Control method and device of refrigeration equipment and refrigeration equipment

Abstract

The invention discloses a control method and device of refrigeration equipment and the refrigeration equipment, and belongs to the technical field of refrigeration equipment. The control method of the refrigeration equipment comprises the steps that task information input by a user is obtained; the task information comprises at least one of text information, voice information, image information and audio and video information; the plan agent splits the task information to obtain at least one sub-task text feature; and the reflection agent outputs a task result corresponding to the task information based on the at least one sub-task text feature. According to the control method of the refrigeration equipment, full-cycle development workflow of planning, decision making, reflection and the like of the intelligent agents is worked, the workflow efficiency among the multiple intelligent agents is improved, the interaction accuracy of the dialogue intelligent agents and the user is improved, and therefore the use experience of the user is improved.

Description

Technical Field

[0001] This application belongs to the field of refrigeration equipment technology, and particularly relates to a control method, device and refrigeration equipment for refrigeration equipment. Background Technology

[0002] Users have an increasing demand for human-computer interaction through cooling devices. Related technologies primarily utilize machine learning and deep learning methods for the research and application of dialogue agents. However, these methods cannot effectively handle complex tasks, and the accuracy of dialogue agents interacting with users is low, negatively impacting the user experience. Summary of the Invention

[0003] This application aims to address at least one of the technical problems existing in the related art. To this end, this application proposes a control method, apparatus, and cooling equipment for a cooling device, which streamlines the entire development process of an intelligent agent, including planning, decision-making, and reflection, thereby improving the workflow efficiency between multiple intelligent agents, enhancing the accuracy of interaction between the dialogue agent and the user, and thus improving the user experience.

[0004] In a first aspect, this application provides a control method for a refrigeration device, the refrigeration device comprising: a planning agent and a reflecting agent; the method comprising:

[0005] The planning agent acquires task information input by the user; the task information includes at least one of the following: text information, voice information, image information, and audio / video information.

[0006] The planning agent breaks down the task information to obtain at least one sub-task text feature, and sends the at least one sub-task text feature to the reflection agent; each sub-task text feature corresponds to execution order information and weight information;

[0007] The reflective agent outputs the task result corresponding to the task information based on the text features of at least one sub-task.

[0008] According to the control method of the refrigeration equipment of this application, by acquiring task information in various forms, users can interact with the refrigeration equipment in multiple ways, with wide application scenarios and flexible usage. By setting up multiple intelligent agents, the planning intelligent agent breaks down the task information into multiple sub-task text features, and the reflection intelligent agent outputs the task results corresponding to the task information based on the multiple sub-task texts, realizing efficient collaborative work between the planning intelligent agent and the reflection intelligent agent. The entire life cycle of intelligent agent development workflow from planning, decision-making and reflection is streamlined, improving the workflow efficiency between multiple intelligent agents, improving the accuracy of the output task results, improving the interaction accuracy between the dialogue intelligent agent and the user, thereby improving the user experience.

[0009] According to the control method for the refrigeration equipment of this application, the planning agent decomposes the task information to obtain at least one sub-task text feature, including:

[0010] The task information is converted into text to obtain the first task text;

[0011] Based on queue and stack algorithms, the first task text is split to obtain at least one sub-task text feature corresponding to the execution order information;

[0012] The text features of the at least one subtask are processed based on a graph neural network to obtain the at least one subtask text features corresponding to the weight information.

[0013] According to the control method for the refrigeration equipment of this application, the planning agent performs text conversion on the task information to obtain a first task text, including:

[0014] When the task information includes the voice information, the planning agent converts the voice information into first text information.

[0015] When the task information includes the image information, the planning agent extracts image features from the image information to obtain the image features corresponding to the image information, and converts the image features into a text description.

[0016] The planning agent obtains the first task text based on at least one of the first text information and the text description.

[0017] According to the control method for the refrigeration equipment of this application, the reflective agent outputs the task result corresponding to the task information based on the text features of the at least one sub-task, including:

[0018] The reflective agent outputs the task result corresponding to the task information based on the context corresponding to the text features of each subtask, the context relationship corresponding to the text features of each subtask, and the multi-step reasoning information corresponding to the text features of each subtask.

[0019] According to the control method of the refrigeration device of this application, the refrigeration device further includes: a memory agent; the reflecting agent outputs the task result corresponding to the task information based on the context corresponding to the text features of each sub-task, the context relationship corresponding to the text features of each sub-task, and the multi-step reasoning information corresponding to the text features of each sub-task, including:

[0020] The memory agent constructs the target relationship between the text features of each subtask based on the context corresponding to the text features of each subtask, the contextual relationship corresponding to the text features of each subtask, and the multi-step reasoning information corresponding to the text features of each subtask.

[0021] Based on the target relationship, the reflective agent outputs the task result corresponding to the task information.

[0022] According to the control method for the refrigeration equipment of this application, the reflective agent outputs the task result corresponding to the task information based on the target relationship, including:

[0023] Based on the target relationship, determine the current subtask and the previous subtask corresponding to the current subtask;

[0024] Based on the execution result of the previous subtask and the execution content of the current subtask, obtain the execution result of the current subtask;

[0025] If the current subtask is the last subtask among all subtasks included in the target relationship, output the task result corresponding to the task information.

[0026] According to the control method for the refrigeration equipment of this application, the reflective agent outputs the task result corresponding to the task information based on the text features of the at least one sub-task, including:

[0027] Based on the task information, determine the output method corresponding to the task scenario;

[0028] Output the task result corresponding to the task information based on the output method corresponding to the task scenario.

[0029] According to the control method for the refrigeration equipment of this application, the step of determining the output mode corresponding to the task scenario based on the task information further includes:

[0030] Based on the task information, train and generate a network model;

[0031] Based on the model parameters corresponding to the generated network model, the network model is used to generate inference tasks.

[0032] Based on the reasoning task, determine the output method corresponding to the task scenario.

[0033] According to the control method for the refrigeration equipment of this application, after obtaining the task information input by the user, the method further includes:

[0034] The reflective agent acquires historical task information; the historical task information includes: the historical task and the historical task result corresponding to the historical task.

[0035] If the first similarity between the task information and the historical task is greater than the target similarity threshold, the task result corresponding to the task information is retrieved from the historical task information and the task result corresponding to the task information is output.

[0036] According to the control method of the refrigeration equipment of this application, the first similarity is calculated based on the contextual similarity between the task information and the historical task and / or the semantic similarity between the task information and the historical task.

[0037] Secondly, this application provides a control device for a refrigeration device, the refrigeration device including a planning agent and a reflecting agent; the device includes:

[0038] The first processing module is used to enable the planning agent to obtain task information input by the user; the task information includes at least one of the following: text information, voice information, image information, and audio-visual information;

[0039] The second processing module is used to enable the planning agent to split the task information to obtain at least one sub-task text feature, and send the at least one sub-task text feature to the reflection agent; each sub-task text feature corresponds to execution order information and weight information;

[0040] The third processing module is used to enable the reflective agent to output the task result corresponding to the task information based on the text features of the at least one sub-task.

