Device control method, apparatus, storage medium, and program product

Abstract

The present disclosure relates to a device control method, device, storage medium and program product. The method comprises: receiving a user instruction; determining, according to the user instruction, a candidate device conforming to a user operation intention and an operation to be performed; obtaining running state information of the candidate device and device specification information, the device specification information being used to define configurable parameters supported by the device; determining, according to the operation to be performed, the running state information and the device specification information, a target device from the candidate device; and controlling the target device to act. In this way, operation abnormalities caused by inconsistent device states or unsupported parameters can be avoided, the success rate of instruction execution is improved, and user experience is improved.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to a device control method, apparatus, storage medium, and program product. Background Technology

[0002] In intelligent control scenarios, numerous devices may be distributed within the same space. When user commands do not conform to standards, the system struggles to accurately select the target device that matches the user's intent from the candidate devices, leading to frequent operational failures or miscontrols. Summary of the Invention

[0003] To overcome the problems existing in related technologies, this disclosure provides a device control method, apparatus, storage medium, and program product.

[0004] According to a first aspect of the present disclosure, a device control method is provided, comprising: Receive user instructions; Based on the user instructions, determine the candidate devices that match the user's operational intent and the operations to be performed; Obtain the operating status information and device specification information of the candidate devices, wherein the device specification information is used to define the configurable parameters supported by the device; Based on the operation to be performed, the running status information, and the equipment specification information, the target device is determined from the candidate devices; Control the target device to operate.

[0005] In the above technical solution, candidate devices and operations to be executed are determined by parsing user commands. This is combined with the real-time operating status and specifications of the devices to accurately identify the target device from multiple candidate devices and control the action. This avoids operational anomalies caused by incompatible device status or unsupported parameters, improves the success rate of command execution, and ultimately enhances the user experience.

[0006] In some possible implementations, determining the target device from the candidate devices based on the operation to be performed, the operating status information, and the device specification information includes: Based on the configurable parameters, determine whether the candidate device has the capability to perform the operation to be performed; Based on the operating status information, candidate devices that are offline or have already met the expected state of the operation to be performed are filtered out; Candidate devices that possess the aforementioned capabilities and have not been filtered are identified as the target devices.

[0007] In the above technical solution, by using both device specifications and real-time status for dual screening, it can be ensured that the target device has both execution capability and is in an available state, thereby improving the success rate of instruction execution and thus improving the user experience.

[0008] In some possible implementations, determining the candidate device that matches the user's operational intent based on the user instruction includes: Based on the user instructions and device information in the target space, determine the candidate devices in the target space that match the user's operating intentions; The target space is any one of the following: The space where the control device that receives the user's instructions is located; The space indicated by the user instruction; The default space that is set in advance.

[0009] In the above technical solution, the range of candidate devices can be narrowed down based on the target space, avoiding traversing all devices in the house, improving processing efficiency, and reducing the probability of accidentally controlling irrelevant devices.

[0010] In some possible implementations, controlling the target device includes: The user instructions are input into the first inference model, and the first inference model outputs the standard control instructions that the target device can execute. The standard control commands are issued to control the action of the target device.

[0011] In the above technical solution, the user instructions are converted into standard control instructions through the first reasoning model. This can accurately parse the user's intentions and generate adapted instructions, reduce instruction ambiguity and execution deviation, and improve the accuracy and reliability of the target device's action control.

[0012] In some possible implementations, the method further includes: Determine whether the user instruction meets the preset nondeterministic processing conditions; The step of inputting the user command into the first inference model and having the first inference model output standard control commands that the target device can execute includes: If the nondeterministic processing conditions are met, the user instruction is input into the first inference model, and the first inference model outputs the standard control instruction.

[0013] In the above technical solution, if the preset nondeterministic processing conditions are met, it can be determined that the current instruction has uncertainties such as missing information, ambiguity, parameter out-of-bounds, or multiple targets. At this time, the first reasoning big model can improve the understanding of complex scenarios and the execution success rate.

[0014] In some possible implementations, the method further includes: If the nondeterministic processing conditions are not met, the standard control command is obtained using a pre-configured mapping rule base.

[0015] In the above technical solution, if the nondeterministic processing conditions are not met, it can be determined that the current instruction information is complete, the intent is clear, the parameters are legal, and the target is unique. At this time, by using the pre-configured mapping rule base, the user instruction can be quickly and efficiently converted into a standard control instruction and executed, reducing inference overhead and improving response speed.

[0016] In some possible implementations, the nondeterministic processing conditions include one or more of the following: The structured description information obtained by semantically parsing the user instructions is invalid; The current state of the target device is consistent with the state expected by the operation to be performed; The target device does not have the capability to perform the operation to be performed; The user instructions involve multiple operations to be executed; The number of target devices is multiple; The semantic parsing result of the user instruction is ambiguous.

[0017] In the above technical solution, by using a nondeterministic judgment mechanism, it is possible to accurately identify whether the instruction is complex, ambiguous, unclear or missing information, so as to select an appropriate method to generate standard control instructions.

[0018] In some possible implementations, the first inference big model is used to generate the standard control instructions by inference based on the user instructions, the equipment specification information, the operating status information, and the equipment information in the target room.

[0019] In the above technical solution, the first inference big model can be used to adapt to diverse user needs and equipment status, efficiently generate accurate standard control commands, and improve the accuracy and reliability of target equipment action control.

[0020] In some possible implementations, the first large inference model is trained in the following manner: Acquire first raw sample data and second raw sample data. The determinism of user instructions in the first raw sample data is higher than that in the second raw sample data. The raw sample data includes user instruction samples and corresponding standard control instruction samples. The first stage of training of the base model is performed using the first original sample data; The pre-defined domain knowledge is used as prompt words to input into the second reasoning model, which processes the second raw sample data to generate thought chain data. Using the aforementioned thought chain data, supervised fine-tuning and cold start training are performed sequentially on the model that has undergone the first stage of training; The model trained by cold start is optimized using reinforcement learning algorithms.

[0021] The above technical solutions can improve the model's processing accuracy and output reliability for complex and ambiguous user commands, and generate standard control commands more accurately and efficiently.

[0022] In some possible implementations, the first large-scale reasoning model is trained to possess at least one of the following decision-making capabilities: Candidate devices are determined from the space where the control device receiving the user instruction is located, or from the space indicated by the user instruction; Devices whose first configuration parameter involved in the user instruction belongs to the configurable parameter are identified as candidate devices; Filter out offline devices and candidate devices whose current state already satisfies the user's instruction operation intent; If the user instruction includes a complete device name or key name, then the candidate device indicated by the device name or key name is determined as the target device.

