Device control method, apparatus and system, device, and medium

Abstract

The present disclosure relates to a device control method, apparatus and system, a device, and a medium. The method comprises: acquiring target voice information; determining current input information on the basis of the target voice information; matching the current input information with a preset target device control database to obtain a matching result; if the matching result indicates that the matching is successful, controlling a target device on the basis of a first device control task corresponding to successfully matched target input information; and if the matching result indicates that the matching fails, inputting the current input information into a device control model to generate a second device control task, and controlling the target device on the basis of the second device control task. According to the present disclosure, calculation costs are greatly reduced, the response speed is increased, and internal information and external information of a device are added into input information, increasing influence factors considered during device control, so that the intent of a user can be understood more accurately, thus improving the accuracy of subsequent device control.

Description

Device control method, device, system, device and medium

[0001] This disclosure claims priority to the Chinese patent application filed with the China Patent Office on July 20, 2023, with application number 202310899189.1 and invention name “A device control method, apparatus, system, equipment and medium”, the entire contents of which are incorporated by reference into this disclosure. Technical Field

[0002] The present disclosure relates to the field of computer technology, and in particular to a device control method, apparatus, system, device, and medium. Background Art

[0003] With the development of natural language technology, the control of equipment is becoming more and more intelligent and humanized. For example, the control of home appliances is becoming more and more intelligent.

[0004] In related technologies, device control is typically achieved by training deep learning models. However, this approach requires large amounts of data and is highly complex, resulting in high costs and slow response times. Furthermore, it often considers fewer influencing factors, leading to low control accuracy. In summary, related technologies struggle to achieve a satisfactory balance of cost, response speed, and control accuracy in device control.

[0005] Summary of the Invention

[0006] (1) Technical issues to be solved

[0007] The technical problem to be solved by the present disclosure is to solve the problem that the control of existing equipment is difficult to achieve good effects in terms of cost, response speed and control accuracy.

[0008] (2) Technical solution

[0009] In order to solve the above technical problems or at least partially solve the above technical problems, the present disclosure provides a device control method, apparatus, system, device and medium.

[0010] A first aspect of an embodiment of the present disclosure provides a device control method, the method comprising:

[0011] Obtain target voice information;

[0012] determining current input information based on the target voice information;

[0013] Matching the current input information in a preset target device control database to obtain a matching result;

[0014] If the matching result is a successful match, controlling the target device based on the first device control task corresponding to the successfully matched target input information;

[0015] If the matching result is a matching failure, the current input information is input into a device control model to generate a second device control task, and the target device is controlled based on the second device control task.

[0016] A second aspect of the embodiments of the present disclosure further provides a device control apparatus, the apparatus comprising:

[0017] An acquisition module is used to acquire target voice information;

[0018] A determination module, configured to determine current input information based on the target voice information;

[0019] A matching module, configured to match the current input information in a preset target device control database to obtain a matching result;

[0020] a first control module configured to control the target device based on a first device control task corresponding to the successfully matched target input information if the matching result is a successful match;

[0021] The second control module is configured to input the current input information into a device control model to generate a second device control task if the matching result is a matching failure, and control the target device based on the second device control task.

[0022] The third aspect of the embodiment of the present disclosure also provides a device control system, including a device control device, a cache space including a target device control database, and a task module provided with a device control model. The device control device is connected to the cache space and the task module for executing the device control method provided in the embodiment of the present disclosure.

[0023] The fourth aspect of the embodiments of the present disclosure also provides an electronic device, which includes: a processor; a memory for storing executable instructions of the processor; the processor is used to read the executable instructions from the memory and execute the instructions to implement the device control method provided in the embodiments of the present disclosure.

[0024] A fifth aspect of the embodiments of the present disclosure further provides a computer-readable storage medium, wherein the storage medium stores a computer program, and the computer program is used to execute the device control method provided by the embodiments of the present disclosure. (3) Beneficial effects

[0025] The technical solution provided by the embodiments of the present disclosure has the following advantages over the prior art:

[0026] The device control solution provided by the embodiment of the present disclosure obtains target voice information; determines current input information based on the target voice information; matches the current input information in a preset target device control database to obtain a matching result; if the matching result is a successful match, the target device is controlled based on the first device control task corresponding to the successfully matched target input information; if the matching result is a failed match, the current input information is input into the device control model to generate a second device control task, and the target device is controlled based on the second device control task. Using the above technical solution, by constructing a target device control database, after obtaining voice information and determining the corresponding current input information, the target input information and device control task corresponding to the current input information can be matched in the target device control database first. When the match fails, the current input information is input into the device control model to generate a device control task in real time, avoiding repeated calculation of the device control task corresponding to the repeated voice, greatly reducing the computing cost, and improving the response speed. In addition, other information is added to the input information determined based on the voice information, and the influencing factors considered when controlling the device are increased, which can more accurately understand the user's intention and thereby improve the accuracy of subsequent device control.

[0027] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0029] In order to more clearly illustrate the embodiments of the present disclosure 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.

[0030] FIG1 is a schematic flow chart of a device control method provided by an embodiment of the present disclosure;

[0031] FIG2 is a schematic diagram of a device control provided by an embodiment of the present disclosure;

[0032] FIG3 is a schematic diagram of another device control provided by an embodiment of the present disclosure;

[0033] FIG4 is a schematic diagram of a multi-round control provided by an embodiment of the present disclosure;

[0034] FIG5 is a schematic diagram of another device control method provided by an embodiment of the present disclosure;

[0035] FIG6 is a schematic diagram of a conversation diagram provided by an embodiment of the present disclosure;

[0036] FIG7 is a schematic structural diagram of a device control system provided by an embodiment of the present disclosure;

[0037] FIG8 is a schematic structural diagram of a device control apparatus provided by an embodiment of the present disclosure;

[0038] FIG9 is a schematic structural diagram of an electronic device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0039] To make the objectives, technical solutions, and advantages of the embodiments of the present disclosure more clear, the technical solutions in the embodiments of the present disclosure are clearly and completely described below. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments of the present disclosure, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present disclosure.

[0040] In the related art, device control is usually achieved by training a deep learning model. For example, since the Large Language Model (LLM) can understand and generate natural language text and has shown good results in many application scenarios, this model can be used for device control. However, this method has the following disadvantages: 1. Slow response speed. LLM is a deep learning model based on neural networks. It has a huge number of parameters and very high computational complexity. When processing device control requests, it is necessary to process and decode the input text, which leads to relatively slow processing speed, often taking several seconds to complete. Therefore, when real-time interaction is required, this delay may make users dissatisfied; 2. High cost. Due to the high computational complexity of LLM, running such a model requires high computing resources. If a service provided by an external third party is called, the cost of calling the service will be particularly high as the number of controls increases. Therefore, in many application scenarios, this may lead to very high energy consumption and costs; 3. Few influencing factors are considered during control, resulting in low control accuracy. In summary, the control of devices in the related art is difficult to achieve good results in terms of cost, response speed and control accuracy.