[0041] According to the control device for the refrigeration equipment of this application, by acquiring task information in various forms, users can interact with the refrigeration equipment in multiple ways, with wide application scenarios and flexible usage methods. By setting up multiple intelligent agents, the planning intelligent agent breaks down the task information into multiple sub-task text features, and the reflection intelligent agent outputs the task results corresponding to the task information based on the multiple sub-task texts, realizing efficient collaborative work between the planning intelligent agent and the reflection intelligent agent. The entire lifecycle of intelligent agent development workflow from planning, decision-making and reflection is streamlined, improving the workflow efficiency between multiple intelligent agents, improving the accuracy of the output task results, and improving the interaction accuracy between the dialogue intelligent agent and the user, thereby improving the user experience.

[0042] Thirdly, this application provides a refrigeration device, comprising:

[0043] At least one intelligent agent involved in the control method of the refrigeration equipment as described in the first aspect.

[0044] Fourthly, this application provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the control method for the refrigeration device as described in the first aspect above.

[0045] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the control method for the refrigeration equipment as described in the first aspect above.

[0046] The above-described one or more technical solutions in the embodiments of this application have at least one of the following technical effects:

[0047] By acquiring task information in various forms, users can interact with the refrigeration equipment in multiple ways, with wide application scenarios and flexible usage. By setting up multiple agents, the planning agent breaks down the task information into multiple sub-task text features, and the reflecting agent outputs the task results corresponding to the task information based on the multiple sub-task texts. This achieves efficient collaboration between the planning agent and the reflecting agent, and streamlines the entire lifecycle of the agent's planning, decision-making, and reflection processes, improving the workflow efficiency between multiple agents, increasing the accuracy of the output task results, and enhancing the accuracy of the interaction between the dialogue agent and the user, thereby improving the user experience.

[0048] Furthermore, by using a planning agent to convert task information into text, various forms of data are unified into text data, reducing data complexity and improving the efficiency of subsequent data processing. The first task text is also split to obtain at least one sub-task text feature, breaking down complex tasks into multiple simple tasks, which facilitates the agent's understanding and execution, forming an effective task execution plan and improving the efficiency and accuracy of user-agent interaction.

[0049] Furthermore, by setting up a memory agent, the reflecting agent can obtain the logical relationship between the text features of each subtask based on the context, contextual relationships, and multi-step reasoning information corresponding to the text features of each subtask. Based on the logical relationship, the task result can be output, which can improve the accuracy of task execution, provide efficient task application services, and improve the user experience.

[0050] Furthermore, by comparing the similarity between new task information and historical tasks, task results can be retrieved from historical task information, and the task results corresponding to new task information can be output efficiently, thereby improving the task response rate and enhancing the user experience.

[0051] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0052] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0053] Figure 1 This is one of the flowcharts illustrating the control method for a refrigeration device provided in the embodiments of this application;

[0054] Figure 2 This is one of the schematic diagrams illustrating the principle of the control method for the refrigeration equipment provided in the embodiments of this application;

[0055] Figure 3 This is a second schematic diagram illustrating the principle of the control method for the refrigeration equipment provided in the embodiments of this application;

[0056] Figure 4 This is the third schematic diagram illustrating the principle of the control method for the refrigeration equipment provided in the embodiments of this application;

[0057] Figure 5 This is the fourth schematic diagram illustrating the principle of the control method for the refrigeration equipment provided in the embodiments of this application;

[0058] Figure 6 This is the fifth schematic diagram illustrating the principle of the control method for the refrigeration equipment provided in the embodiments of this application;

[0059] Figure 7 This is the sixth schematic diagram illustrating the principle of the control method for the refrigeration equipment provided in the embodiments of this application;

[0060] Figure 8 This is the seventh schematic diagram illustrating the principle of the control method for the refrigeration equipment provided in the embodiments of this application;

[0061] Figure 9 This is the eighth schematic diagram illustrating the principle of the control method for the refrigeration equipment provided in the embodiments of this application;

[0062] Figure 10 This is the ninth schematic diagram illustrating the principle of the control method for the refrigeration equipment provided in the embodiments of this application;

[0063] Figure 11 This is the tenth schematic diagram of the principle of the control method for the refrigeration equipment provided in the embodiments of this application;

[0064] Figure 12 This is eleventh of the schematic diagrams illustrating the principle of the control method for the refrigeration equipment provided in the embodiments of this application;

[0065] Figure 13 This is a schematic diagram of the structure of the refrigeration equipment provided in the embodiments of this application;

[0066] Figure 14 This is a schematic diagram of the structure of the refrigeration equipment provided in the embodiments of this application;

[0067] Figure 15 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0068] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0069] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0070] The following description, in conjunction with the accompanying drawings, details the control method, control device, refrigeration equipment, and readable storage medium for the refrigeration equipment provided in this application, through specific embodiments and application scenarios.

[0071] The control method for the refrigeration equipment can be applied to the terminal, and can be executed by the hardware or software in the terminal.

[0072] The terminal includes, but is not limited to, portable communication devices such as mobile phones or tablets. It should also be understood that, in some embodiments, the terminal may not be a portable communication device, but rather a desktop computer.

[0073] The following embodiments describe a terminal including a display and a touch-sensitive surface. However, it should be understood that the terminal may include one or more other physical user interface devices such as a physical keyboard, mouse, and joystick.

[0074] The freshness detection method provided in this application can be executed by a storage device, such as by a microcontroller; in some embodiments, it can also be executed by a server connected to the storage device, or by a server and a microcontroller jointly.

[0075] The control method for a refrigeration device provided in this application embodiment can be executed by a refrigeration device, a control device for the refrigeration device installed on the refrigeration device, a server electrically connected to the refrigeration device, or a user terminal communicatively connected to the refrigeration device, including but not limited to mobile terminals and non-mobile terminals.

[0076] like Figure 1 As shown, the control method of the refrigeration equipment includes steps 110, 120 and 130.

[0077] The term "refrigeration equipment" can be understood as a broad category of refrigeration and storage equipment, including but not limited to refrigerators, freezers, display cases, beverage cabinets, wine cabinets, refrigerated display cases, and refrigerated vending machines. Refrigeration equipment has diverse structural forms and a wide range of applications.

[0078] The cooling equipment includes: planning agents and reflective agents.

[0079] Step 110: The planning agent obtains the task information input by the user;

[0080] In this step, the task information is the information about the user's interaction with the refrigeration equipment.

[0081] Task information includes at least one of the following: text information, voice information, image information, and audio / video information.

[0082] In actual execution, task information can be obtained through various input methods such as real-time or offline voice, text, images, and video.

[0083] like Figure 2 As shown, task information can be obtained through the task information acquisition module.

[0084] like Figure 3 As shown, during actual execution, task information input by the user can be received through the data channel and the context-aware module.

[0085] The data channel and context-aware module can receive user-inputted task information through one or more of various data channels, such as 5G / 6G, WiFi, APP, microphone, eye tracker, microphone array, mobile phone, tablet, global wide area network, computer terminal, sensor, artificial intelligence (AI) acquisition, Bluetooth, and acquisition tools.

[0086] In actual execution, one or more of the following methods can be used: hardware collectors, software collectors, and tool collectors. This will enable the acquisition of data from sources such as refrigeration equipment and its own information, internal and external environmental context information, human-computer interaction, multi-turn dialogue, knowledge Q&A, and file formats.