[0023] In the above technical solution, through multi-dimensional decision-making capabilities such as spatial limitation, parameter matching, state filtering and name recognition, the first reasoning big model can accurately lock the target device, thereby improving the success rate of instruction execution and the convenience of user operation.

[0024] In some possible implementations, the first large-scale reasoning model is trained to possess at least one of the following decision-making capabilities: If multiple candidate devices belong to the lighting category, the priority of each candidate device is determined according to the predefined sub-category priority table. If the name of a candidate device conflicts with the name of the room it is in, and the user instruction does not completely match the name of the device, then the priority of the candidate device is reduced. Select the candidate device with the highest priority as the target device.

[0025] In the above technical solution, the accuracy of intent matching can be improved and the reliability of control can be guaranteed by using sub-category priority sorting, naming conflict reduction and highest priority selection mechanism.

[0026] In some possible implementations, the first large-scale inference model is trained to possess the following decision-making capabilities: If the target device does not support the first configuration parameter involved in the user instruction, then a second configuration parameter with a preset similarity relationship to the first configuration parameter is determined from the device specification information, and an alternative operation corresponding to the second configuration parameter is determined.

[0027] The above technical solution can enhance the flexibility of equipment control, avoid direct rejection when the target equipment has limited capabilities, and meet user needs and improve user experience as much as possible.

[0028] In some possible implementations, obtaining the standard control instructions using a pre-configured mapping rule base includes: The user instructions are semantically parsed to generate structured description information; Based on the structured description information, the corresponding mapping rule is matched from the mapping rule library; Based on the matched mapping rules, candidate standard instructions are generated; The candidate standard instructions are verified using the equipment specification information of the target device; The candidate standard instructions that pass the verification are determined as the standard control instructions.

[0029] In the above technical solution, by mapping rules to device specifications, it can be ensured that the generated instructions conform to the user's intent and can be correctly executed by the device.

[0030] According to a second aspect of the present disclosure, a device control apparatus is provided, which is configured to perform the steps of the device control method provided in the first aspect of the present disclosure.

[0031] According to a third aspect of the present disclosure, a device control apparatus is provided, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to execute the executable instructions in the memory to implement the steps of the device control method provided in the first aspect of this disclosure.

[0032] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the device control method provided in the first aspect of the present disclosure.

[0033] According to a fifth aspect of the present disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the device control method provided in the first aspect of the present disclosure.

[0034] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0035] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0036] Figure 1 This is a flowchart illustrating a device control method according to an exemplary embodiment.

[0037] Figure 2 This is a flowchart illustrating a device control method according to an exemplary embodiment.

[0038] Figure 3 This is a block diagram illustrating a device control apparatus according to an exemplary embodiment.

[0039] Figure 4 This is a block diagram illustrating a device control apparatus according to an exemplary embodiment. Detailed Implementation

[0040] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0041] It should be noted that all actions involving the acquisition of signals, information, or data in this disclosure are performed with authorization from the owner of the relevant device.

[0042] Figure 1 This is a flowchart illustrating a device control method according to an exemplary embodiment. This device control method can be applied to smart home systems, industrial automation control systems, or IoT device management platforms. Figure 1 As shown, the device control method may include steps S101 to S105.

[0043] In step S101, a user instruction is received.

[0044] In one embodiment, the user can issue commands via a terminal device (such as a smart speaker, smartphone, or central control panel). These commands can be voice or text commands. For example, the user can say "turn on the air conditioner" or type "turn on the bedroom light."

[0045] In step S102, candidate devices that match the user's operating intentions and the operations to be performed are determined according to the user's instructions.

[0046] In one embodiment, user commands are parsed using natural language understanding or keyword matching to determine one or more candidate devices and corresponding operations to be performed. Taking the user command "turn on the living room light" as an example, candidate devices may include the main living room light and spotlights, and the operation to be performed may be "turn on".

[0047] In step S103, the operating status information and equipment specification information of the candidate equipment are obtained.

[0048] In one embodiment, the device's operating status information may include the current on / off state, online status, and operating parameters (such as the current brightness value of a lamp or the current temperature setting of an air conditioner). Thus, the operating status information can be used to determine whether a candidate device is in a state where an operation can be performed, thereby avoiding issuing commands to offline devices and avoiding repeatedly performing invalid operations on devices that have already met the desired state.

[0049] In one embodiment, device specification information (SPCE) is used to define the configurable parameters supported by the device. Configurable parameters are parameters that can be adjusted by the user, or parameters that can be adjusted by function keys identified on the corresponding control device of the device. Taking a lighting fixture as an example, the corresponding configurable parameters may include attributes such as brightness, color temperature, and color; taking an air conditioner as an example, the corresponding configurable parameters may include attributes such as temperature, operating mode (e.g., cooling / heating / air supply), fan speed, and airflow direction. Thus, by using device specification information, devices capable of performing the desired operation can be selected, thereby excluding devices that may fail due to mismatches between configurable parameters and the requirements of the desired operation.

[0050] In step S104, the target device is determined from the candidate devices based on the operation to be performed, the running status information, and the device specification information.

[0051] In one embodiment, the target device can be determined from candidate devices in the following manner: Based on configurable parameters, determine whether the candidate device has the capability to perform the operation to be performed; Based on the operating status information, filter out candidate devices that are offline or have already met the expected status of the operation to be performed; Candidate devices that possess this capability and have not been filtered are identified as target devices.

[0052] For example, if the first configuration parameter involved in the user instruction is a configurable parameter defined in the corresponding device specification information, it can be determined that the candidate device has the ability to perform the operation to be performed; otherwise, it can be determined that the candidate device does not have the ability to perform the operation to be performed. As another example, if the first configuration parameter involved in the user instruction is a configurable parameter defined in the device specification information, and the expected value of the parameter involved in the user instruction is within the corresponding allowed value range, it can be determined that the candidate device has the ability to perform the operation to be performed; otherwise, it can be determined that the candidate device does not have the ability to perform the operation to be performed.

[0053] For example, if a user command requests that the light color be adjusted to yellow: if the configurable parameters of the main living room light include a color attribute, and that attribute supports yellow as a valid value, then it can be determined that the main living room light has this capability; if the configurable parameters of the living room spotlights do not include a color attribute, or if the color attribute does not support yellow in its color list, then it can be determined that the living room spotlights do not have this capability.

