recipes (i.e., skill-based intents) are assigned to skill services by the dispatch service. The skill service then changes the message processing method for each intent and synchronously returns the execution results (the content to be broadcast to the user) to the dispatch service.

[0118] For example, when playing music, the network request URL for the playlist and audio resource is directly returned; when querying the weather, the text content of the city weather is returned; when querying a recipe, the text content of the recipe is also returned. This text content is directly returned to the scheduling service.

[0119] Step S307: Perform speech processing on the execution results of multiple intentions and multiple devices, and send the processed results to the user (i.e. Figure 3 voice reply result).

[0120] The scheduling service integrates the device control service, skill service, and intent parsing context. It then organizes and optimizes the final broadcast content, prioritizing the synchronized results from the device control service, followed by text content like weather and recipes, and finally audio resources like music.

[0121] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus the necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods of each embodiment of the present application.

[0122] Figure 4 is a structural block diagram of a multi-intention control device based on a large model according to an embodiment of the present application; Figure 4 As shown, including:

[0123] A recognition module 42 is configured to perform intent recognition on the voice instruction corresponding to the target object to determine multiple intentions to be executed indicated by the voice instruction;

[0124] An input module 44 is configured to input each intent to be executed into a preset large language model, so that the preset large language model extracts keywords corresponding to each intent to be executed, and identifies the intent type corresponding to each intent to be executed based on the keywords corresponding to each intent to be executed, wherein the preset large language model is generated based on generalized preset keywords and preset intent types corresponding to the generalized preset keywords, and the intent types include: device control intent and skill intent;

[0125] The determination module 46 is used to determine the conversion method of each intention to be executed according to the intention type, and convert each intention to be executed into each target instruction according to the conversion method, so that each target device to be executed executes each target instruction, wherein the target instruction includes at least one of the following: control instructions and skill instructions.

[0126] Through the above-mentioned device, the voice instruction corresponding to the target object is recognized to determine multiple intentions to be executed indicated by the voice instruction; each intention to be executed is input into a preset large language model generated based on the generalized preset keywords and the preset intention types corresponding to the generalized preset keywords, so that the preset large language model identifies the intention type corresponding to each intention to be executed according to the keywords corresponding to each intention to be executed, wherein the intention types include: device control type intentions and skill type intentions; the conversion method of each intention to be executed is determined according to the intention type, and each intention to be executed is converted into each target instruction according to the corresponding conversion method, so that each target device to be executed executes each target instruction, wherein the target instruction may be a control type instruction and / or a skill type instruction. That is, according to the embodiment of the present application, the keywords corresponding to each intention to be executed can be extracted based on the preset large language model, and then the intention type corresponding to each intention to be executed can be identified according to the keywords corresponding to each intention to be executed. Furthermore, different conversion methods corresponding to the intention can be determined according to different intention types, and the intentions to be executed of different types can be converted into target instructions according to different conversion methods. Through the embodiment of the present application, the problem in the related art that each intention in multiple intentions cannot be accurately converted into a corresponding control instruction can be solved. Then, by determining the intention type of the intention to be executed and determining different conversion methods, each intention in the multiple intentions is accurately converted into a corresponding control instruction.

[0127] In an exemplary embodiment, the determination module 46 is also used to determine the device information of multiple first devices to be executed when the multiple intentions to be executed include device control intentions, and determine the conversion methods corresponding to the multiple device control intentions based on the device information, and determine the sending strategy for each control instruction corresponding to each device control intention, wherein the target device to be executed includes: one or more first devices to be executed; converting each device control intention into each control instruction according to the conversion method, and sending each control instruction to each first device to be executed according to the sending strategy, so that the first device to be executed executes the control instruction.

[0128] In an exemplary embodiment, the determination module 46 is also used to determine the first control version supported by each first to-be-executed device separately when the device information indicates that the multiple first to-be-executed devices are not the same to-be-executed device, wherein the instruction format and device control protocol of the target instructions corresponding to different control versions are different; convert each to-be-executed intention into each first control class instruction according to each first control version, wherein each first control class instruction is an instruction corresponding to each first control version, and the control class instructions include: each first control class instruction; sending each first control class instruction to each first to-be-executed device separately, so that each first to-be-executed device executes each first control class instruction.