[0087] When the hardware acquisition device is an independent peripheral, data can be collected by an external device.

[0088] Among them, knowledge-based question answering and context-aware information of internal and external environments can include data such as refrigerator knowledge, casual conversation, songs, music, and recipes.

[0089] Understandably, acquiring task information involves data collection for tasks such as interaction between refrigeration equipment and various data information, user information interaction, human-computer interaction, intelligent question answering, multi-turn dialogue, personal intelligent assistant, data transmission, data generation tools, large model prediction multi-task, and multi-task combination.

[0090] Continue to refer to Figure 3 In actual execution, the form of task information obtained through multiple data channels may be different. For example, task information obtained through a microphone array is voice information, data obtained through a mobile phone is text information, and task information obtained through a camera is image information, etc.

[0091] In actual execution, at least one of text information, voice information, image information, and audio-visual information can be obtained through multiple data channels.

[0092] It is understandable that the acquired text, voice, image, and audio / video information are all task information.

[0093] Step 120: The planning agent breaks down the task information to obtain at least one sub-task text feature, and sends the at least one sub-task text feature to the reflection agent.

[0094] In this step, the subtask text features are the features obtained by breaking down the task information into the smallest units.

[0095] The text features of the subtasks correspond to execution order information and weight information.

[0096] The execution order information refers to the execution order of the text features of each subtask during the execution of the task corresponding to the task information.

[0097] The weight information represents the strength of the relationship between the text features of two adjacent subtasks.

[0098] like Figure 2 As shown, the task information obtained can be processed by the task information processing module so that the corresponding task can be executed subsequently.

[0099] In actual execution, the planning agent can break down complex online tasks and complex offline tasks into their smallest units.

[0100] Online complex tasks are tasks issued to users in real time.

[0101] For example, complex online tasks could be tasks like "What's left in the fridge?" or "Are there any apples left in the fridge?"

[0102] Offline complex tasks can include tasks such as generating timed broadcasts of question and answer information and recommending question and answer information.

[0103] For example, complex offline tasks can include tasks such as sending information on near-expiry ingredients, sending occasional Q&A messages about food preservation, and sending information pushes.

[0104] Taking the task information "What's left in the refrigerator?" as an example, the planning agent breaks down the task information, and the resulting sub-task text features may include: querying the ingredients in the refrigeration equipment and outputting the specific ingredients in the refrigeration equipment.

[0105] Of course, in actual execution, the task information can be split in any feasible way, and this application does not limit it.

[0106] In actual execution, after the planning agent splits the task information in any feasible way, multiple sub-task text features can be obtained. In the process of splitting the task information, attention is paid to the execution order between the text features of each sub-task and the degree of correlation between two adjacent sub-task text features in order to obtain more information.

[0107] like Figure 4 As shown, the planning agent splits the task based on the acquired task information, resulting in multiple sub-task text features.

[0108] The planning agent sends at least one subtask text feature obtained from the decomposition to the reflecting agent, so that the reflecting agent can output the task result based on the text features of each subtask.

[0109] Step 130: The reflecting agent outputs the task result corresponding to the task information based on the text features of at least one subtask.

[0110] In this step, the task result is the result obtained by executing the instructions corresponding to the task information.

[0111] like Figure 2 As shown, the task execution module can execute the instructions corresponding to the task information to meet user needs.

[0112] After the planning agent breaks down the task information, the reflection agent confirms or asynchronously calls back the text features of each subtask obtained from the breakdown, and then outputs the task result corresponding to the task information after executing the text features of each subtask.

[0113] In some embodiments, a reflective agent can provide multimodal outputs and embodied intelligence.

[0114] Embodied intelligence is the ability to interact with the environment in real time through perception and interaction.

[0115] Continue to refer to Figure 4 In actual implementation, by constructing a workflow autonomous system for dialogue multi-agents, the application business processes of multiple dialogue agents are built, including collaborative application tasks such as planning and reflection.

[0116] Continue to refer to Figure 4 Dialogue multi-agent systems can be categorized into planning agents and reflecting agents based on the application task.

[0117] As shown in Table 1, in actual execution, the tasks that the dialogue multi-agent can perform include, but are not limited to, the tasks shown in the table below.

[0118] Table 1

[0119]

[0120] According to the control method for refrigeration equipment provided in the embodiments of this application, by acquiring various forms of task information, users can interact with the refrigeration equipment in multiple ways, with wide application scenarios and flexible usage methods. By setting up multiple intelligent agents, the planning intelligent agent decomposes the task information to obtain multiple sub-task text features, and the reflection intelligent agent outputs the task results corresponding to the task information based on the multiple sub-task texts, realizing efficient collaborative work between the planning intelligent agent and the reflection intelligent agent. The entire lifecycle of intelligent agent development workflow from planning, decision-making and reflection is streamlined, improving the workflow efficiency between multiple intelligent agents, improving the accuracy of the output task results, improving the interaction accuracy between the dialogue intelligent agent and the user, thereby improving the user experience.

[0121] In some embodiments, step 120 may further include:

[0122] The task information is converted into text to obtain the first task text;

[0123] Based on queue and stack algorithms, the first task text is split to obtain at least one sub-task text feature with execution order information.

[0124] The text features of at least one subtask are processed using a graph neural network to obtain at least one subtask text feature with corresponding weight information.

[0125] In this embodiment, the first task text is text obtained by converting various forms of task information into text form.

[0126] Queue algorithms use a first-in, first-out (FIFO) data structure.

[0127] The stack algorithm is a last-in-first-out (LIFO) data structure.

[0128] In some embodiments, after obtaining task information, the task information can be cleaned and its format converted.

[0129] In this embodiment, when the task information is audio and video information, appropriate tools or scripts can be used to separate the voice and video to obtain effective voice and video data.

[0130] When the task information is text, stop word processing and deduplication can be performed on the text information.

[0131] In some embodiments, the planning agent can use queue and stack algorithms to split the first task text to obtain multiple subtasks of the task information to be executed. Each subtask can be stored in a queue or stack according to the execution order.

[0132] For example, if the first task text is "What ingredients are left in the refrigerator?", then the first task text can be split into "Query what ingredients are in the refrigerator" and "Output the ingredients by category".

[0133] In actual execution, the output can be categorized, such as outputting ingredients based on categories like fruits, meats, seafood, and beverages.

[0134] Of course, in actual implementation, ingredients can also be output in any other feasible way, and this application does not limit this.

[0135] After obtaining the text features of each subtask, the text features of each subtask can be processed by a graph neural network. The graph neural network represents the text features of each subtask as nodes, and the relationship between the text features of each subtask as edges. The edges between two adjacent text features of each subtask have weight information.

[0136] In actual execution, after processing the text features of each subtask through a graph neural network, the relationships and strengths between the text features of each subtask can be obtained.

[0137] According to the control method for refrigeration equipment provided in the embodiments of this application, the task information is converted into text by a planning agent, unifying various forms of data into text data, reducing data complexity, and improving the efficiency of subsequent data processing; and the first task text is split to obtain at least one sub-task text feature, breaking down complex tasks into multiple simple tasks to form an effective task execution plan, which is convenient for the agent to understand and execute, and improves the interaction efficiency and accuracy between the user and the agent.