[0054] For example, the user command is "turn on the living room lights," and candidate devices include the main living room light, spotlights, and LED strip lights. The system retrieves the operating status information of each device. If the LED strip light is currently offline (e.g., due to a network disconnection or power failure), it is filtered out from the candidate devices. This avoids issuing invalid commands to uncontrollable devices.

[0055] For example, the user command is "turn off the main living room light," and the desired state for the operation to be performed is "off." The system retrieves the operating status information of the main living room light. If the current switch status is "off," the device is determined to have met the desired state and is filtered out. This avoids repeatedly executing invalid commands and reduces unnecessary control overhead.

[0056] For example, the user command is "turn on the living room light," and candidate devices include the main living room light and living room spotlights. Based on configurable parameters, it is determined that both support the turn-on operation; based on the operating status information, if the living room spotlights are offline, they are filtered out; finally, the main living room light, which is capable of being turned on, online, and currently off, is identified as the target device, and thus the main living room light is controlled to turn on.

[0057] By using both device specifications and real-time status for dual screening, we can ensure that the target device is both capable of execution and in an available state, thereby improving the success rate of command execution and thus enhancing the user experience.

[0058] In step S105, the target device is controlled to operate.

[0059] In one embodiment, a standard control command is sent to the target device. This standard control command can be generated based on the target device and the operation to be performed. The standard control command is an executable command conforming to the target device's interface protocol. The executable command includes a unique identifier for the target device, a target configuration parameter identifier corresponding to the operation to be performed, and a target value for the target configuration parameter. If the target device executes the command successfully, a success response can be returned; if the target device fails to execute the command successfully, the user can be prompted that the operation failed.

[0060] In the above technical solution, candidate devices and operations to be executed are determined by parsing user commands. This is combined with the real-time operating status and specifications of the devices to accurately identify the target device from multiple candidate devices and control the action. This avoids operational anomalies caused by incompatible device status or unsupported parameters, improves the success rate of command execution, and ultimately enhances the user experience.

[0061] In some possible implementations, in step S102, determining a candidate device that matches the user's operational intent based on the user's instruction includes: Based on user instructions and device information in the target space, candidate devices in the target space that match the user's operational intent are identified.

[0062] For example, the target space is the living room, and the devices in the living room include the main light, spotlights, a television, and an air conditioner. The user command is "turn on the light." Semantic analysis can determine that candidate devices may include the main light and spotlights. In this way, the range of candidate devices can be narrowed down based on the target space, avoiding a complete search of all devices in the house, improving processing efficiency, and reducing the probability of accidentally controlling irrelevant devices.

[0063] In one embodiment, the target space is the space where the control device receiving the user's command is located. For example, if a user issues the command "turn on the lights" through a smart speaker in the living room, the target space is the living room. In this way, the user does not need to specify a space; the system can filter devices based on their location, improving ease of operation and optimizing the user experience.

[0064] In one embodiment, the target space is the space indicated by the user's command. For example, if a user issues the command "Turn on the air conditioner in the bedroom," explicitly mentioning "bedroom," then the target space is the bedroom. This allows for precise response to the user's operational needs for a specific space, avoiding accidental control of devices in other spaces.

[0065] In one embodiment, the target space is a pre-defined default space. For example, if the user does not specify a space and the control device cannot locate it, the pre-defined default space can be used as the target space. The user can pre-set the default space through an APP (Application). In this way, a fallback strategy can be provided when instruction information is insufficient, quickly narrowing down the scope and ensuring the efficiency of device screening.

[0066] In some possible implementations, in step S102, determining a candidate device that matches the user's operational intent based on the user's instruction includes: Based on user instructions and device information in the target space, determine candidate devices in the target space that match the user's operational intent; Based on configurable parameters, determine whether the candidate device has the capability to perform the operation to be performed; Based on the operating status information, filter out candidate devices that are offline or have already met the expected status of the operation to be performed; Candidate devices that possess the above capabilities and have not been filtered are identified as target devices.

[0067] In this way, by narrowing the candidate range through spatial constraints and combining dual screening of equipment specifications and real-time status, it can be ensured that the target equipment is both capable of execution and in an available state, thereby improving the success rate of instruction execution and avoiding invalid operations and miscontrols.

[0068] In some possible implementations, step S105, controlling the target device to operate, includes: User commands are input into the first inference model, and the first inference model outputs standard control commands that the target device can execute. Issue standard control commands to control the actions of the target equipment.

[0069] In one embodiment, the first large-scale inference model can be built based on a large language model (LLM). This model can be trained using machine learning, and the input information to the first large-scale inference model can include user commands. The output information of the first large-scale inference model can be standard control commands. The first large-scale inference model can be stored locally on the electronic device and invoked locally each time it is used, or it can be stored on a third-party platform and invoked from the third party each time it is used; no specific limitation is made here.

[0070] In this way, by converting user commands into standard control commands through the first reasoning model, user intentions can be accurately parsed and adapted commands can be generated, reducing command ambiguity and execution deviations, and improving the accuracy and reliability of target device action control.

[0071] In one embodiment, the first inference big model is used to generate standard control commands through inference based on user instructions, equipment specification information, operating status information, and equipment information in the target room.

[0072] For example, if the user command is "brighten the bedroom light", the first inference model obtains the device information in the bedroom, such as the bedroom includes a light and an air conditioner. Further, it obtains the device specification information and corresponding operating status information of the bedroom light. If the current brightness of the bedroom light is 40% and its device specification information includes brightness as a configurable parameter, that is, it supports brightness adjustment, the standard control command "increase the brightness of the bedroom light by 10%" can be generated through inference.

[0073] In this way, the first inference model can be used to adapt to diverse user needs and equipment status, efficiently generate accurate standard control commands, and improve the accuracy and reliability of target equipment action control.

[0074] In one embodiment, the first large inference model is trained in the following manner: Acquire first raw sample data and second raw sample data, wherein the determinism of user instructions in the first raw sample data is higher than that in the second raw sample data, and the raw sample data includes user instruction samples and corresponding standard control instruction samples. The first stage of training for the pedestal model is performed using the first set of original sample data; The pre-defined domain knowledge is used as prompts to input into the second reasoning model, which processes the second raw sample data to generate thought chain data. Using the thought chain data, supervised fine-tuning and cold start training are performed sequentially on the model that has undergone the first stage of training; The model trained by cold start is optimized using reinforcement learning algorithms.