[0041] In order to solve the above problems, an embodiment of the present disclosure provides a device control method, which is introduced below in conjunction with specific embodiments.

[0042] FIG1 is a flow chart of a device control method provided by an embodiment of the present disclosure. The method can be executed by a device control device, wherein the device can be implemented using software and / or hardware and can generally be integrated into an electronic device. As shown in FIG1 , the method includes:

[0043] Step 101: Acquire target voice information.

[0044] Among them, the target voice information can be the voice audio emitted by the user when he currently needs to control a device, which can be collected by a voice collection device, and the specific device to be controlled is not limited, for example, it can be an air conditioner, a television, etc. The embodiment of this disclosure takes the control of a single household appliance as an example.

[0045] Step 102: Determine current input information based on the target voice information.

[0046] Among them, the input information can be information used to determine subsequent device control tasks, and can be a collection of information of multiple dimensions. The current input information can be input information determined based on the current target voice information. In the embodiment of the present disclosure, the current input information can include the target text corresponding to the target voice information and the device information of the target device corresponding to the target voice information. The device information includes internal information and external information. The target text can be the text obtained by voice recognition of the target voice information; the target device can be the device that currently needs to be controlled based on the target voice information; the device information can include both internal information and external information. The internal information can be information such as the status and functional parameters of the target device. For example, when the target device is a refrigeration air conditioner, the internal information can include switch status information, target temperature, air volume, etc. The external information can be information about the environment where the target device is located, such as temperature, humidity, weather, light intensity, carbon dioxide concentration, etc.

[0047] In some embodiments, determining the current input information based on the target voice information may include: converting the target voice information into target text and determining the target device corresponding to the target text; obtaining the device information of the target device, and combining the target text and the device information of the target device to determine the current input information.

[0048] After acquiring the target voice information, the device control device can convert the target voice information into target text through voice recognition, for example, by using Automatic Speech Recognition (ASR) technology. The device control device can then determine the target device to be controlled based on the target text, and then acquire the internal and external information of the target device, and determine the combination of the target text, internal and external information as the current input information. For example, when the target text is "turn on the air conditioner," the internal information of the target device may include the air conditioner switch status being off, the target temperature being 25°C, the air volume being medium, and the mode being cooling. The external information may include the temperature being 30°C.

[0049] In some embodiments, obtaining device information of the target device may include: retrieving internal information of the target device; and obtaining external information of the target device based on the device type of the target device, the geographical location of the target device, sensors of the associated device, and / or user settings.

[0050] When the device control device obtains the device information of the target device, it can send a call instruction to the target device for the internal information therein so that the target device returns the internal information; for the external information therein, it can obtain the corresponding external information according to the device type of the target device. Different device types can correspond to different external information. For example, when the target device is an air conditioner, the external information is temperature; when the target device is a smart curtain, the external information can be light intensity; it can also obtain external information from the Internet based on the geographical location of the target device; it can also obtain external information through the sensors of the associated devices of the target device. The associated devices can be other devices with the same geographical location as the target device, and the specific number is not limited; it can also obtain the external information of the target device of the user device.

[0051] In the above scheme, the internal information of the target device can be obtained through call, and the acquisition of external information can be achieved in a variety of ways, which improves the efficiency and reliability of information acquisition and helps to improve the accuracy of subsequent home appliance control.

[0052] Step 103: Match the current input information in a preset target device control database to obtain a matching result.

[0053] Among them, the target device control database can be a database newly added in the embodiment of the present disclosure that stores multiple standard dialogue control information. Each standard dialogue control information includes a standard input information and a corresponding device control task, that is, the target device control database stores a one-to-one mapping relationship between multiple standard input information and multiple device control tasks. The standard input information has the same data structure as the above-mentioned current input information. The device control task can be the specific task information for executing device control. Specifically, the device control task can be sent to the device to generate an electronic control instruction to control the target device to perform a specific operation. The target device control database can be stored in the cache space, which can reduce the resource and cost consumption caused by calling a large model and improve response efficiency.

[0054] In some embodiments, matching the current input information in a preset device control database to obtain a matching result may include: semantically matching the current input information with the standard input information of multiple standard dialogue control information included in the target device control database one by one to determine the matching degree; if there is standard input information with a matching degree greater than a matching degree threshold, then determining the matching result as a successful match, and determining the standard input information as the target input information with a successful match; otherwise, determining the matching result as a failed match.

[0055] The matching degree threshold may be a minimum threshold for a semantic matching device, and may be set specifically according to actual conditions. In the embodiment of the present disclosure, the matching degree threshold may be set larger, and when the matching degree is greater than the matching degree threshold, it may be considered a complete match.

[0056] After determining the current input information, the device control device can semantically match the current input information with the standard input information of each standard dialogue control information in the target device control database and determine the matching degree. It can then judge whether each matching degree is greater than the matching degree threshold. If there is a standard input information with a matching degree greater than the matching degree threshold with the current input information, the matching result is determined to be a successful match, and the standard input information is determined to be the target input information that has been successfully matched; if each matching degree is less than the matching degree threshold, the matching result can be determined to be a failed match.

[0057] In the above scheme, when there are at least two standard input information whose matching degree with the current input information is greater than the matching degree threshold, the standard input information with the highest matching degree can be determined as the target input information with successful matching, or the at least two standard input information can be fed back to the user to determine the standard input information selected by the user as the target input information with successful matching, or the standard input information selected by the largest number of other users can be determined as the target input information with successful matching.

[0058] Step 104 , determine whether the matching result is successful. If so, execute step 105 ; otherwise, execute step 106 .

[0059] Step 105: Control the target device based on the first device control task corresponding to the successfully matched target input information.

[0060] The first device control task may be a device control task corresponding to target input information determined according to the target device control data.

[0061] After the device control device determines that the match result is successful, it can obtain the first device control task corresponding to the successfully matched target input information and send the first device control task to the target device to control the target device to perform the relevant operation. For example, if the target text of the target voice information is "Turn on the air conditioner", the first device control task can be an instruction to turn on the air conditioner. This first device control task is sent to the air conditioner to turn it on.