[0138] Continue to refer to Figure 3 In some embodiments, the planning agent performs text conversion on the task information to obtain the first task text, which may further include:

[0139] If the task information includes voice information, the voice information is converted into first text information;

[0140] When the task information includes image information, image feature extraction is performed on the image information to obtain the corresponding image features, and the image features are then converted into text descriptions.

[0141] The planning agent obtains the first task text based on at least one of the first text information and the text description.

[0142] In this embodiment, the first text information is the information obtained after converting the voice information into text.

[0143] The text description is the text that describes the image information after the image information has been converted into text.

[0144] In actual execution, the voice information of refrigeration equipment has spatiotemporal characteristics and contextual information features. A speech recognition model architecture based on convolutional enhancement transformer blocks can be adopted to obtain the contextual information of the voice information.

[0145] It is understandable that the local feature information of the voice information of the refrigeration equipment is very important, and the local feature information of the voice information of the refrigeration equipment can be obtained through convolution enhancement.

[0146] In some embodiments, speech information can be converted into first text information using an end-to-end deep neural network model based on a Gaussian neural network model and an encoding / decoding structure.

[0147] For example, the Kaldi open-source speech recognition architecture can be used to convert speech information into first text information.

[0148] Of course, other feasible methods can also be used to convert voice information into first text information, and this application does not limit this.

[0149] In actual execution, when the task information includes image information, image recognition technology can be used to extract image features from the image information, obtain the image features corresponding to the image information, and convert the image features into text descriptions.

[0150] In some embodiments, when the task information includes audio and video information, a deep neural network fusion model structure based on a diffusion network model can be used to convert the audio and video information into text information.

[0151] Continue to refer to Figure 3 In other embodiments, audio and video information can be separated. Considering that the sentences of users interacting with the cooling device may be complex, such as inconsistent sentence lengths, different word structures, and the presence of contextual relationships, an image recognition method based on a migration and distillation diffusion deep fusion network model using spatiotemporal and long-distance dependent features can be used to convert audio and video information into text information.

[0152] In practice, the distillation-diffusion deep fusion model integrates the transfer and teacher model outputs of the distillation-diffusion deep fusion model to mine rich semantic features of sentence sequences.

[0153] Specifically, in the training of deep fusion models using distillation, migration, or quantization diffusion, the teacher model is introduced into a temporally and spatially continuous student model, resulting in the student model having the same parameters as the teacher model.

[0154] The deep fusion model of distillation, migration, or quantization diffusion is converted into a student model with discrete time steps and a short number of steps. This process is repeated N times until the teacher model is distilled to half the number of steps of the student model, i.e., N / 2.

[0155] The specific value of N can be determined based on the actual situation, and this application does not impose any restrictions.

[0156] In some embodiments, when the text information is a file, the text information in the file can be extracted using appropriate tools.

[0157] For example, when the text information is in a PDF file, the text content of the PDF file can be extracted using PDF tools and an Optical Character Recognition (OCR) interface.

[0158] PDF files can be documents such as academic papers or instruction manuals for refrigeration equipment.

[0159] In actual execution, after obtaining the first text information and text description, data fusion and task filtering can be performed on the first text information and text description to obtain the first task text, so that the large model can be trained and generate information based on the first task text.

[0160] Among them, filtering refers to data cleaning, data generation, and extraction of specific target slot rules from the first text information and text description to obtain high-quality first task text.

[0161] In actual implementation, it can be done through methods such as Figure 3The data processing module shown processes task information and, through the data fusion module, transforms and merges the acquired text information, voice information, image information, and audio-visual information to obtain the first text information.

[0162] In some embodiments, if the image and audio / video information themselves are valid, they can be directly applied to subsequent large models.

[0163] According to the control method of the refrigeration equipment provided in the embodiments of this application, by converting different forms of task information into text, high-quality data is obtained in the same data format to prepare data for downstream tasks, which is conducive to improving the efficiency of subsequent large-scale model processing tasks based on text information.

[0164] like Figure 3 As shown, in some embodiments, after step 110, the method may further include:

[0165] The continuously acquired task information is accumulated to obtain historical cumulative data.

[0166] In this embodiment, the collected historical multimodal data can be converted into text data, and the text data can be vectorized and mapped to tasks such as transfer learning of the same vector control.

[0167] In actual implementation, such as Figure 3 As shown, task information can be accumulated through the historical data accumulation module.

[0168] According to the control method of the refrigeration equipment provided in the embodiments of this application, by accumulating historical data and mapping the historical data to the same vector space, it is beneficial to feature extraction with a deep convolutional network model to obtain data representation. In addition, it can further ensure that the subsequent output is comprehensive and rich in high-quality information.

[0169] In some embodiments, step 130 may further include:

[0170] The reflective agent outputs the task results corresponding to the task information based on the context corresponding to the text features of each subtask, the contextual relationships corresponding to the text features of each subtask, and the multi-step reasoning information corresponding to the text features of each subtask.

[0171] In this embodiment, the context refers to the linguistic environment or background information of the subtask text features.

[0172] The context may include subtask text features with a number of pre-target subtask text features or a number of post-target subtask text features.

[0173] The specific number of targets can be determined based on the actual situation, and this application does not impose any restrictions.

[0174] Contextual relationships refer to the connections or influences between different subtask text features based on the context.

[0175] Contextual relationships can include: collocation relationships between text features of each subtask, syntactic structural relationships between text features of each subtask, and semantic logical relationships between text features of each subtask.

[0176] Multi-step reasoning information refers to information obtained through multi-step logical reasoning or analysis to arrive at the correct conclusion.

[0177] Multi-step reasoning information can include logical reasoning relationships such as causal relationships and parallel relationships.

[0178] Taking the reflective agent as an example, based on the text features of each subtask in the task "What food is left in the refrigerator? Then query what food is in the refrigerator and output the specific food items," the task results corresponding to the task information are output.

[0179] In actual execution, the reflective agent performs a complete analysis of the text features of each subtask. For example, the preceding text before "What other ingredients are in the refrigerator?" is "empty", and the following text is "Query what ingredients are in the refrigerator". Based on the text features of each subtask, the agent constructs the relationship between each subtask and infers how each subtask and each agent can work together, thereby outputting the task results corresponding to the task information.

[0180] Of course, other feasible methods can be used to analyze the context corresponding to the text features of each subtask, the contextual relationships corresponding to the text features of each subtask, and the multi-step reasoning information corresponding to the text features of each subtask, such as more detailed hierarchical analysis, or understanding based on words and phrases. This application does not limit these methods.

[0181] In some embodiments, the cooling device further includes a memory agent, which, based on the context corresponding to the text features of each sub-task, the contextual relationships corresponding to the text features of each sub-task, and the multi-step reasoning information corresponding to the text features of each sub-task, outputs the task result corresponding to the task information, and may further include:

[0182] The memory agent constructs the target relationship between the text features of each subtask based on the context corresponding to the text features of each subtask, the contextual relationship corresponding to the text features of each subtask, and the multi-step reasoning information corresponding to the text features of each subtask.

[0183] The reflective agent outputs the task results corresponding to the task information based on the target relationship.