[0075] For example, the first set of raw sample data can be generated using a rule-driven, batch synthesis approach. For instance, it involves extracting the specifications of each device, including device name, category, supported attributes (such as on / off, brightness, color temperature, and temperature), and the value range of each attribute (such as brightness 0-100%, temperature 16-30℃). Simultaneously, a natural language template library is pre-built, containing phrases such as "Open <device name>" and "Set the brightness of <device name> to <value>%". Then, the device specification information is populated into the templates, automatically generating a massive number of user command samples and simultaneously generating precisely corresponding standard control commands, forming the first set of raw sample data corresponding to a simple scenario.

[0076] For example, logs of user commands that are difficult to process and occur in real-world online scenarios can be collected, such as complex cases involving multiple device operations and ambiguous expressions. These complex commands, along with background knowledge such as device lists, spatial layout information, and business strategies, can be input into a large language model as context. The model is then required to infer the correct standard control commands based on the given information, and multiple optimizations can be performed to improve the annotation quality. Multiple large model instances can be used to independently annotate the same sample. If the results are consistent and match the actual online execution results, this sample is identified as the second set of original sample data. Difficult samples with inconsistent results are manually reviewed. In this way, high-quality second set of original sample data can be obtained for complex scenarios.

[0077] For example, user commands from the first set of raw sample data are used as input, and the corresponding standard control commands are used as the target output to train the base model until the first stage of training ends. In this way, the first stage of training allows the model to initially learn the basic mapping ability from natural language commands to standard control commands.

[0078] For example, pre-defined domain knowledge, such as business strategies, can be input into the second reasoning model in the form of prompts, along with user instructions from the second original sample data, to obtain thought chain data. Each piece of data may include user instructions, reasoning steps, and corresponding standard control instructions.

[0079] For example, using the thought chain data as a training set, the model trained in the first stage is subjected to supervised fine-tuning, enabling it to learn to output reasoning processes and standard control commands. A cold start training is then performed to enhance its reasoning ability.

[0080] For example, GRPO or DAPO reinforcement learning algorithms can be used to iteratively optimize the model after a cold start. A reward function can be set, and reward items can include: correct instruction format, existence of the device ID in the target space, whether the parameter values ​​are within the specification range, and whether the output is completely consistent with the standard answer. After multiple rounds of iterative optimization, the model can achieve an optimal balance in terms of output accuracy, format standardization, and consistency of inference logic, thereby improving the model's ability to handle complex user instructions.

[0081] In the above technical solution, basic capabilities are learned through first raw sample data with higher certainty, and then the logical reasoning of the model is enhanced by thinking chain data obtained based on second raw sample data. Then, reinforcement learning algorithms are used to further optimize the model. In this way, the model can improve the processing accuracy and output reliability of complex and ambiguous user commands, and generate standard control commands more accurately and efficiently.

[0082] For example, the first large-scale inference model was trained to have the following decision-making capabilities: Candidate devices are determined from the space where the control device receiving the user instruction is located, or from the space indicated by the user instruction.

[0083] For example, if a user issues the command "turn on the lights" through a smart speaker in the living room, the target space is determined to be the living room; if the user's command is "turn on the bedroom lights," the target space is determined to be the bedroom. In this way, by limiting the range of candidate devices, cross-space mis-controls can be reduced and response accuracy can be improved.

[0084] For example, the first large-scale inference model was trained to have the following decision-making capabilities: Devices whose first configuration parameter involved in the user command is a configurable parameter are identified as candidate devices.

[0085] For example, if the user command is "adjust the brightness to 80%", devices that support brightness adjustment can be filtered out as candidate devices. In this way, devices that do not support the required parameters can be quickly eliminated, narrowing down the range of candidate devices.

[0086] For example, the first large-scale inference model was trained to have the following decision-making capabilities: Filter out offline devices and candidate devices whose current state already meets the user's command operation intent.

[0087] For example, if the user command is "turn off the living room light," and the main living room light is already off, then that command will be filtered out. This avoids invalid operations.

[0088] For example, the first large-scale inference model was trained to have the following decision-making capabilities: If the user instruction includes a complete device name or key name, the candidate device indicated by the device name or key name will be identified as the target device.

[0089] For example, if the user command is "Turn on the main living room light," which includes the complete device name "main living room light," then the main living room light will be directly identified as the target device. As another example, if the user command is "Turn on the small light," which includes the complete button name "small light" on the switch, then the device corresponding to the "small light" button (such as a bedside nightlight) will be directly identified as the target device. This allows for quick response to precise user commands, thereby improving operational efficiency.

[0090] For example, the first large-scale inference model was trained to have the following decision-making capabilities: If multiple candidate devices belong to the lighting category, the priority of each candidate device is determined according to the predefined sub-category priority table.

[0091] For example, in a predefined sub-category priority table, primary lighting equipment has a higher priority than alternative lighting equipment, and alternative lighting equipment has a higher priority than auxiliary lighting equipment. Primary lighting equipment can include ceiling lights, chandeliers, etc.; alternative lighting equipment can include wall lights, spotlights, etc.; and auxiliary lighting equipment can include bedside lamps, table lamps, etc. When the user command is "turn on the lights," the primary lighting equipment with the highest priority can be selected as the candidate target according to this priority table. In this way, when faced with multiple candidate lighting equipment, the optimal candidate target can be quickly and accurately determined, improving the efficiency and accuracy of equipment control and avoiding blind operation.

[0092] For example, the first large-scale inference model was trained to have the following decision-making capabilities: If the name of a candidate device conflicts with the name of the room it is in, and the user command does not fully match the device name, the priority of the candidate device will be reduced.

[0093] For example, if a user command is "turn on the bedroom light," and there is a candidate device in the bedroom named "living room light," the priority of that candidate device can be lowered. This avoids mis-controls caused by inconsistent naming.

[0094] For example, the first large-scale inference model was trained to have the following decision-making capabilities: Select the candidate device with the highest priority as the target device.

[0095] For example, candidate devices can be prioritized as follows: main light > spotlights > light strips, with the main light selected as the target device. This avoids control errors caused by chaotic device selection, meeting basic user needs while preventing resource waste.

[0096] For example, the first large-scale inference model was trained to have the following decision-making capabilities: If the target device does not support the first configuration parameter involved in the user command, then a second configuration parameter with a preset similarity to the first configuration parameter is determined from the device specification information, and an alternative operation corresponding to the second configuration parameter is determined.