[0062] Optionally, corresponding reply information may be stored for each input message in each standard dialogue control message in the target device control database. This reply information may be generated based on the input information and the execution result of the device control task. After determining the first device control task, first reply information corresponding to the successfully matched target input information may be obtained and sent to the target device, so that the target device can feedback the first reply information to the user. Specifically, this first reply information may be displayed on the screen or synthesized and played through speech synthesis.

[0063] Step 106: Input the current input information into the device control model to generate a second device control task, and control the target device based on the second device control task.

[0064] The device control model may be a pre-trained or general model that analyzes and calculates the current input information in real time to generate the corresponding device control task. The device control model may be a deep learning model, such as an LLM. The second device control task may be the device control task corresponding to the current input information determined based on the device control model.

[0065] After the device control device determines that the matching result is a match failure, it indicates that a corresponding device control task needs to be generated in real time. The current input information can be input into the device control model for real-time analysis to obtain a corresponding second device control task, and the second device control task is sent to the target device to control the target device to perform the relevant operation. Optionally, the device control device can also generate corresponding second reply information based on the input information and the execution result of the second device control task, and send the second reply information to the target device, so that the target device feeds back the first reply information to the user. Specifically, the first reply information can be displayed on the screen or the first reply information can be synthesized and played through voice.

[0066] Exemplarily, Figure 2 is a schematic diagram of a device control provided by an embodiment of the present disclosure. As shown in Figure 2, the figure shows the specific process of device control, including: when a new target voice information enters the system, the target voice information and device information are obtained to generate current input information, and the current input information is first matched in the target device control database. If the match is successful, the device control task corresponding to the successfully matched target input information can be extracted as the first device control task, and the target device is controlled to perform the corresponding operation through the first device control task; if the match fails, the current input information is input into the device control model for processing to generate a second device control task, and the target device is controlled to perform the corresponding operation through the second device control task.

[0067] For example, FIG3 is a schematic diagram of another device control provided by an embodiment of the present disclosure. As shown in FIG3 , the diagram illustrates the data flow between modules of the device control apparatus of an embodiment of the present disclosure when performing device control. The device control apparatus may include the semantic representation generation module (Embedding Generator), database management module (Cache Manager), semantic matching module (Semantic Match), task execution module (Task Executor), and task planning module (LLM planner) shown in the diagram.

[0068] The semantic representation module is used to calculate the semantic representation vector of the user's input information. The input includes input text and device information, and the output is the corresponding semantic representation vector. The input text comes from the user's text input or the text converted from the user's voice information using voice technology.

[0069] The database management module includes a cache and a vector database (vector DB). These two databases together constitute the target device control database. The vector database stores semantic representation vectors (embeddings) of multiple input information. The database also stores task information for the device control tasks corresponding to each input information. This information can be stored in JSON format and can be, for example, a SQL database. Each semantic representation vector of input information corresponds to the task information for a device control task. The database management module maintains and updates the vector database and database, updating the database and vector data with device control tasks and input information generated in real time after a match failure.

[0070] The semantic matching module is used to calculate the matching degree between the semantic representation vector corresponding to the input information and each semantic representation vector in the vector database. When there is a matching degree greater than the matching degree threshold, the matching is successful, otherwise, the matching fails.

[0071] The task planning module is used to generate a high-level device control task consisting of a sequence of subtasks based on the current input information. It also retrieves low-level device control tasks from a knowledge base through a retrieval mechanism. This module generates device control tasks corresponding to the input information using the device control model. Device control tasks include both high-level and low-level device control tasks. This module includes the device control model, which can be composed of a prompt generator, a high-level task planning module (LLM Planner), and a low-level task planner. The prompt generator generates prompts based on the input information. The generated prompts contain four components: task and goal descriptions, in-context learning examples, internal and external information from the device information, and current state information (including text from the input information, available devices, action information, and the execution status of the device control task). The context learning examples, available devices, and action information are retrieved from a cache or a pre-built knowledge base through a retrieval mechanism. The internal information, external information and execution status information of the device control task in the device information are obtained through the interaction between the user and the environment and other sensors.

[0072] The task execution module is used to call the corresponding task execution mechanism to implement the device control task according to the device control task matched in the target device control database or obtained through the device control model.

[0073] In addition, the device control apparatus may further include a response generation module (Response Generator), which is used to generate a user response based on input information and the task execution result of the device control task, and return it to the user or the voice module for voice synthesis and broadcasting.

[0074] The specific process may include: the input information in the figure includes text input from the user or text converted from the user's voice information using voice technology, as well as the device information of the device to be controlled; the input information is generated into a corresponding semantic vector representation by the semantic representation generation module; the semantic vector representation is then sent to the database management module, and semantically matched with all input information pointing to other nodes under the current node in the vector database of the database management module through the semantic matching module. If the match is successful, the successful node (i.e., the successfully matched target input information) is directly used as the new dialogue state, and the device control task in this state (i.e., the target device control task corresponding to the successfully matched target input information) is passed as the input of the task execution module; otherwise, the input information is input into the task planning module for task planning, and the corresponding device control task is obtained. The output of the task planning module is a task representation that can be executed by the task execution module, i.e., the device control task. The task execution module completes the task execution, i.e., controls the target device to perform related operations based on the device control task, and returns the execution status information. The task execution status information can then be input into the reply generation module to generate reply information, which is returned to the user, completing the dialogue interaction with the user and simultaneously updating the database of the database management module to achieve automatic or semi-automatic data update.

[0075] The device control method of the embodiment of the present disclosure can cover single-round and multi-round control scenarios of the device. For example, Figure 4 is a schematic diagram of a multi-round control provided by the embodiment of the present disclosure. As shown in Figure 4, the target device is an air conditioner as an example, and the text corresponding to the user's multi-round control voice information is displayed. The direction of the arrow in the figure indicates the control sequence. The control sequence is: the room is too hot, turn up the wind, sweep the wind left and right, and switch to fresh air mode; the room is too hot, adjust the temperature to 25 degrees, it is still too hot, and adjust the temperature lower; the room is too hot, adjust the temperature to 25 degrees, and adjust the temperature higher. These are just examples.