[0129] In an exemplary embodiment, the determination module 46 is also used to determine the second control version supported by the first device to be executed when the device information indicates that the multiple first devices to be executed are the same device to be executed; convert each device control class intention into each second control class instruction according to the second control version, wherein each second control class instruction is a control instruction corresponding to the second control version, and the control class instruction includes: the second control class instruction; determine the first priority of each second control class instruction, and send each second control class instruction to the first device to be executed in sequence according to the first priority, so that the first device to be executed executes each second control class instruction.

[0130] In an exemplary embodiment, the determination module 46 is also used to determine the second device to be executed corresponding to the skill-type intention when the multiple intentions to be executed include a skill-type intention, and convert the skill-type intention into a skill-type instruction; send the skill-type instruction to the second device to be executed, so that the second device to be executed displays the first execution result corresponding to the skill-type instruction on the target screen corresponding to the second device to be executed, and / or, voice broadcast the first execution result.

[0131] In an exemplary embodiment, the input module 44 is further used to perform an intention conflict check on each of the intentions to be executed based on the preset large language model to determine whether there is a conflicting intention among the multiple intentions to be executed, wherein the conflicting intention is an intention in which the execution results corresponding to the multiple intentions are conflicting, and / or an intention in which the execution result corresponding to the intention is consistent with the current execution state of the target device to be executed; in the case of determining that there are multiple first conflicting intentions in which the execution results corresponding to the multiple intentions are conflicting among the multiple intentions to be executed, the historical execution record of the target device to be executed is obtained through the preset large language model, and the historical intention set corresponding to the target device to be executed in the current time period is obtained according to the historical execution record, wherein the historical execution record at least includes: the historical intention corresponding to each historical instruction executed by the target device to be executed; each first conflicting intention is tested for similarity with each intention in the historical intention set, And modify the multiple first conflicting intentions according to the similarity detection result; perform a secondary intention conflict check on the modified multiple first conflicting intentions based on the preset large language model to determine whether there is a conflicting intention among the modified multiple first conflicting intentions; when it is determined that there is a conflicting intention among the modified multiple first conflicting intentions, determine the execution frequency of the historical instructions corresponding to each modified first conflicting intention according to the historical execution record, and delete the first conflicting intention with an execution frequency less than the preset execution frequency among the modified multiple first conflicting intentions according to the execution frequency, so that there is no conflicting intention among the multiple intentions to be executed; when there is any conflicting intention among the multiple intentions to be executed, the execution result corresponding to the intention is consistent with the current execution state of the target device to be executed, delete the any conflicting intention through the preset large language model, so that there is no conflicting intention among the multiple intentions to be executed.

[0132] In an exemplary embodiment, the determination module 46 is also used to obtain each second execution result corresponding to each target device to be executed, and determine the second priority corresponding to each second execution result; integrate multiple second execution results into a message list according to the second priority, and display the message list on the screen in the control device, wherein the second execution results in the message list are arranged based on the second priority.

[0133] An embodiment of the present application further provides a storage medium, which includes a stored program, wherein the program executes any of the above methods when it is run.

[0134] Optionally, in this embodiment, the storage medium may be configured to store program codes for executing the following steps:

[0135] S1, performing intent recognition on the voice command corresponding to the target object to determine multiple intentions to be executed indicated by the voice command;

[0136] S2. Input each intention to be executed into a preset large language model, so that the preset large language model extracts keywords corresponding to each intention to be executed, and identifies the intent type corresponding to each intention to be executed based on the keywords corresponding to each intention to be executed, wherein the preset large language model is a model generated based on generalized preset keywords and preset intent types corresponding to the generalized preset keywords, and the intent types include: device control intent and skill intent;

[0137] S3, determining the conversion method of each intention to be executed according to the intention type, and converting each intention to be executed into each target instruction according to the conversion method, so that each target device to be executed executes each target instruction, wherein the target instruction includes at least one of the following: control type instruction, skill type instruction.

[0138] An embodiment of the present application further provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0139] Optionally, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0140] Optionally, in this embodiment, the processor may be configured to execute the following steps through a computer program:

[0141] S1, performing intent recognition on the voice command corresponding to the target object to determine multiple intentions to be executed indicated by the voice command;

[0142] S2. Input each intention to be executed into a preset large language model, so that the preset large language model extracts keywords corresponding to each intention to be executed, and identifies the intent type corresponding to each intention to be executed based on the keywords corresponding to each intention to be executed, wherein the preset large language model is a model generated based on generalized preset keywords and preset intent types corresponding to the generalized preset keywords, and the intent types include: device control intent and skill intent;

[0143] S3, determining the conversion method of each intention to be executed according to the intention type, and converting each intention to be executed into each target instruction according to the conversion method, so that each target device to be executed executes each target instruction, wherein the target instruction includes at least one of the following: control type instruction, skill type instruction.