[0097] For example, if the user command is "mute the air conditioner," but the air conditioner (target device) does not have volume control capabilities (meaning its specifications do not include the volume attribute specified by the user), a second configuration parameter similar to mute, such as fan speed, can be identified from the air conditioner's specifications. An alternative operation corresponding to this second configuration parameter can then be determined; for example, setting the fan speed to level 1 can reduce the air conditioner's operating noise. The preset similarity relationships can be pre-defined based on actual needs.

[0098] This enhances the flexibility of equipment control, avoids direct rejection when the target equipment has limited capabilities, and strives to meet user needs and improve the user experience.

[0099] For example, the first large-scale inference model was trained to have the following decision-making capabilities: If all identified target devices are located within the same area, then control the actions of each target device.

[0100] For example, if the user command is "turn off the lights," and the analysis determines that the target devices include the main living room light, spotlights, and LED strip lights, all of which are located within the living room area, then all of these devices can be controlled to turn off without requiring individual user confirmation. This enables rapid batch control, improving operational efficiency and user experience.

[0101] For example, the first large-scale inference model was trained to have the following decision-making capabilities: If multiple target devices are identified but not in the same area, a selection query is sent to the user.

[0102] For example, if the user command is "turn off the lights," the parsing process identifies the target devices as the main living room light, the main bedroom light, and the study lamp, located in three different areas: the living room, bedroom, and study. At this point, it's difficult to determine whether the user intends to turn off all the lights in the house or only the lights in their current area. Therefore, a selection query can be posed to the user: "Do you want to turn off all the lights in the house, or only the lights in the living room?" Once the user clarifies their choice, the corresponding control action can be executed.

[0103] In this way, cross-regional miscontrol can be avoided, and users can be guided to supplement their intent information by proactively asking questions, thereby improving the user-friendliness of the interaction and the accuracy of decision-making.

[0104] Figure 2 This is a flowchart illustrating a device control method according to an exemplary embodiment. Figure 2 As shown, the device control method provided in this disclosure may include steps S106 to S109.

[0105] In step S106, it is determined whether the user instruction meets the preset nondeterministic processing conditions. If yes, step S107 is executed; otherwise, step S108 is executed.

[0106] If the preset nondeterministic processing conditions are met, it can be determined that the current instruction has uncertainties such as missing information, ambiguity, parameter out-of-bounds, or multiple targets. At this time, step S107 is executed, and the first inference big model can improve the understanding of complex scenarios and the success rate of execution. If the nondeterministic processing conditions are not met, it can be determined that the current instruction information is complete, the intent is clear, the parameters are legal, and the target is unique. At this time, step S108 is executed, and the user instruction can be quickly and efficiently converted into a standard control instruction and executed using the pre-configured mapping rule base, reducing inference overhead and improving response speed.

[0107] In one embodiment, the nondeterministic processing condition includes: the structured description information obtained by semantic parsing the user instruction is invalid.

[0108] For example, a user command is "Turn on the living room light." After semantic parsing, the resulting structured description is {Device Code: x1; Device Type: Light, Space: Living Room, Operation: Turn On}. This structure is complete, and all field values ​​are valid. If the parsing result is abnormal, such as x1 not being declared or bound to a specific device, then the structured description information is invalid, satisfying the non-deterministic processing condition. Thus, by judging the validity, it is possible to identify whether the parsed result possesses the basic information structure for execution, thereby distinguishing between deterministic and non-deterministic commands. This allows the use of the first inference model to handle commands with ambiguous information, thereby improving the success rate of command execution.

[0109] In one embodiment, the nondeterministic processing condition includes: the current state of the target device is consistent with the state expected by the operation to be performed.

[0110] For example, if the user command is "adjust the bedroom air conditioner temperature to 26℃", and the current temperature of the target device, the bedroom air conditioner, is 24℃, which is inconsistent with the desired 26℃, then the nondeterministic processing condition is not met. If the current temperature is already 26℃, then the state is consistent, and the nondeterministic processing condition is met.

[0111] In one embodiment, the nondeterministic processing condition includes: the target device does not have the capability to perform the operation to be performed.

[0112] For example, the first configuration parameter involved in the user instruction is not a configurable parameter defined in the corresponding device specification information. For instance, if the user instruction is "adjust the bedside lamp light to white", and the bedside lamp specification information does not contain a color attribute, meaning the bedside lamp does not support color adjustment, then this nondeterministic processing condition is met.

[0113] For example, the first configuration parameter involved in the user instruction is a configurable parameter defined in the device specification information, but the expected value of the parameter involved in the user instruction does not fall within the corresponding allowed value range. For instance, the user instruction is "set the air conditioner temperature to 15℃". The allowed temperature range in the air conditioner specification information is 18℃ to 30℃. 15℃ is outside this range, so it meets the nondeterministic processing condition.

[0114] In this way, if the device's capabilities determine that an instruction is difficult to execute, an alternative solution can be implemented using the first inference model to meet user needs and improve the user experience as much as possible.

[0115] In one embodiment, the nondeterministic processing condition includes: the user instruction involves multiple operations to be performed.

[0116] For example, if the user instruction is "turn on the living room light and turn off the bedroom air conditioner," this instruction contains two different operations to be executed: "turn on the living room light" and "turn off the bedroom air conditioner." In this case, the nondeterministic processing condition is met. If the instruction is only "turn on the living room light," which contains only a single operation, then this condition is not met. Thus, complex tasks can be guided to the first inference model for task decomposition, ensuring that each operation is executed reasonably and orderly, avoiding omissions or conflicts.

[0117] In one embodiment, the nondeterministic processing condition includes: the number of target devices is multiple.

[0118] For example, if the user command is "turn on the lights," and there are four target devices in the living room—a main light, spotlights, LED strips, and wall lights—then the nondeterministic processing condition is met. However, if the command is "turn on the living room main light," and there is only one target device, then the condition is not met. Thus, the first inference model can be used to clearly define the target device, avoiding execution ambiguity and miscontrol.

[0119] In one embodiment, the nondeterministic processing condition includes: the semantic parsing result of the user instruction is ambiguous.

[0120] For example, if the user command is "turn on that light," where "that" is ambiguous and lacks a clear spatial or device reference, the semantic parsing result is ambiguous, thus satisfying the nondeterministic processing condition. If the command is "turn on the main living room light," the semantics are clear and unambiguous, then this condition is not satisfied. For another example, if the user command is "turn on the living room light," there are two different execution paths: path one turns on a specific device named "living room light" (e.g., a lamp that the user has custom-named "living room light"); path two turns on multiple lights in the "living room" (e.g., the main living room light, living room spotlights). In this case, "living room light" can refer to a specific aliased device or all the lights in the living room space; the semantic parsing results in multiple path ambiguities, thus satisfying the nondeterministic processing condition.