[0184] In this embodiment, the target relationship is the relationship obtained by analyzing the text features of each subtask.

[0185] In actual execution, the memory agent analyzes the context corresponding to the text features of each subtask, the contextual relationships corresponding to the text features of each subtask, and the multi-step reasoning information corresponding to the text features of each subtask. Based on the context corresponding to the text features of each subtask, the contextual relationships corresponding to the text features of each subtask, and the multi-step reasoning information corresponding to the text features of each subtask, the agent constructs relationships to obtain the target relationship.

[0186] After obtaining the target relationship, the reflective agent outputs the task results corresponding to the task information based on the target relationship.

[0187] According to the control method of the refrigeration equipment provided in the embodiments of this application, the memory agent analyzes the context, context relationship and multi-step reasoning information corresponding to the text features of each sub-task to obtain the semantic information and logical relationship of the text features of each sub-task. Based on the analysis results, a workflow is constructed. The intelligent agent reflects on the relationship between each sub-task and outputs the task results corresponding to the task information. This effectively completes the construction of multiple intelligent agent business processes, improves workflow efficiency and the accuracy of human-computer interaction.

[0188] In actual execution, by analyzing the text features of each subtask obtained from the decomposition, the logical relationship between the text features of each subtask is obtained. The intelligent agent then reflects on the logical relationship between each subtask and outputs the task results corresponding to the task information.

[0189] According to the control method of the refrigeration equipment provided in the embodiments of this application, by setting up a memory agent, the reflecting agent can obtain the logical relationship between the text features of each sub-task based on the context, context relationship and multi-step reasoning information corresponding to the text features of each sub-task, and thus output the task result based on the logical relationship, which can improve the accuracy of task execution, efficiently provide task application services and improve the user experience.

[0190] In some embodiments, the reflective agent outputs the task result corresponding to the task information based on the target relationship, and may further include:

[0191] Based on the target relationship, determine the current subtask and the previous subtask corresponding to the current subtask;

[0192] Based on the execution result of the previous subtask and the execution content of the current subtask, obtain the execution result of the current subtask.

[0193] If the current subtask is the last subtask among all subtasks included in the target relationship, output the task result corresponding to the task information.

[0194] In this embodiment, the previous subtask is the task preceding the subtask executed at the current acquisition time.

[0195] The execution result is the feedback after the task is executed.

[0196] The current subtask is the task being executed at the current data collection moment.

[0197] The execution content refers to the actions that the subtask actually needs to perform.

[0198] The last subtask is the last task that needs to be executed to complete the task information.

[0199] In some embodiments, the reflective agent can implement callbacks after encapsulating task functions and task algorithms.

[0200] In this embodiment, during actual execution, after the task information is split, the split tasks can be confirmed or asynchronously called back. Based on the execution result of the previous sub-task and the execution content of the current sub-task, the execution result of the current sub-task can be obtained, and each sub-task can be executed step by step to output the task result corresponding to the task information.

[0201] In actual execution, the task results can be output by calling the toolset formed by the encapsulation of the large model.

[0202] like Figure 5 As shown, taking a refrigerator as an example of a refrigeration device, data can be collected through an input acquisition module or a data acquisition device. Then, multiple agents work together based on the task information collected by the acquisition module to perform multimodal adaptation. The task information is processed through a multimodal large model, and the processing results are returned to the refrigerator to display the task results to the user.

[0203] According to the control method of the refrigeration equipment provided in the embodiments of this application, the intelligent agent executes each sub-task step by step through reflection, and outputs the task results through the corresponding toolset to complete the task requirements, effectively generate the solution generator for the current task, improve the accuracy and response speed of task matching, meet more business needs, and improve the user experience.

[0204] In some embodiments, the reflective agent outputs the task result corresponding to the task information based on at least one sub-task text feature, and may further include:

[0205] Based on the task information, determine the output method corresponding to the task scenario;

[0206] Output the task results corresponding to the task information based on the output method corresponding to the task scenario.

[0207] In this embodiment, during actual execution, the intelligent agent routing and task scenario function adaptation can be completed based on task information, such as the specific content of the question or statement entered by the user.

[0208] For example, the task information could be the user's input of "What ingredients are in the refrigerator?" or "What ingredients are in the refrigerator? Please announce them via voice."

[0209] After obtaining the task information, the agent is optimally routed to find an agent that can analyze the task information, thereby adapting the task scenario and generating the scenario function of the question-and-answer task. This enables the reflective agent to output the task result corresponding to the task information based on the adapted scenario.

[0210] According to the control method of the refrigeration equipment provided in the embodiments of this application, the optimal routing and task scenario function adaptation are performed through task information, effectively determining the intelligent agent that can execute the task information, improving task execution efficiency, and thus outputting task information based on the actual application scenario, improving the accuracy of task execution, and enhancing the user experience.

[0211] like Figure 6 As shown, in some embodiments, determining the output method corresponding to the task scenario based on task information may further include:

[0212] Train a generative network model based on task information;

[0213] Based on the model parameters corresponding to the generative network model, the network model is used to generate inference tasks.

[0214] Based on the reasoning task, determine the output method corresponding to the task scenario.

[0215] In this embodiment, the generative network model is a network model that generates new data samples that are similar to real data.

[0216] The model parameters are obtained by training the generative network model.

[0217] In some embodiments, model parameters can also be inference knowledge.

[0218] The reasoning task is to verify the prompting reasoning based on the model parameters obtained during the training phase of the generative network model and to determine the logical relationships of the prompts.

[0219] In practical implementation, a prompting engineering reasoning method based on Generative Adversarial Networks (GANs) can be constructed to complete the prompting engineering reasoning task. On the one hand, the prompting engineering reasoning method based on GANs uses the scoring method or the scoring result obtained based on the scoring method as the judgment condition; the discriminant network model verifies the prompting reasoning and judges the prompting logical relationship based on the judgment condition. On the other hand, the prompting engineering reasoning method based on GANs integrates the prompting information compression method. For redundant or excessively long information, the prompting information is compressed to obtain concise and accurate prompting information, so as to determine the output method corresponding to the task scenario based on the prompting information.

[0220] Among them, adversarial networks (GANs) include generative network models, discriminative network models, and variational autoencoder (VAE) generative GANs.

[0221] In actual execution, task information can be used as prompts. During the training phase of the GAN (Generative Adversarial Network), the pre-training of the autonomous generative network model is completed. Then, based on the model parameters or reasoning knowledge generated during the pre-training phase of the generative network model, the discriminative network model is used to verify the prompt reasoning and determine the logical relationship of the prompts, i.e., the discriminative network model performs the reasoning task.

[0222] In some embodiments, the dataset used for pre-training the generated network model can be high-quality data that is unsupervised, semi-supervised, supervised, or self-supervised.

[0223] In this embodiment, by using high-quality data, the prompts can be analyzed more comprehensively, more closely, and more effectively to enhance autonomous reasoning and learning of human reasoning behavior.

[0224] In actual execution, after the discriminant network model generates the inference task, the output method corresponding to the task scenario can be determined based on the actual content of the inference task.

[0225] In actual implementation, it can be done through methods such as Figure 6 The illustrated prompting engineering reasoning method module based on adversarial networks (GANs) processes task information to obtain the reasoning task (i.e., the prompting reasoning result).