[0144] An embodiment of the present application further provides a computer program product, including a computer program, which is used by a processor to execute the steps in any of the above method embodiments.

[0145] Optionally, in this embodiment, the computer program product may be processed by a processor to execute the following steps:

[0146] S1, performing intent recognition on the voice command corresponding to the target object to determine multiple intentions to be executed indicated by the voice command;

[0147] S2. Input each intention to be executed into a preset large language model, so that the preset large language model extracts keywords corresponding to each intention to be executed, and identifies the intent type corresponding to each intention to be executed based on the keywords corresponding to each intention to be executed, wherein the preset large language model is a model generated based on generalized preset keywords and preset intent types corresponding to the generalized preset keywords, and the intent types include: device control intent and skill intent;

[0148] S3, determining the conversion method of each intention to be executed according to the intention type, and converting each intention to be executed into each target instruction according to the conversion method, so that each target device to be executed executes each target instruction, wherein the target instruction includes at least one of the following: control type instruction, skill type instruction.

[0149] Optionally, in this embodiment, the above-mentioned storage medium may include but is not limited to: a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and other media that can store program codes.

[0150] Optionally, specific examples in this embodiment may refer to the examples described in the above embodiments and optional implementation modes, and this embodiment will not be described in detail here.

[0151] Obviously, those skilled in the art should understand that the modules or steps of the present application described above can be implemented using a general-purpose computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices. Alternatively, they can be implemented using program code executable by the computing device, so that they can be stored in a storage device and executed by the computing device. In some cases, the steps shown or described can be performed in a different order than herein, or they can be made into separate integrated circuit modules, or multiple modules or steps can be made into a single integrated circuit module for implementation. Thus, the present application is not limited to any specific combination of hardware and software.

[0152] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A multi-intention control method based on a large model, characterized in that: include: Performing intent recognition on the voice command corresponding to the target object to determine multiple intentions to be executed indicated by the voice command; Inputting each intention to be executed into a preset large language model, so that the preset large language model extracts keywords corresponding to each intention to be executed, and identifies the intent type corresponding to each intention to be executed according to the keywords corresponding to each intention to be executed, wherein the preset large language model is a model generated according to generalized preset keywords and preset intent types corresponding to the generalized preset keywords, and the intent types include: device control intent and skill intent; Determine the conversion method of each intention to be executed according to the intention type, and convert each intention to be executed into each target instruction according to the conversion method, so that each target device to be executed executes each target instruction, wherein the target instruction includes at least one of the following: control type instruction, skill type instruction.

2. The multi-intention control method based on a large model according to claim 1, characterized in that: Determining a conversion method for each of the intentions to be executed according to the intention type, and converting each of the intentions to be executed into each target instruction according to the conversion method, so that each target device to be executed executes each target instruction, including: In a case where the multiple intentions to be executed include device control intentions, determining device information of multiple first devices to be executed, and determining conversion methods corresponding to the multiple device control intentions based on the device information, and determining a sending strategy for each control instruction corresponding to each device control intention, wherein the target devices to be executed include: one or more first devices to be executed; Each device control class intention is converted into each control class instruction according to the conversion method, and each control class instruction is sent to each first device to be executed according to the sending strategy, so that the first device to be executed executes the control class instruction.

3. The multi-intention control method based on a large model according to claim 2, characterized in that: Determining device information of a plurality of first to-be-executed devices, and determining conversion methods corresponding to a plurality of device control class intentions based on the device information, and determining a sending strategy for each control class instruction corresponding to each device control class intention, including: When the device information indicates that the multiple first to-be-executed devices are not the same to-be-executed device, respectively determining a first control version supported by each of the first to-be-executed devices, wherein target instructions corresponding to different control versions have different instruction formats and device control protocols; Converting each of the to-be-executed intentions into each of the first control class instructions according to each of the first control versions, wherein each of the first control class instructions is an instruction corresponding to each of the first control versions, and the control class instructions include: each of the first control class instructions; Each of the first control class instructions is sent to each of the first to-be-executed devices respectively, so that each of the first to-be-executed devices executes each of the first control class instructions.