[0121] When there is ambiguity in the semantic parsing results of user commands, the first reasoning model can be used in conjunction with dialogue history (such as the devices the user operated in the previous round) and the user's commonly used device habits to perform disambiguation reasoning, thereby improving the accuracy of command understanding.

[0122] In step S107, user instructions are input into the first inference big model, and the first inference big model outputs standard control instructions.

[0123] The specific implementation of step S107 has been described in detail in the previous embodiments and will not be repeated here.

[0124] In step S108, standard control commands are obtained using a pre-configured mapping rule base.

[0125] In one embodiment, step S108 is implemented in the following manner: Perform semantic parsing on user commands to generate structured description information; Based on the structured description information, match the corresponding mapping rules from the mapping rule base; Based on the matched mapping rules, candidate standard instructions are generated; The candidate standard instructions are verified using the equipment specification information of the target device; Candidate standard instructions that pass the verification are determined as standard control instructions.

[0126] For example, the user command is semantically parsed to generate a structured description. For instance, if the user command is "turn on the air conditioner's ventilation mode," the parsed result is structured information: the candidate device is an air conditioner, and the operation to be performed is to turn on the ventilation mode. Based on this structured information, the corresponding mapping rule is matched from the mapping rule base. Matching conditions include action type (e.g., turning on) and device attributes (e.g., ventilation mode). If multiple rules exist for matching, the rule with the highest priority is selected.

[0127] The matched rules contain a pre-configured list of multiple candidate parameter configurations (i.e., targets), which are used to attempt to generate standard instructions that can be executed by the device in sequence. For example, the first candidate configuration of this rule is to set the mode parameter to ventilation mode; the second candidate configuration is to set the ventilation parameter to the on state.

[0128] First, try the first candidate configuration to generate a candidate standard command: the action is "operation," and the mode parameter is set to ventilation mode. Then, use the target device's specifications to verify whether the device supports the mode setting function, whether a ventilation mode option exists, and whether the option is within a valid value range. If the verification passes, the candidate command is determined as the standard control command.

[0129] If the first configuration verification fails (e.g., the device does not support the mode parameter), the second candidate configuration is tried, generating a candidate standard command: the action is "operation," the ventilation parameter is set to "on," and the device is verified to support the parameter and its value. If the second candidate configuration verification passes, the candidate command is determined as the standard control command. If all candidate configurations fail verification, an unsupported message is returned.

[0130] In this way, by mapping rules to device specifications, it can be ensured that the generated instructions conform to the user's intent and can be correctly executed by the device.

[0131] In step S109, standard control commands are issued to control the action of the target device.

[0132] exist Figure 2 The technical solution presented utilizes a non-deterministic judgment mechanism to guide complex, ambiguous, unclear, or information-deficient instructions through a first-level reasoning model, while clear and unambiguous instructions are quickly mapped into a rule base. This not only improves instruction comprehension and execution success rates in complex scenarios but also ensures efficient response in simpler scenarios.

[0133] In some possible implementations, the device control method provided in this disclosure further includes: If the user command does not belong to the preset command library, then through semantic parsing, candidate devices that match the user's operation intention and the operation to be executed are determined; If the user command belongs to the preset command library, the corresponding target device is determined according to the preset first mapping relationship.

[0134] For example, the preset instruction library can store high-frequency, deterministic instructions and their corresponding target device mapping relationships. For instance, "turn off all lights" is mapped to all types of lights in the house. When the user instruction is "turn off all lights," it can be determined that the instruction belongs to the preset instruction library. The corresponding target device and operation sequence can be determined directly based on the first mapping relationship, and the instruction can be executed quickly.

[0135] In this way, for high-frequency or patterned instructions in the preset instruction library, complex semantic parsing and candidate device screening can be skipped to achieve rapid response and improve execution efficiency; for non-preset instructions, real-time semantic parsing is used for flexible processing to ensure execution accuracy.

[0136] In some possible implementations, the device control method provided in this disclosure further includes: Obtain the action results of the target device; Generate natural language feedback based on the results of actions.

[0137] In one embodiment, if the user instruction belongs to a preset instruction library, the natural language feedback corresponding to the action result is determined according to the preset second mapping relationship; if the user instruction does not belong to the preset instruction library, the action result and the user instruction are input into the speech generation model, and the speech generation model outputs natural language feedback.

[0138] For example, if the user command is "Turn off all lights" (which belongs to the preset command library), after execution, all lights in the house will be successfully turned off. Based on the feedback script corresponding to "Turn off all lights successfully" in the second mapping relationship, the natural language feedback "All lights in the house have been turned off for you" can be generated.

[0139] For example, if the user command is "adjust the living room spotlights to cool light" (not part of the preset command library), and the user successfully adjusts the color temperature of the living room spotlights to cool light mode based on the user command, then the action result (the color temperature of the living room spotlights has been adjusted to cool light) and the user command can be input into the dialogue generation model. The model will then generate the natural language feedback "The living room spotlights have been adjusted to cool light." If the execution fails (e.g., the spotlights do not support color temperature adjustment), then the failure result is input, and the model generates the feedback "Sorry, the current living room spotlights do not support color temperature adjustment and cannot be adjusted to cool light." In this way, by differentiating between preset and non-preset commands and adopting different feedback generation strategies, the efficiency of feedback in high-frequency scenarios is ensured, while the model utilizes flexible and natural interactive dialogue to improve the user experience.

[0140] In another embodiment, regardless of whether the user instruction belongs to the preset instruction library, the action result and the user instruction can be input into the speech generation model, and the speech generation model outputs natural language feedback to achieve unified generation and flexible expression of feedback speech.

[0141] Figure 3 This is a block diagram illustrating a device control apparatus according to an exemplary embodiment. (Refer to...) Figure 3 The equipment control device 300 includes: Receiver module 301 is used to receive user instructions; The first determining module 302 is used to determine, based on the user instruction, a candidate device that matches the user's operating intention and an operation to be performed. The acquisition module 303 is used to acquire the operating status information and device specification information of the candidate device, wherein the device specification information is used to define the configurable parameters supported by the device; The second determining module 304 is used to determine the target device from the candidate devices based on the operation to be executed, the running status information and the device specification information; The control module 305 is used to control the operation of the target device.

[0142] In the above technical solution, candidate devices and operations to be executed are determined by parsing user commands. This is combined with the real-time operating status and specifications of the devices to accurately identify the target device from multiple candidate devices and control the action. This avoids operational anomalies caused by incompatible device status or unsupported parameters, improves the success rate of command execution, and ultimately enhances the user experience.