[0226] According to the control method of the refrigeration equipment provided in the embodiments of this application, by using task information as prompt content, a generative network model is trained to improve the autonomous reasoning of the prompt content, the comprehensiveness of the learning content, and the closeness to human reasoning behavior. In this way, a reasoning task is generated through a discriminator, and based on the reasoning task, the output mode corresponding to the task scenario is determined, thereby improving the accuracy of the determined output mode and enhancing the user experience in subsequent applications.

[0227] like Figure 7 As shown, in some embodiments, after step 110, the method may further include:

[0228] The reflective agent acquires historical task information; historical task information includes: the historical task and the corresponding historical task results.

[0229] If the first similarity between the task information and the historical tasks is greater than the target similarity threshold, the task result corresponding to the task information is retrieved from the historical task information and the task result corresponding to the task information is output.

[0230] In this embodiment, historical task information refers to task information acquired before the current acquisition time.

[0231] The first similarity is the similarity between the new task information obtained at the current collection time and the previously accumulated historical tasks.

[0232] In some embodiments, the first similarity is calculated based on the contextual similarity between the task information and the historical task and / or the semantic similarity between the task information and the historical task.

[0233] In this embodiment, the task information can be decomposed into multiple smallest task units, and the similarity between the task information and historical tasks can be calculated based on the context information corresponding to the smallest task unit.

[0234] Continue to refer to Figure 7 In some embodiments, calculating the first similarity based on the contextual similarity between task information and historical tasks includes:

[0235] Input the task information into the embedded vector model to obtain the entities corresponding to the task information and the relationships corresponding to the task information.

[0236] The first similarity is determined based on the entities corresponding to the task information, the relationships corresponding to the task information, the vector library, and the knowledge base.

[0237] In actual implementation, the embedded vector model is a model that maps high-dimensional data to a low-dimensional vector space.

[0238] The entities corresponding to task information can be entities such as users, locations, and transactions within the task information.

[0239] The relationships corresponding to task information can be the associations between different entities.

[0240] Vector libraries and knowledge bases can be pre-built based on historical task information or dynamically built.

[0241] In actual execution, after the task information is transformed into a low-dimensional vector through the embedding vector model, the entities and relationships corresponding to the task information are obtained. Other vectors similar to the obtained entities and relationships are searched in the vector library or knowledge base. The context similarity is determined by calculating the similarity between the vectors, that is, the first similarity is obtained.

[0242] According to the control method of the refrigeration equipment provided in the embodiments of this application, the complexity of the task information is effectively reduced by converting the task information into a low-dimensional vector, and the first similarity is effectively determined by calculating the similarity between the converted vector and the vector library or knowledge base, so as to accurately obtain the semantic information of entities and relations in a specific context and improve the accuracy of calculation.

[0243] In actual execution, the semantic similarity between task information and historical tasks can be calculated in any feasible way, and this application does not limit the specific calculation method.

[0244] Of course, in actual implementation, the first similarity can also be calculated in any other feasible way.

[0245] The target similarity threshold is a preset value used to determine whether new task information is approximately the same as historical task information.

[0246] The target similarity threshold can be the average similarity of the target, the maximum similarity of the target, or the minimum similarity of the target. The target similarity threshold can also be determined based on user definition, which is not limited in this application.

[0247] The specific values ​​of the average similarity, maximum similarity, or minimum similarity of the targets can be determined based on the actual execution situation, and this application does not impose any restrictions.

[0248] For example, in actual execution, multiple first similarities can be set together and calculated uniformly to compress the first similarities into a unified measurement method. Based on the compressed data, statistical analysis can be performed to obtain one or more of the target average similarity, target maximum similarity, or target minimum similarity.

[0249] In actual execution, after obtaining the task information, the task information can be broken down into the smallest task units, that is, at least one sub-task text feature can be obtained. The context vector retrieval method is used for each smallest task unit to retrieve historical task information to obtain historical tasks with high similarity to the task information.

[0250] In actual execution, when new task information is obtained, the reflective agent can compare the similarity between the new task information and the previous task information. When the similarity between the new task information and the previous task information reaches the target similarity threshold, the task result corresponding to the historical task can be output to complete the new task information without having to execute the same task again.

[0251] like Figure 7 As shown in the Retrieval-Augmented Generation (RAG) module, in actual execution, when the task information is in the form of text sentences such as questions, table-based question-and-answer, instructions, and queries, the corresponding results can be retrieved through semantic similarity, thereby outputting the required results.

[0252] According to the control method of the refrigeration equipment provided in the embodiments of this application, by comparing the similarity between new task information and historical tasks, task results can be retrieved from historical task information, and task results corresponding to new task information can be efficiently output, thereby improving the task response rate and enhancing the user experience.

[0253] Continue to refer to Figure 8 In some embodiments, the task results can be output using a multimodal large model, such as a text-audio-video large model and an image large model.

[0254] like Figure 8 As shown in the text-audio-video large model module, the text-audio-video large model is used to complete large model architecture construction tasks such as text data encoding and its model fusion module, model structure optimization, and context pruning, such as the task of constructing a large model architecture by combining and optimizing the attention block (Att Block) and the feed-forward network block (FFN Block).

[0255] For different data types, different downstream tasks, and different scenario functions, select the best-performing fusion block. For example, use residual networks, RMS Norm normalization, and multi-head attention mechanisms to form Att blocks for encoding, so as to improve the performance of large models and the accuracy of content generation capabilities.

[0256] Continue to refer to Figure 8The specific execution steps of the text-audio-video large model are as follows: complete the pruning of the embedding layer encoder, which can divide the embedding layer into several sub-blocks, more fine-grained coding units, and effectively extract more detailed granular semantic features.

[0257] For example, pruning is performed on the input and output data of the embedding layer within the neurons. Specifically, RMSNorm normalization is calculated on the input and output results. If the RMSNorm is less than the input or output, a pruning threshold is set. If it is less than the set pruning threshold, the pruning operation is performed. The pruning corresponds to the removal of rows or columns of the transpose matrix, i.e., deletion. In other words, the transpose matrix retains useful parameters while deleting network parameters that have little impact on model pre-training and fine-tuning. This has no impact on large model pre-training and fine-tuning. This results in the embedding layer with important weights. These weights are fully reflected and verified for efficient computation and generalization ability in large model pre-training. There is no residual network, reducing the number of network layers, improving computational efficiency, and improving the performance of this layer.

[0258] The activation layer pruning task is completed, which involves activation layer pruning with multi-head attention mechanism, residual and RMSnorm normalization calculation, i.e., the Att Block fusion block. The RMSnorm normalization calculation module is after the multi-head attention mechanism, i.e., post-Norm. It first identifies unimportant neurons in the output of the large model activation layer and prunes them, i.e., deletes unimportant neurons, such as GeLU or ReLU layers, to reduce the parameters of the large model. This is equivalent to performing compression pruning of the activation layer with multi-head attention mechanism without affecting the model performance and pre-training quality, improving the optimization performance of the internal structure of the large model, and reflecting the further efficiency improvement of parallel computing of the internal structure of the model.