[0076] The device control solution provided by the embodiment of the present disclosure obtains target voice information; determines current input information based on the target voice information; matches the current input information in a preset target device control database to obtain a matching result; if the matching result is a successful match, the target device is controlled based on the first device control task corresponding to the successfully matched target input information; if the matching result is a failed match, the current input information is input into the device control model to generate a second device control task, and the target device is controlled based on the second device control task. Using the above technical solution, by constructing a target device control database, after obtaining voice information and determining the corresponding current input information, the target input information and device control task corresponding to the current input information can be matched in the target device control database first. When the match fails, the current input information is input into the device control model to generate a device control task in real time, avoiding repeated calculation of the device control task corresponding to the repeated voice, greatly reducing the computing cost, and improving the response speed. In addition, the input information determined based on the voice information adds internal and external device information, increasing the influencing factors considered when controlling the device, and can more accurately understand the user's intention, thereby improving the accuracy of subsequent device control.

[0077] In some embodiments, the device control method of the embodiment of the present disclosure may further include: constructing a target device control database. For example, FIG5 is a schematic diagram of another device control method provided by the embodiment of the present disclosure. As shown in FIG5 , in a feasible implementation, constructing the target device control database may include the following steps:

[0078] Step 501: Build an initial device control database.

[0079] Among them, the initial device control database can be understood as a device control database that has not undergone data verification and may not be very accurate. The initial device control database may include multiple initial dialogue control information, and each initial dialogue control information may include initial input information and corresponding device control tasks.

[0080] In some embodiments, constructing an initial device control database may include: obtaining multiple historical dialogue control information, where each historical dialogue control information includes historical input information and a corresponding device control task; and constructing the device control database based on the multiple historical dialogue control information. Optionally, constructing the device control database based on the multiple historical dialogue control information may include: filtering the multiple historical dialogue control information based on device usage time and / or information quantity to obtain multiple initial dialogue control information; and storing the multiple initial dialogue control information in a database to obtain the initial device control database.

[0081] Among them, the historical dialogue control information can be a large amount of historical input information and device control task information corresponding to the device control generated by the actual user or test user at historical time. It is control information in the actual scenario. The specific quantity is not limited. The historical input information has the same data structure as the above-mentioned current input information.

[0082] Specifically, the device control device can obtain multiple historical conversation control information from log data, then filter the multiple historical conversation control information based on device usage time and / or information quantity to obtain multiple initial conversation control information, and store them in a database to obtain an initial device control database. Optionally, the initial device control database can be stored in the form of a conversation graph. Specifically, multiple nodes can be constructed based on the device control tasks in each initial conversation control information, and directed edges between each node can be constructed based on the initial input information in each initial conversation control information. The conversation graph corresponding to the initial device control database can include multiple nodes and directed edges, each node representing a device control task in the initial conversation control information, and each directed edge storing the initial input information in the initial conversation control information, indicating the direction of the node. Optionally, each initial conversation control information can also include response information corresponding to the input information. In addition to the aforementioned device control tasks, each node can also include the response information.

[0083] Optionally, multiple historical dialogue control information are filtered based on the device usage time and / or the amount of information, including: deleting historical dialogue control information in which the amount of information is less than a quantity threshold and the electronic control instructions corresponding to the device control task do not belong to the preset instructions; and / or deleting historical dialogue control information in which the time point is outside the device usage time range.

[0084] Device usage time can be the specific time a device is used, for example, an air conditioner is used in the summer, while a heater is used in the winter. Message quantity can be the total number of historical control messages from the same dialog, and the quantity threshold can be a minimum value set for the message quantity. An electronic control instruction can be a binary instruction generated by inputting a device control task into the electronic control platform. This electronic control instruction can be considered an instruction to control the device. Preset instructions can be instructions set for a device to implement a corresponding function. Preset instructions can vary for different devices.

[0085] Specifically, when filtering multiple historical dialogue control messages, the device control device can determine the number of messages in each of the multiple historical dialogue control messages, and determine the number of historical dialogue control messages that meet a threshold number of messages. Based on these historical dialogue control messages, the device control task is input into the electronic control platform to obtain electronic control instructions. The device control device then determines whether each electronic control instruction is a preset instruction. If so, the historical dialogue control information may be infrequent but typical, such as an air conditioner self-cleaning instruction, and can be retained. If not, the historical dialogue control information appears infrequently and is not typical, and the historical dialogue control information is deleted. Furthermore, the device control device can determine whether the time point in each historical dialogue control message falls outside the device usage time range. If the time point in a historical dialogue control message falls outside the device usage time range, the message can be deleted. The above strategy for filtering historical dialogue control information is merely an example and can be implemented using other applicable methods.

[0086] In the above scheme, when constructing the initial device control database, the historical dialogue control information can be filtered or screened according to a certain filtering strategy before creating the database, which helps to improve the matching hit rate of the initial device control database in subsequent use, thereby improving the efficiency of device control.

[0087] Step 502: Perform data verification on the initial device control database to obtain the target device control database.

[0088] Among them, data verification can be an operation used to verify the correctness of each initial dialogue control information in the initial device control database. In the embodiment of the present disclosure, by adding this data verification process, errors caused by the model are avoided and a target device control database with higher accuracy is obtained.

[0089] In some embodiments, performing data verification on the initial device control database to obtain the target device control database may include: extracting at least two initial dialogue control information with the same initial input information from the initial device control database, and combining them to obtain a set to be verified; inputting at least two device control tasks in the set to be verified into the electronic control platform to obtain at least two electronic control instructions; merging at least two initial dialogue control information with the same electronic control instructions in the set to be verified, and sending the initial dialogue control information with different electronic control instructions to the verification platform for verification to obtain the target device control database.

[0090] The set to be verified may be a set of at least two initial dialogue control information that need to be verified, extracted from multiple initial dialogue control information. The number of sets to be verified may be one or more. Each set to be verified corresponds to the same initial input information, and the initial input information of different sets to be verified is different.

[0091] Specifically, after constructing the initial equipment control database, the equipment control device can determine whether the initial input information of each initial dialogue control information is the same, extract the initial dialogue control information with the same initial input information and combine them to obtain a set to be verified; for each set to be verified, the equipment control tasks of at least two initial dialogue control information included therein can be input into the electronic control platform to generate at least two corresponding electronic control instructions; then, for each set to be verified, at least two initial dialogue control information with the same electronic control instructions are merged and processed, and the initial dialogue control information with different electronic control instructions from other initial dialogue control information is sent to the verification platform for verification personnel to verify, and returned to the initial equipment control database after verification. After each set to be verified is processed, the processed initial equipment control database is determined as the target equipment control database, and the target equipment control database includes a database of multiple standard dialogue control information, each standard dialogue control information includes a standard input information and a corresponding equipment control task.

[0092] For example, assuming that a set to be verified includes three initial dialogue control information, the electronic control instructions obtained by inputting three device control tasks into the electronic control platform are 1, 1 and 2 respectively, the two initial dialogue control information with electronic control instructions of 1 are merged, and the initial dialogue control information with electronic control instruction of 2 is sent to the verification platform for verification.