[0143] In some possible implementations, the second determining module 304 is used to determine the target device from the candidate devices in the following manner: Based on the configurable parameters, determine whether the candidate device has the capability to perform the operation to be performed; Based on the operating status information, candidate devices that are offline or have already met the expected state of the operation to be performed are filtered out; Candidate devices that possess the aforementioned capabilities and have not been filtered are identified as the target devices.

[0144] In some possible implementations, the first determining module 302 is used to determine candidate devices that conform to the user's operational intent by: Based on the user instructions and device information in the target space, determine the candidate devices in the target space that match the user's operating intentions; The target space is any one of the following: The space where the control device that receives the user's instructions is located; The space indicated by the user instruction; The default space that is set in advance.

[0145] In some possible implementations, the control module 305 includes: The first inference submodule is used to input the user instructions into the first inference big model, and the first inference big model outputs the standard control instructions that the target device can execute. The issuing submodule is used to issue the standard control commands to control the action of the target device.

[0146] In some possible implementations, the control module 305 further includes: The judgment submodule is used to determine whether the user instruction meets the preset nondeterministic processing conditions; The first inference submodule is further configured to, if the nondeterministic processing conditions are met, input the user instruction into the first inference big model, and have the first inference big model output the standard control instruction.

[0147] In some possible implementations, the control module 305 further includes: The second reasoning submodule is used to obtain the standard control command by utilizing a pre-configured mapping rule base if the nondeterministic processing conditions are not met.

[0148] In some possible implementations, the nondeterministic processing conditions include one or more of the following: The structured description information obtained by semantically parsing the user instructions is invalid; The current state of the target device is consistent with the state expected by the operation to be performed; The target device does not have the capability to perform the operation to be performed; The user instructions involve multiple operations to be executed; The number of target devices is multiple; The semantic parsing result of the user instruction is ambiguous.

[0149] In some possible implementations, the first inference big model is used to generate the standard control instructions by inference based on the user instructions, the equipment specification information, the operating status information, and the equipment information in the target room.

[0150] In some possible implementations, the first large inference model is trained in the following manner: Acquire first raw sample data and second raw sample data. The determinism of user instructions in the first raw sample data is higher than that in the second raw sample data. The raw sample data includes user instruction samples and corresponding standard control instruction samples. The first stage of training of the base model is performed using the first original sample data; The pre-defined domain knowledge is used as prompt words to input into the second reasoning model, which processes the second raw sample data to generate thought chain data. Using the aforementioned thought chain data, supervised fine-tuning and cold start training are performed sequentially on the model that has undergone the first stage of training; The model trained by cold start is optimized using reinforcement learning algorithms.

[0151] In some possible implementations, the first large-scale reasoning model is trained to possess at least one of the following decision-making capabilities: Candidate devices are determined from the space where the control device receiving the user instruction is located, or from the space indicated by the user instruction; Devices whose first configuration parameter involved in the user instruction belongs to the configurable parameter are identified as candidate devices; Filter out offline devices and candidate devices whose current state already satisfies the user's instruction operation intent; If the user instruction includes a complete device name or key name, then the candidate device indicated by the device name or key name is determined as the target device.

[0152] In some possible implementations, the first large-scale reasoning model is trained to possess at least one of the following decision-making capabilities: If multiple candidate devices belong to the lighting category, the priority of each candidate device is determined according to the predefined sub-category priority table. If the name of a candidate device conflicts with the name of the room it is in, and the user instruction does not completely match the name of the device, then the priority of the candidate device is reduced. Select the candidate device with the highest priority as the target device.

[0153] In some possible implementations, the first large-scale inference model is trained to possess the following decision-making capabilities: If the target device does not support the first configuration parameter involved in the user instruction, then a second configuration parameter with a preset similarity relationship to the first configuration parameter is determined from the device specification information, and an alternative operation corresponding to the second configuration parameter is determined.

[0154] In some possible implementations, the second inference submodule is used to obtain the standard control commands in the following manner: The user instructions are semantically parsed to generate structured description information; Based on the structured description information, the corresponding mapping rule is matched from the mapping rule library; Based on the matched mapping rules, candidate standard instructions are generated; The candidate standard instructions are verified using the equipment specification information of the target device; The candidate standard instructions that pass the verification are determined as the standard control instructions.

[0155] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0156] In another exemplary embodiment, this disclosure also provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the steps of the device control method provided in this disclosure.

[0157] In another exemplary embodiment, this disclosure also provides a computer program product comprising a computer program executable by a programmable device, the computer program having a code portion for performing the device control method described above when executed by the programmable device.

[0158] Figure 4 This is a block diagram illustrating a device control apparatus 1900 according to an exemplary embodiment. For example, apparatus 1900 may be provided as a server. (Refer to...) Figure 4 The device 1900 includes a processing component 1922, which further includes one or more processors, and memory resources represented by memory 1932 for storing instructions, such as application programs, that can be executed by the processing component 1922. The application programs stored in memory 1932 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 1922 is configured to execute instructions to perform device control methods.

[0159] Device 1900 may also include a power supply component 1926 configured to perform power management of device 1900, a wired or wireless network interface 1950 configured to connect device 1900 to a network, and an input / output interface 1958. Device 1900 can operate on an operating system stored in memory 1932.

[0160] Those skilled in the art will also understand that the various illustrative logical blocks and steps listed in the embodiments of this application can be implemented by electronic hardware, computer software, or a combination of both. Whether such functionality is implemented through hardware or software depends on the specific application and the overall system design requirements. Those skilled in the art can implement the described functionality using various methods for each specific application, but such implementation should not be construed as exceeding the scope of protection of the embodiments of this application.

[0161] It should be understood that, unless otherwise specifically indicated, features of various embodiments of this disclosure described herein can be combined with each other. As used herein, the term “and / or” includes any one of the relevant listed items and any combination of any two or more; similarly, “at least one of…” includes any one of the relevant listed items and any combination of any two or more.

[0162] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In this description, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0163] Furthermore, the term “exemplary” is used herein to mean serving as an example, instance, or illustration. Any aspect or design described herein as “exemplary” is not necessarily to be construed as advantageous compared to other aspects or designs. Rather, the use of the term “exemplary” is intended to present the concept in a concrete manner. As used herein, the term “or” is intended to mean an inclusive “or” rather than an exclusive “or.” That is, unless otherwise specified or clear from the context, “X applies A or B” is intended to mean any of the natural inclusive arrangements. That is, “X applies A or B” satisfies any of the foregoing instances if X applies A; X applies B; or both X applies A and B. Additionally, unless otherwise specified or clear from the context to refer to the singular form, the articles “a” and “an” as used in this application and the appended claims are generally understood to mean “one or more.”