[0259] To construct an FFN Block, first complete N FFN network layers, residuals, and RMSnorm to form the FFN Block. By fusing the FFN Block, network performance is improved and the resources occupied by RMSnorm normalization computation are reduced, which is beneficial for obtaining high-value and useful contextual semantic feature information.

[0260] Complete the activation layer pruning with full multi-head self-attention mechanism, construct AttBlock with residual layers and RMSnorm. The specific process is the same as the activation layer pruning task. In this case, RMSnorm normalization is performed to achieve better data representation features. Also, identify and prune unimportant neurons in the output of large model activation layers, i.e. delete unimportant neurons, such as GeLU or ReLU layers, and output the normalization calculation for the next layer.

[0261] Complete the feedforward FFN layer operation, and on this basis, complete the residual network layer and RMSNorm normalization operation, and finally output the encoded output result.

[0262] Firstly, the residual and RMSorm normalization operations and feedforward network operations are performed on the data input and output results. This can be a fully connected network. Finally, the encoded result is output.

[0263] like Figure 9 As shown in the image large model module, the image large model is used to complete the construction of the image recognition and classification architecture and the LLMAdapter.

[0264] Large image models can include: image recognition classifier-student models, such as distillation, teacher-student models, or quantization techniques, such as food image classifier modules, including images of ingredients, food, beverages, soy sauce, vinegar, etc., thus completing tasks such as building and pre-training large visual models using Vi-transformer-based structural models.

[0265] For example, extracting contextual image information from food images solves and improves the efficiency and accuracy of large-scale visual classifiers.

[0266] The specific steps for large image modeling are as follows:

[0267] Image data is segmented into small regions, which means that the image data is reduced to a two-dimensional data format by word segmentation method and unified into the same two-dimensional data vector space.

[0268] Mapping a two-dimensional vector space to a one-dimensional vector space of the same length is called generating a patch embedding.

[0269] To enable learnable and predictive classification, positional embedding vectors are obtained, which can capture multi-layered sequence information such as position, part-of-speech tag, and word order, providing more information for the classification embedding vectors.

[0270] The obtained vectors are encoded using a transformer encoder block, during which normalization, multi-head attention mechanism and multi-layer perception are performed.

[0271] Multi-layer perceptual classification identifies the category of ingredients, food, or fruits and vegetables.

[0272] In actual implementation, with the text-audio-video large model and the image large model as end-to-end encoding architecture, the backend of the large model also includes a decoder, which completes tasks such as decoding the encoded data, compressing the large model, and adapting to downstream collaboration to improve the composition structure optimization of the large model itself, the complementarity, dependence and correlation of the semantics of pre-trained resources.

[0273] Continue to refer to Figure 10 In some embodiments, the multimodal large model can be optimized and updated to improve the large model's continuous learning and enhanced autonomous learning capabilities.

[0274] like Figure 10 As shown in the multimodal reinforcement learning model training and LLM parameter update module, in the actual execution process, multimodal adaptation can be performed based on the data obtained from the task, and then the score distribution or score sequence of the reward model can be obtained by fine-tuning the large model.

[0275] Secondly, based on the reward model, high-quality datasets are combined with human feedback behaviors, habits, and even hobbies and interests. For example, refrigerator food items are used to create users' question-and-answer history and real-time recorded information, which are then trained through reinforcement learning to generate a large model.

[0276] To generate robust and optimally stable large models that include general network parameters, domain-specific data, and general data.

[0277] In some embodiments, reinforcement learning with proximal policy optimization (PPO) or deep policy optimization (DPO) can be used to enhance the continuous learning and adaptability of multimodal learning, such as text and images.

[0278] In some embodiments, after the multimodal large model is trained and optimized, downstream multitasking tasks such as deployment of the multimodal large model and release of application services can be performed.

[0279] In this embodiment, the deployed device can provide cloud services, including voice, image, text, and video content.

[0280] like Figure 11 As shown in the management modules for model application and deployment, deployment is carried out according to model evaluation, model encapsulation, model plugins, model service activation, interface calling, interface results, and matching downstream tasks (i.e., calling the intelligent agent toolset).

[0281] like Figure 12 As shown, the model can be deployed on a cooling device, which can also receive task information issued by the user through a data channel and a context-aware module.

[0282] The control device for the refrigeration equipment provided in the embodiments of this application, by deploying a trained multimodal large model, realizes a visual, user-friendly, and quantifiable multi-task joint application service interface, thereby improving the user experience.

[0283] In actual implementation, large Transformer models can be built with (1-N) built-in architectures, or with decoder-only architectures, encoder-only architectures, and encoder-decoder architectures, etc. These include, but are not limited to, large models with graph neural networks, attention mechanisms, Transformer models and their variants or improvements, distillation network large models, latent / diffusion large models, U-net networks and other deep network large models, and fused deep network large models, such as Transformer-based BiLSTM+MMCNN-RNN, BiLSTM+MMCNN+Attention, MMCNN+BiLSTM+Attention fused large models, GRU+CNN, GRU+CNN+Attention models, deep reinforcement learning and reward models, RAG's LLM series of neural networks or deep neural network models, as well as Gaussian mixture deep neural network models.

[0284] The control method for refrigeration equipment provided in this application can be executed by a control device for the refrigeration equipment. This application uses the example of a control device for the refrigeration equipment executing the control method to illustrate the control device for the refrigeration equipment provided in this application.

[0285] This application also provides a control device for a refrigeration equipment.

[0286] like Figure 14 As shown, the control device of the refrigeration equipment includes: a first processing module 1410, a second processing module 1420 and a third processing module 1430.

[0287] The first processing module 1410 is used to enable the planning agent to obtain task information input by the user; the task information includes at least one of the following: text information, voice information, image information, and audio-visual information;

[0288] The second processing module 1420 is used to enable the planning agent to decompose the task information, obtain at least one sub-task text feature, and send the at least one sub-task text feature to the reflection agent; each sub-task text feature corresponds to execution order information and weight information.

[0289] The third processing module 1430 is used to enable the reflective agent to output the task result corresponding to the task information based on the text features of at least one sub-task.

[0290] The control device for refrigeration equipment provided in the embodiments of this application acquires various forms of task information, enabling users to interact with the refrigeration equipment in multiple ways. It has a wide range of application scenarios and flexible usage. By setting up multiple agents, the planning agent breaks down the task information into multiple sub-task text features, and the reflection agent outputs the task results corresponding to the task information based on the multiple sub-task texts. This achieves efficient collaborative work between the planning agent and the reflection agent, streamlining the entire development workflow from planning, decision-making, and reflection of the agents. This improves the workflow efficiency between multiple agents, increases the accuracy of the output task results, and enhances the interaction accuracy between the dialogue agent and the user, thereby improving the user experience.

[0291] In some embodiments, the second processing module 1420 may also be used for:

[0292] The task information is converted into text to obtain the first task text;

[0293] Based on queue and stack algorithms, the first task text is split to obtain at least one sub-task text feature with execution order information.

[0294] The text features of at least one subtask are processed using a graph neural network to obtain at least one subtask text feature with corresponding weight information.

[0295] In some embodiments, the second processing module 1420 may also be used for:

[0296] If the task information includes voice information, the voice information is converted into first text information;

[0297] When the task information includes image information, image features are extracted from the image information to obtain the image features corresponding to the image information, and the image features are converted into text descriptions.