[0093] Since the initial device control database can be stored in the form of a dialog graph, the generated target device control database can also be stored in the form of a dialog graph. The dialog graph has a tree structure. The dialog graph corresponding to the target device control database includes multiple nodes and directed edges. Each node represents a device control task in the standard dialog control information, and each directed edge stores the standard input information in the standard dialog control information, indicating the direction of the node.

[0094] In some embodiments, the device control method may also include: obtaining first voice information, and determining the corresponding first input information based on the first voice information; matching the first input information with each directed edge starting from the root node in the dialogue graph corresponding to the target device control database, determining that the node corresponding to the successfully matched directed edge is the first node, and controlling the target device based on the device control task of the first node; obtaining second voice information, and determining the corresponding second input information based on the second voice information; matching the second input information with each directed edge starting from the first node in the dialogue graph corresponding to the target device control database, determining that the node corresponding to the successfully matched directed edge is the second node, and controlling the target device based on the device control task of the second node.

[0095] The first voice information may be the first one in the continuous voice information, and the second voice information may be the voice information following the first voice information in the continuous voice information. It is understandable that voice information may continue to be input after the second voice information, and the specific number is not limited, that is, the continuous voice information may include multiple voice information.

[0096] The device control device can obtain the first voice information and determine the corresponding first input information in the manner of the above embodiment. Then, in the dialogue graph corresponding to the target device control database, the first input information is matched with the standard input information stored in each directed edge starting from the root node, and the node pointed to by the successfully matched directed edge is determined to be the first node, and the target device is controlled based on the device control task of the first node; then, the second voice information can be obtained, and the corresponding second node is determined in the same manner as the first voice information, except that the matching starts from the first node when matching in the dialogue graph, and the target device is controlled based on the device control task of the second node. Subsequently, voice information can be continued to be obtained, and the target control can be continued to be controlled to achieve continuous control of the target device. By storing the target device control database in the form of a dialogue graph, continuous control of the device can be quickly achieved based on the structure of the dialogue graph when controlling the device, thereby improving the efficiency of device control.

[0097] For example, FIG6 is a schematic diagram of a conversation graph provided by an embodiment of the present disclosure. As shown in FIG6 , the graph shows a conversation graph for an air conditioner as the target device in the target device control database. The root node of the conversation graph is the air conditioner and includes five nodes and five directed edges. In addition to the root node, each node represents a device control task in standard conversation control information, and each directed edge stores standard input information in the standard conversation control information. Arrows in the graph indicate the direction of the nodes. For example, the root node points to three nodes. The first and second nodes have the same device control task, but the standard input information of the directed edges is different. The third node's standard input information and device control task are both different from the other two nodes. The second node also points to two nodes. The conversation graph in the figure is for illustrative purposes only and is not intended to be limiting.

[0098] In the above scheme, the target device control database is stored in the form of a dialogue graph, which can better adapt to the system's understanding and decision-making needs based on the dialogue context. In addition, the device control device can also automatically combine the missed (i.e., failed matching) input information with frequency and LLM self-consistency technology (self-consistency technique) to add high-frequency and task-consistent dialogue control information to the target device control database and cache it in the cache space.

[0099] Optionally, the device control method of the disclosed embodiment may further include: eliminating data from the target device control database based on a cache elimination algorithm. The cache elimination algorithm may include a least recently used (LRU) algorithm, a least frequently used (LFU) algorithm, and an adaptive replacement cache (Adaptive Replacement Cache) algorithm, etc. The cache elimination algorithm ensures the timeliness, trend, and dynamic hit rate of data and further reduces system maintenance costs.

[0100] For example, Figure 7 is a structural diagram of a device control system provided by an embodiment of the present disclosure. As shown in Figure 7, an embodiment of the present disclosure also provides a device control system 700, including a device control device 701, a cache space 702 including a target device control database, and a task module 703 provided with a device control model. The device control device 701 is connected to the cache space 702 and the task module 703 for executing the device control method of the above-mentioned embodiment of the present disclosure.

[0101] The device control solution of the embodiment of the present disclosure constructs an initial device control database and performs data verification on the initial device control database to obtain a target device control database. By adding a data verification process, errors caused by the model are avoided, and a target device control database with higher accuracy is obtained, thereby improving the matching hit rate and accuracy in the subsequent matching process, thereby improving the accuracy of subsequent understanding of the user's voice information as a whole, and effectively improving the accuracy of device control.

[0102] Next, we'll use a specific example to further illustrate the process of building the target device control database. The dialog graph is structured as a tree called a cache tree. This cache tree has a root node, from which multiple directed edges emerge. Each directed edge stores standard input information from the standard dialog control message, indicating the direction of the node. Each directed edge points to a node, which can represent a specific device control task or a reply message to the user.

[0103] Assuming that the historical dialogue control information comes from a test user and the target device is a home appliance, a simulated home environment can be created, allowing a large number of test user modes to control the home appliance. The system automatically records and creates a dialogue diagram for the target device control database. The specific process includes: a) Creating an empty cache tree in the background. This cache tree is used to store the words spoken by the test user and the device control tasks of the LLM for these words, and the device information of the home appliance can be set, including internal information and external information. b) When the test user interacts with the home appliance through voice, the voice can be transcribed into text using ASR. This text and the above-mentioned device information are then converted into a piece of text (i.e., input information) and sent to the LLM for processing to generate the corresponding device control task. The input information {1. text, 2. internal information, 3. external information} and the device control task are recorded in the cache tree, and the specific information quantity of the input information and the device control task is recorded.

[0104] For example, in the initial state, the cache tree has only one root node. When the test user says "It's so hot", the system will convert {query: "It's so hot", device information: [off, target temperature: 25, air volume: medium, mode: air supply]} into a plain text description, such as "You are a smart device control robot. The air conditioner is currently off, the target temperature is set to 25 degrees, the air volume is medium, and the mode is air supply. Now the user says 'It's so hot'. How should you operate the air conditioner and what should you say to the user? Strictly output in the format of action: {control: "Turn on the air conditioner", response: "The air conditioner has been turned on for you"}." Then this description is determined as input information and input into the LLM. The LLM reads this conversation and outputs "action:{control:["Turn on the air conditioner", "Change the mode to cooling"], response:"The air conditioner has been turned on for you and switched to cooling mode"}", where "action:{control:["Turn on the air conditioner", "Change the mode to cooling"]" is the device control task, and "response:"The air conditioner has been turned on for you and switched to cooling mode"}" is the response. The cache tree then records this information, adding a directed edge and a node. If the test user then says "More wind," the cache tree will add another directed edge and the node it points to. If, in another conversation, the test user says "It's so hot" and "More wind," and after saying "It's so hot," the LLM output is still "action:{control:["Turn on the air conditioner", "Change the mode to cooling"], response:"The air conditioner has been turned on for you and switched to cooling mode"}", and the cache tree will add another directed edge and the node it points to. If in another conversation, the test user says "It's so hot" and "The wind is stronger", and after saying "It's so hot", the LLM output is different from the previous result, for example, it becomes "action:{control:["Start air conditioning","Set mode to cooling"], response:"The air conditioning has been turned on for you and switched to cooling mode"}", then the cache tree will continue to add a directed edge and the node pointed to by the directed edge.