4. The multi-intention control method based on a large model according to claim 2, characterized in that: Determining device information of a plurality of first to-be-executed devices, and determining conversion methods corresponding to a plurality of device control class intentions based on the device information, and determining a sending strategy for each control class instruction corresponding to each device control class intention, including: In a case where the device information indicates that the multiple first to-be-executed devices are the same to-be-executed device, determining a second control version supported by the first to-be-executed device; Converting each device control class intention into each second control class instruction according to the second control version, wherein each second control class instruction is a control instruction corresponding to the second control version, and the control class instructions include: the second control class instruction; Determine a first priority of each second control class instruction, and send each second control class instruction to the first to-be-executed device in sequence according to the first priority, so that the first to-be-executed device executes each second control class instruction.

5. The multi-intention control method based on a large model according to claim 1, characterized in that: Determining a conversion method for each of the intentions to be executed according to the intention type, and converting each of the intentions to be executed into each target instruction according to the conversion method, so that each target device to be executed executes each target instruction, including: In a case where the multiple intentions to be executed include a skill-type intention, determining a second to-be-executed device corresponding to the skill-type intention, and converting the skill-type intention into a skill-type instruction; The skill-type instruction is sent to the second device to be executed, so that the second device to be executed displays the first execution result corresponding to the skill-type instruction on the target screen corresponding to the second device to be executed, and / or voice broadcasts the first execution result.

6. The multi-intention control method based on a large model according to claim 1, characterized in that: After inputting each intention to be executed into the preset large language model, the method further includes: Performing an intent conflict check on each of the intentions to be executed based on the preset large language model to determine whether there are conflicting intentions among the multiple intentions to be executed, wherein the conflicting intentions are intentions for which the execution results corresponding to the multiple intentions conflict, and / or intentions for which the execution results corresponding to the intentions are consistent with the current execution state of the target device to be executed; In a case where it is determined that there are multiple first conflicting intentions among the multiple intentions to be executed, wherein the execution results corresponding to the multiple intentions conflict, obtaining a historical execution record of the target device to be executed through the preset large language model, and obtaining a historical intention set corresponding to the target device to be executed in the current time period based on the historical execution record, wherein the historical execution record at least includes: a historical intention corresponding to each historical instruction executed by the target device to be executed; Performing a similarity test on each first conflicting intention and each intention in the historical intention set, and modifying the plurality of first conflicting intentions according to the similarity test results; performing a secondary intention conflict check on the modified multiple first conflicting intentions based on the preset large language model to determine whether there is a conflicting intention among the modified multiple first conflicting intentions; In a case where it is determined that a conflicting intention exists among the modified multiple first conflicting intentions, determining the execution frequency of historical instructions corresponding to each modified first conflicting intention according to the historical execution record, and deleting the first conflicting intention among the modified multiple first conflicting intentions whose execution frequency is less than the preset execution frequency according to the execution frequency, so that no conflicting intention exists among the multiple intentions to be executed; If there is any conflicting intention among the multiple intentions to be executed, and the execution result corresponding to the intention is consistent with the current execution state of the target device to be executed, the any conflicting intention is deleted through the preset large language model so that there is no conflicting intention among the multiple intentions to be executed.

7. The multi-intention control method based on a large model according to claim 1, characterized in that: After converting each of the to-be-executed intentions into each target instruction according to the conversion manner, the method further includes: Obtain each second execution result corresponding to each target to-be-executed device, and determine a second priority corresponding to each second execution result; A plurality of second execution results are integrated into a message list according to the second priority, and the message list is displayed on a screen in the control device, wherein the second execution results in the message list are arranged based on the second priority.

8. A multi-intention control device based on a large model, characterized in that: include: A recognition module, configured to perform intent recognition on the voice command corresponding to the target object to determine multiple intentions to be executed indicated by the voice command; An input module, configured to input each intention to be executed into a preset large language model, so that the preset large language model extracts keywords corresponding to each intention to be executed, and identifies the intent type corresponding to each intention to be executed based on the keywords corresponding to each intention to be executed, wherein the preset large language model is a model generated based on generalized preset keywords and preset intent types corresponding to the generalized preset keywords, and the intent types include: device control intent and skill intent; A determination module is used to determine the conversion method of each intention to be executed according to the intention type, and convert each intention to be executed into each target instruction according to the conversion method, so that each target device to be executed executes each target instruction, wherein the target instruction includes at least one of the following: control instructions and skill instructions.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein the method according to any one of claims 1 to 7 is executed when the program is executed.

10. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to execute the method according to any one of claims 1 to 7 through the computer program.

Images