[0164] Similarly, although this disclosure has been shown and described with respect to one or more implementations, equivalent variations and modifications will occur to those skilled in the art upon reading and understanding this specification and the accompanying drawings. This disclosure includes all such modifications and variations and is limited only by the scope of the claims. In particular, with respect to the various functions performed by the components described above (e.g., elements, resources, etc.), unless otherwise indicated, the terminology used to describe such components is intended to correspond to any component (functionally equivalent) that performs the specific function of the described component, even if structurally not equivalent to the disclosed structure. Furthermore, although specific features of this disclosure may have been disclosed with respect to only one of several implementations, such features may be combined with one or more other features of other implementations, as may be desired and advantageous to any given or particular application. Moreover, with regard to the terms “comprising,” “owning,” “having,” “having,” or variations thereof as used in the detailed description or claims, such terms are intended to be inclusive in a manner similar to the term “including.”

[0165] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the appended claims.

[0166] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A device control method characterized by, include: Receive user instructions; Based on the user instructions, determine the candidate devices that match the user's operational intent and the operations to be performed; Obtain the operating status information and device specification information of the candidate devices, wherein the device specification information is used to define the configurable parameters supported by the device; Based on the operation to be performed, the running status information, and the equipment specification information, the target device is determined from the candidate devices; Control the target device to operate.

2. The method of claim 1, wherein, The step of determining the target device from the candidate devices based on the operation to be executed, the operating status information, and the device specification information includes: Based on the configurable parameters, determine whether the candidate device has the capability to perform the operation to be performed; Based on the operating status information, candidate devices that are offline or have already met the expected state of the operation to be performed are filtered out; Candidate devices that possess the aforementioned capabilities and have not been filtered are identified as the target devices.

3. The method of claim 1, wherein, The step of determining candidate devices that match the user's operational intent based on the user instruction includes: Based on the user instructions and device information in the target space, determine the candidate devices in the target space that match the user's operating intentions; The target space is any one of the following: The space where the control device that receives the user's instructions is located; The space indicated by the user instruction; The default space that is set in advance.

4. The method of claim 1, wherein, The control of the target device includes: The user instructions are input into the first inference model, and the first inference model outputs the standard control instructions that the target device can execute. The standard control commands are issued to control the action of the target device.

5. The method of claim 4, wherein, The method further includes: Determine whether the user instruction meets the preset nondeterministic processing conditions; The step of inputting the user command into the first inference model and having the first inference model output standard control commands that the target device can execute includes: If the nondeterministic processing conditions are met, the user instruction is input into the first inference model, and the first inference model outputs the standard control instruction.

6. The method of claim 5, wherein, The method further includes: If the nondeterministic processing conditions are not met, the standard control command is obtained using a pre-configured mapping rule base.

7. The method of claim 5, wherein, The nondeterministic processing conditions include one or more of the following: The structured description information obtained by semantically parsing the user instructions is invalid; The current state of the target device is consistent with the state expected by the operation to be performed; The target device does not have the capability to perform the operation to be performed; The user instructions involve multiple operations to be executed; The number of target devices is multiple; The semantic parsing result of the user instruction is ambiguous.

8. The method of claim 4, wherein, The first inference model is used to generate the standard control instructions based on the user instructions, the equipment specification information, the operating status information, and the equipment information in the target room.

9. The method of claim 4, wherein, The first large-scale inference model was trained in the following way: Acquire first raw sample data and second raw sample data. The determinism of user instructions in the first raw sample data is higher than that in the second raw sample data. The raw sample data includes user instruction samples and corresponding standard control instruction samples. The first stage of training of the base model is performed using the first original sample data; The pre-defined domain knowledge is used as prompt words to input into the second reasoning model, which processes the second raw sample data to generate thought chain data. Using the aforementioned thought chain data, supervised fine-tuning and cold start training are performed sequentially on the model that has undergone the first stage of training; The model trained by cold start is optimized using reinforcement learning algorithms.

10. The method according to claim 4, characterized in that, The first large-scale reasoning model was trained to possess at least one of the following decision-making abilities: Candidate devices are determined from the space where the control device receiving the user instruction is located, or from the space indicated by the user instruction; Devices whose first configuration parameter involved in the user instruction belongs to the configurable parameter are identified as candidate devices; Filter out offline devices and candidate devices whose current state already satisfies the user's instruction operation intent; If the user instruction includes a complete device name or key name, then the candidate device indicated by the device name or key name is determined as the target device.

11. The method of claim 4, wherein, The first large-scale reasoning model was trained to possess at least one of the following decision-making abilities: If multiple candidate devices belong to the lighting category, the priority of each candidate device is determined according to the predefined sub-category priority table. If the name of a candidate device conflicts with the name of the room it is in, and the user instruction does not completely match the name of the device, then the priority of the candidate device is reduced. Select the candidate device with the highest priority as the target device.

12. The method of claim 4, wherein, The first large-scale reasoning model was trained to have the following decision-making capabilities: If the target device does not support the first configuration parameter involved in the user instruction, then a second configuration parameter with a preset similarity relationship to the first configuration parameter is determined from the device specification information, and an alternative operation corresponding to the second configuration parameter is determined.

13. The method according to claim 6, characterized in that, The process of obtaining the standard control commands using a pre-configured mapping rule base includes: The user instructions are semantically parsed to generate structured description information; Based on the structured description information, the corresponding mapping rule is matched from the mapping rule library; Based on the matched mapping rules, candidate standard instructions are generated; The candidate standard instructions are verified using the equipment specification information of the target device; The candidate standard instructions that pass the verification are determined as the standard control instructions.

14. An apparatus control device characterized by comprising: The device control apparatus is configured to perform the device control method according to any one of claims 1-13.

15. An apparatus control device characterized by comprising: include: processor; Memory used to store processor-executable instructions; The processor is configured to execute the executable instructions in the memory to implement the steps of the device control method according to any one of claims 1-13.

16. A computer readable storage medium having stored thereon a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the device control method according to any one of claims 1-13.

17. A computer program product, characterized in that, A computer program including a computer program that, when executed by a processor, implements the steps of the device control method of any one of claims 1-13.

Images