[0105] It's important to note that due to the presence of noise in home environments and the variability of human speech, saying "it's so hot" to a home appliance twice might result in different ASR transcriptions. For example, one transcription might be "it's so hot." Therefore, the cache tree may contain a large number of similar expressions (e.g., "it's so hot," "it's so hot," and "hmm, it's so hot") that have the same meaning but are transcribed into many different texts due to the randomness of spoken language and noise. These different ASR transcriptions are retained in the cache tree because the same situation also occurs in real-world environments.

[0106] c) Filtering. The specific process may include: deleting nodes and their corresponding directed edges (edges pointing to and departing from them) whose information count is below a certain threshold (e.g., n=10). If the number of information occurrences is below a preset threshold, and the electronic control instructions for the device control task corresponding to the node are not preset instructions, then the node may be deleted. If the number of information for a node is below the preset threshold, but the electronic control instructions for the device control task corresponding to the node are preset instructions, this indicates that the node often corresponds to a function that the home appliance has but the user uses very rarely. For example, if you say "self-clean" to the air conditioner, it will generate the self-clean electronic control instruction, but the user may only say it once a year. Such nodes will be retained. And / or, deleting nodes and their corresponding directed edges whose corresponding time points are outside the device usage range of the current home appliance. For example, the device usage range of an air conditioner without heating function is June to August, and the device usage range of a dual-purpose air conditioner is June to August and November to March.

[0107] d) While building the cache tree in the simulation environment, select some real users (e.g., 0.1% of real users) to access the LLM during specific time periods (e.g., for a non-heating air conditioner, we can select only June to August; for a dual-purpose air conditioner, we can select June to August and November to March). The backend records the users' actual {1. query text 2. device internal information 3. external information} and the LLM's processing results.

[0108] After the cache tree is initially established, for example, the number of nodes in the cache tree reaches a preset threshold, such as 10,000, the cache tree at this time is called the initial device control database, and enters the verification stage. The cache tree obtained after the verification is called the target device control database. The verification process may include: for each node, sequentially check whether there are multiple directed edges corresponding to exactly the same input information. If so, check whether the device control tasks in the nodes pointed to by these directed edges generate exactly the same electronic control instructions on the device; if there are different electronic control commands, it means that the output of the LLM is likely to be wrong, so a warning is output, and manual review is required to delete the nodes corresponding to the wrong LLM output; if all electronic control commands are exactly the same, it means that the output of the LLM is likely to be correct and diverse. In this case, these nodes are merged and the amount of information is accumulated. All nodes and all edges are manually verified and corrected for a second time.

[0109] Once the cache tree is established, its actual usage process is as follows: 1. When the user speaks their first sentence, the system backend transcribes the sentence into text and queries the backend system for all internal and available external information about the device, combining these information to produce the current input. 2. The current input is matched against each directed edge originating from the root node, identifying those with a matching degree greater than a threshold. The threshold can be set to a higher value, and in this case, the current input is considered to fully match the directed edge. The following example illustrates what "fully matches" means: a) The user says "Turn on the air conditioner." The internal information is [Air conditioner status: Off, Target temperature: 25, Air volume: Medium, Mode: Cooling], and the external information is None. The cache tree currently matches the text corresponding to the user's speech, finding that it is identical to the text on this edge. The internal information on this edge includes the device status, which is "Off," indicating a full match. Since the cache tree does not require any additional information, the additional internal information provided by the backend system is ignored, and the matching process concludes with a "fully matched" conclusion.

[0110] 3. If the current input information completely matches a directed edge, the system directly executes the device control task corresponding to the node pointed to by the edge. When the user says the second sentence, it returns to the first step and executes each subsequent step in sequence. The only change is that it no longer starts from the root node, but starts from the current node to match the directed edges of each candidate node. The candidate nodes are all the nodes pointed to by the current node, that is, all the tail nodes connected by directed edges with the current node as the head node. 4. If no directed edge completely matches the current input information, the current input information is sent to LLM for processing. From then on, when the user says the nth sentence, it is all handed over to LLM for processing, and no matching in the cache tree is performed.

[0111] The device control method of the embodiment of the present disclosure can provide a single-round / multi-round dialogue control solution based on the device control database in the cache library, which has the following advantages: the high-frequency dialogue history (including single-round requests, or multi-round dialogues) and typical dialogue requirements are stored in the device control database by utilizing the cache mechanism and cached through the dialogue graph. During real-time operation, it can first match and determine whether there is a hit based on the cache library, which can avoid repeated calculations of typical dialogues. While improving the dialogue capability based on the large model, it can reduce operating costs and improve effect efficiency, and greatly reduce the number of calls to the real-time device control model, thereby reducing calls and calculation costs; the corresponding results can be directly obtained by matching in the cache, which greatly saves the intermediate calculation process and improves the request performance; the high-frequency dialogue scenarios are diverted by the cache mechanism, which can greatly reduce the availability and scalability of the system, and can largely maintain the dependence on high-resource consumption services while reducing the dependence on high-resource consumption services. The automated caching mechanism, combined with the cache elimination algorithm, can significantly reduce the maintenance cost of cached data while ensuring the cache hit rate. The experience improvement based on the LLM's few-shot decision-making capabilities and the large language model's powerful language comprehension, generalization, contextual learning, and emergent chain reasoning capabilities reduce the error propagation of pipeline systems to a certain extent compared to traditional pipeline model systems based on intermediate task decomposition, thereby providing better few-shot understanding and planning capabilities. This is especially true in embodied AI systems, i.e., home control systems that require physical interaction with the real physical environment. LLM can, to a certain extent, provide high-level planning capabilities for user instructions based on a small number of examples and dynamic environments, thus providing a better user experience for long-tail demands that cannot be cached.

[0112] FIG8 is a schematic diagram of the structure of a device control device provided by an embodiment of the present disclosure. The device can be implemented by software and / or hardware and can generally be integrated into an electronic device. As shown in FIG8 , the device includes:

[0113] Acquisition module 801, used to acquire target voice information;

[0114] a determination module 802 for determining current input information based on the target voice information, wherein the current input information includes a target text corresponding to the target voice information and device information of a target device corresponding to the target voice information, wherein the device information includes internal information and external information;

[0115] A matching module 803 is configured to match the current input information with a preset target device control database to obtain a matching result;

[0116] A first control module 804 is configured to control the target device based on the first device control task corresponding to the successfully matched target input information if the matching result is a successful match;

[0117] The second control module 805 is configured to input the current input information into a device control model to generate a second device control task if the matching result is a matching failure, and control the target device based on the second device control task.

[0118] Optionally, the current input information includes a target text corresponding to the target voice information and device information of a target device corresponding to the target voice information, and the device information includes internal information and external information.

[0119] Optionally, the external information is information about the environment where the target device is located.

[0120] Optionally, the determining module 802 includes:

[0121] The first unit is configured to convert the target voice information into a target text and determine a target device corresponding to the target text;

[0122] The second unit is configured to obtain device information of the target device, and determine a combination of the target text and the device information of the target device as the current input information.

[0123] Optionally, the second unit is used to:

[0124] Retrieving internal information of the target device;

[0125] External information of the target device is acquired according to a device type of the target device, a geographical location of the target device, sensors of associated devices, and / or user settings.

[0126] Optionally, the matching module 803 is configured to:

[0127] semantically matching the current input information with standard input information of a plurality of standard dialogue control information included in the target device control database one by one to determine a matching degree, wherein each of the standard dialogue control information includes standard input information and a corresponding device control task;

[0128] If there is standard input information with a matching degree greater than the matching degree threshold, the matching result is determined to be a successful match, and the standard input information is determined to be the successfully matched target input information; otherwise, the matching result is determined to be a failed match.

[0129] Optionally, the target device control database is stored in the form of a dialogue graph, and the dialogue graph corresponding to the target device control database includes multiple nodes and directed edges, each node represents a device control task in a standard dialogue control information, and each directed edge stores a standard input information in the standard dialogue control information, indicating the direction of the node.

[0130] Optionally, the device further comprises a continuous control module, configured to:

[0131] Acquire first voice information, and determine corresponding first input information based on the first voice information;

[0132] Matching the first input information with each directed edge starting from the root node in the conversation graph corresponding to the target device control database, determining that the node corresponding to the successfully matched directed edge is the first node, and controlling the target device based on the device control task of the first node;

[0133] Acquire second voice information, and determine corresponding second input information based on the second voice information;

[0134] The second input information is matched with each directed edge starting from the first node in the dialogue graph corresponding to the target device control database, the node corresponding to the directed edge with successful matching is determined to be the second node, and the target device is controlled based on the device control task of the second node.

[0135] Optionally, the device further comprises a building module for:

[0136] Constructing the target device control database.

[0137] Optionally, building blocks include:

[0138] The first unit is used to build an initial device control database;

[0139] The second unit is configured to perform data verification on the initial device control database to obtain a target device control database.

[0140] Optionally, the first unit includes:

[0141] A history subunit, configured to obtain a plurality of historical dialogue control information, wherein each of the historical dialogue control information includes historical input information and a corresponding device control task;

[0142] A construction subunit is used to construct the device control database based on the multiple historical dialogue control information.

[0143] Optionally, build subunits for:

[0144] filtering the plurality of historical dialogue control information based on device usage time and / or information quantity to obtain a plurality of initial dialogue control information;

[0145] The plurality of initial dialogue control information are stored in a database to obtain the initial device control database.

[0146] Optionally, the construction subunit is specifically used to:

[0147] Deleting the historical dialogue control information in which the number of information is less than a threshold and the electronic control instruction corresponding to the device control task does not belong to the preset instruction;

[0148] And / or, deleting the historical dialogue control information whose time point is outside the usage time range of the device from the plurality of historical dialogue control information.

[0149] Optionally, the second unit is used to:

[0150] Extracting at least two initial dialogue control information with the same initial input information from the initial device control database and combining them to obtain a set to be verified, wherein each initial dialogue control information includes initial input information and a corresponding device control task;

[0151] Inputting at least two device control tasks in the set to be verified into the electronic control platform to obtain at least two electronic control instructions;

[0152] At least two initial dialogue control information with the same electric control instructions in the set to be verified are merged, and the initial dialogue control information with different electric control instructions is sent to the verification platform for verification to obtain the target device control database.

[0153] Optionally, the target device control database is stored in a cache space.

[0154] The device control apparatus provided in the embodiments of the present disclosure can execute the device control method provided in any embodiment of the present disclosure, and has functional modules and beneficial effects corresponding to the execution method.

[0155] An embodiment of the present disclosure further provides a computer program product, including a computer program / instruction, which implements the device control method provided by any embodiment of the present disclosure when the computer program / instruction is executed by a processor.

[0156] FIG9 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present disclosure. As shown in FIG9 , the electronic device 900 includes one or more processors 901 and a memory 902 .

[0157] The processor 901 may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 900 to perform desired functions.

[0158] The memory 902 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory may include, for example, read-only memory (ROM), a hard disk, a flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 901 may execute the program instructions to implement the device control method of the embodiment of the present disclosure described above and / or other desired functions. Various contents such as input signals, signal components, noise components, etc. may also be stored in the computer-readable storage medium.

[0159] In one example, the electronic device 900 may further include an input device 903 and an output device 904 , and these components are interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0160] In addition, the input device 903 may also include, for example, a keyboard, a mouse, and the like.

[0161] The output device 904 can output various information to the outside, including determined distance information, direction information, etc. The output device 904 can include, for example, a display, a speaker, a printer, a communication network and its connected remote output device, etc.

[0162] Of course, for the sake of simplicity, FIG9 only shows some of the components related to the present disclosure in the electronic device 900, omitting components such as buses, input / output interfaces, etc. In addition, the electronic device 900 may further include any other appropriate components according to specific application scenarios.

[0163] In addition to the above-mentioned methods and devices, the embodiments of the present disclosure may also be a computer program product, which includes computer program instructions. When the computer program instructions are executed by a processor, the processor executes the device control method provided by the embodiments of the present disclosure.

[0164] The computer program product may be written in any combination of one or more programming languages ​​to implement the operations of the disclosed embodiments, including object-oriented programming languages ​​such as Java, C++, and conventional procedural programming languages ​​such as C or similar programming languages. The program code may be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0165] In addition, the embodiment of the present disclosure may also be a computer-readable storage medium having computer program instructions stored thereon. When the computer program instructions are executed by a processor, the processor is enabled to execute the device control method provided by the embodiment of the present disclosure.

[0166] The computer-readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can, for example, include but is not limited to a system, device or component of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination thereof. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0167] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0168] The foregoing description is intended only to provide specific embodiments of the present disclosure, intended to enable those skilled in the art to understand and implement the present disclosure. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure is not intended to be limited to the embodiments described herein, but rather to be construed in the broadest manner consistent with the principles and novel features disclosed herein. Industrial Applicability

[0169] The device control method provided by the present invention avoids repeated calculation of device control tasks corresponding to repeated voice, greatly reduces computing costs, improves response speed, and adds other information to the input information determined based on voice information, increasing the influencing factors considered when controlling the device. It can understand user intentions more accurately, thereby improving the accuracy of subsequent device control, and has strong industrial applicability.

Claims

1. A device control method, wherein: include: Obtain target voice information; Determining current input information based on the target voice information; Matching the current input information in a preset target device control database to obtain a matching result; If the matching result is a successful match, controlling the target device based on the first device control task corresponding to the successfully matched target input information; If the matching result is a matching failure, the current input information is input into a device control model to generate a second device control task, and the target device is controlled based on the second device control task.

2. The method according to claim 1, wherein: The current input information includes the target text corresponding to the target voice information and the device information of the target device corresponding to the target voice information, and the device information includes internal information and external information.

3. The method according to claim 2, wherein: The external information is the environment information where the target device is located.

4. The method according to any one of claims 1 to 3, wherein: Determining current input information based on the target voice information includes: Convert the target speech information into a target text, and determine a target device corresponding to the target text; The device information of the target device is acquired, and the target text and the device information of the target device are combined to determine as the current input information.

5. The method according to claim 4, wherein: Obtaining device information of the target device, including: Retrieving internal information of the target device; The external information of the target device is acquired according to the device type of the target device, the geographical location of the target device, the sensor of the associated device and / or the user setting.

6. The method according to any one of claims 1 to 5, wherein: The current input information is matched in a preset target device control database to obtain a matching result, including: Semantically matching the current input information with standard input information of a plurality of standard dialogue control information included in the target device control database one by one to determine a matching degree, wherein each of the standard dialogue control information includes standard input information and a corresponding device control task; If there is standard input information with a matching degree greater than the matching degree threshold, the matching result is determined to be a successful match. The standard input information is determined as the target input information that has been matched successfully; otherwise, the matching result is determined to be a matching failure.

7. The method according to any one of claims 1 to 6, wherein: The target device control database is stored in the form of a dialogue graph. The dialogue graph corresponding to the target device control database includes multiple nodes and directed edges. Each node represents a device control task in a standard dialogue control information. Each directed edge stores a standard input information in the standard dialogue control information, indicating the direction of the node.

8. The method according to claim 7, wherein: The method further comprises: Acquire first voice information, and determine corresponding first input information based on the first voice information; Matching the first input information with each directed edge starting from the root node in the dialog graph corresponding to the target device control database, determining that the node corresponding to the directed edge with successful matching is the first node, and controlling the target device based on the device control task of the first node; Acquire second voice information, and determine corresponding second input information based on the second voice information; The second input information is matched with each directed edge starting from the first node in the dialogue graph corresponding to the target device control database, the node corresponding to the directed edge with successful matching is determined to be the second node, and the target device is controlled based on the device control task of the second node.

9. The method according to any one of claims 1 to 8, wherein: The method further comprises: The target device control database is constructed.

10. The method according to claim 9, wherein: Constructing the target device control database includes: Build the initial equipment control database; The initial device control database is verified to obtain a target device control database.

11. The method according to claim 10, wherein: Build the initial equipment control database, including: Acquire multiple historical dialogue control information, wherein each of the historical dialogue control information includes historical input information and a corresponding device control task; The device control database is constructed based on the plurality of historical dialogue control information.

12. The method according to claim 11, wherein: Building the device control database based on the plurality of historical dialogue control information includes: Filtering the plurality of historical dialogue control information based on device usage time and / or information quantity to obtain a plurality of initial dialogue control information; The plurality of initial dialogue control information are stored in a database to obtain the initial device control database.

13. The method according to claim 12, wherein: Filtering the plurality of historical conversation control information based on device usage time and / or information quantity includes: Deleting the historical dialogue control information in which the number of information is less than the number threshold and the electric control instruction corresponding to the device control task does not belong to the preset instruction; And / or, the historical dialogue control information whose time point is outside the use time range of the device is deleted from the plurality of historical dialogue control information.

14. The method according to claim 10, wherein: Performing data verification on the initial device control database to obtain a target device control database includes: Extracting at least two initial dialogue control information with the same initial input information from the initial device control database, and combining them to obtain a set to be verified, wherein each of the initial dialogue control information includes initial input information and a corresponding device control task; Inputting at least two equipment control tasks in the set to be verified into the electronic control platform to obtain at least two electronic control instructions; At least two initial dialogue control information with the same electric control instructions in the set to be verified are merged, and the initial dialogue control information with different electric control instructions is sent to the verification platform for verification to obtain the target device control database.

15. The method according to any one of claims 1 to 14, wherein: The target device control database is stored in a cache space.

16. A device control apparatus, wherein: include: An acquisition module, used to acquire target voice information; A determination module, used to determine current input information based on the target voice information; A matching module, used to match the current input information in a preset target device control database to obtain a matching result; A first control module, configured to control the target device based on a first device control task corresponding to the target input information that successfully matches, if the matching result is a successful match; The second control module is used to input the current input information into the device control model to generate a second device control task if the matching result is a matching failure, and control the target device based on the second device control task.

17. A device control system, wherein: It comprises a device control device, a cache space including a target device control database and a task module provided with a device control model. The device control device is connected with the cache space and the task module and is used to execute the device control method described in any one of claims 1 to 15.

18. An electronic device, wherein: The electronic device comprises: processor; a memory for storing instructions executable by the processor; The processor is used to read the executable instructions from the memory and execute the instructions to implement the device control method described in any one of claims 1-15 above.

19. A computer-readable storage medium, wherein: The storage medium stores a computer program, and the computer program is used to execute the device control method described in any one of claims 1 to 15.

20. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the device control method described in any one of claims 1-15 is implemented.

Images