[0298] The planning agent obtains the first task text based on at least one of the first text information and the text description.

[0299] In some embodiments, the third processing module 1430 can also be used for:

[0300] The reflective agent outputs the task results corresponding to the task information based on the context corresponding to the text features of each subtask, the contextual relationships corresponding to the text features of each subtask, and the multi-step reasoning information corresponding to the text features of each subtask.

[0301] In some embodiments, the cooling device further includes: a memory agent; the third processing module 1430 can also be used for:

[0302] The memory agent constructs the target relationship between the text features of each subtask based on the context corresponding to the text features of each subtask, the contextual relationship corresponding to the text features of each subtask, and the multi-step reasoning information corresponding to the text features of each subtask.

[0303] The reflective agent outputs the task results corresponding to the task information based on the target relationship.

[0304] In some embodiments, the third processing module 1430 can also be used for:

[0305] Based on the target relationship, determine the current subtask and the previous subtask corresponding to the current subtask;

[0306] Based on the execution result of the previous subtask and the execution content of the current subtask, obtain the execution result of the current subtask.

[0307] If the current subtask is the last subtask among all subtasks included in the target relationship, output the task result corresponding to the task information.

[0308] In some embodiments, the third processing module 1430 can also be used for:

[0309] Based on the task information, determine the output method corresponding to the task scenario;

[0310] Output the task results corresponding to the task information based on the output method corresponding to the task scenario.

[0311] In some embodiments, the third processing module 1430 can also be used for:

[0312] Train a generative network model based on task information;

[0313] Based on the model parameters corresponding to the generative network model, the network model is used to generate inference tasks.

[0314] Based on the reasoning task, determine the output method corresponding to the task scenario.

[0315] In some embodiments, the device may further include a fourth processing module:

[0316] The reflective agent acquires historical task information; historical task information includes: the historical task and the corresponding historical task results.

[0317] If the first similarity between the task information and the historical task is greater than the target similarity threshold, the task result corresponding to the task information is retrieved from the historical task information and the task result corresponding to the task information is output.

[0318] The control device for the refrigeration equipment in this application embodiment can be an electronic device or a component within an electronic device, such as an integrated circuit or a chip. The electronic device can be a terminal or other devices besides a terminal. For example, the electronic device can be a mobile phone, tablet computer, laptop computer, PDA, in-vehicle electronic device, mobile internet device (MID), augmented reality (AR) / virtual reality (VR) device, robot, wearable device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), etc. It can also be a server, network attached storage (NAS), personal computer (PC), television (TV), ATM, or self-service machine, etc. This application embodiment does not specifically limit the scope of the device.

[0319] The control device for the refrigeration equipment in this application embodiment can be a device with an operating system. This operating system can be Android, iOS, or other possible operating systems; this application embodiment does not specifically limit the specific operating system used.

[0320] The control device for the refrigeration equipment provided in this application embodiment can achieve... Figures 1 to 13 The various processes implemented in the method implementation examples will not be described again here to avoid repetition.

[0321] This application also provides a refrigeration device.

[0322] In some embodiments, the refrigeration device includes at least one intelligent agent based on the control method for the refrigeration device as described in any of the above embodiments.

[0323] In this embodiment, the refrigeration equipment can be understood as a refrigeration storage device in a broad sense, including but not limited to refrigerators, freezers, display cases, beverage cabinets, wine cabinets, refrigerated display cases, and refrigerated vending machines. Refrigeration equipment has diverse structural forms and a wide range of applications.

[0324] like Figure 14 As shown, taking a refrigerator as an example, the refrigeration equipment can include multiple layers, including a freezing layer, a refrigeration layer, and a constant temperature layer.

[0325] Refrigeration equipment can also be equipped with human-computer interaction objects, including: gestures, voice, text, audio and video, image acquisition, microphone arrays, radar signal acquisition, OCR devices, documents, VR and AR series.

[0326] The input and output formats of refrigeration equipment include APP, PC, web, mini-program, and large screen.

[0327] The refrigeration equipment can support protocols such as TCP / IP, Wi-Fi, and Bluetooth.

[0328] The refrigeration equipment can support networks such as 5G, Wi-Fi, wired network, and 6G.

[0329] Applications of refrigeration equipment include, but are not limited to: household, hotel, industrial, indoor and outdoor, military and aerospace.

[0330] The embedding methods for refrigeration equipment can include: flush mounting, vehicle-mounted embedding, and integration.

[0331] In some embodiments, at least one agent may further include: a planning agent and a reflecting agent.

[0332] In this embodiment, the planning agent breaks down the task information to obtain at least one sub-task.

[0333] Reflect on the task results corresponding to the output of the intelligent agent and the task information.

[0334] The planning agent and the reflecting agent are electrically connected.

[0335] In some embodiments, the reflective agent may further include a memory agent.

[0336] In this embodiment, the memory agent constructs target relationships between subtasks based on at least one subtask.

[0337] According to the refrigeration device provided in the embodiments of this application, by acquiring task information in various forms, users can interact with the refrigeration device in multiple ways, with wide application scenarios and flexible usage methods. By setting up multiple intelligent agents, the planning intelligent agent breaks down the task information to obtain multiple sub-task text features, and the reflection intelligent agent outputs the task results corresponding to the task information based on the multiple sub-task texts, realizing efficient collaborative work between the planning intelligent agent and the reflection intelligent agent. The entire lifecycle of intelligent agent development workflow from planning, decision-making and reflection is streamlined, improving the workflow efficiency between multiple intelligent agents, improving the accuracy of the output task results, improving the interaction accuracy between the dialogue intelligent agent and the user, thereby improving the user experience.

[0338] In some embodiments, such as Figure 15As shown, this application embodiment also provides an electronic device 1500, including a processor 1501, a memory 1502, and a computer program stored in the memory 1502 and executable on the processor 1501. When the program is executed by the processor 1501, it implements the various processes of the above-described control method embodiment for the refrigeration device and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0339] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.

[0340] This application also provides a non-transitory computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the various processes of the control method embodiment of the above-described refrigeration device and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0341] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0342] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the control method for the above-described refrigeration device.

[0343] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0344] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the various processes of the above-described control method embodiment for the refrigeration device, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0345] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0346] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0347] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the related technology, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0348] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

[0349] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0350] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

Claims

1. A control method of a refrigeration apparatus, characterized by, The refrigeration equipment comprises a planning agent and a reflection agent; the method comprises: The planning agent acquires user-inputted task information; the task information comprises at least one of text information, voice information, image information and audio-video information; The planning agent splits the task information to obtain at least one sub-task text feature, and sends the at least one sub-task text feature to the reflection agent; each sub-task text feature corresponds to execution sequence information and weight information; The reflection agent outputs a task result corresponding to the task information based on the at least one sub-task text feature.

2. The control method of a refrigerating appliance according to claim 1, characterized in that, The planning agent splits the task information to obtain at least one sub-task text feature, comprising: Text conversion is performed on the task information to obtain first task text; The first task text is split based on a queue algorithm and a stack algorithm to obtain the at least one sub-task text feature corresponding to the execution sequence information; The at least one sub-task text feature is processed based on a graph neural network to obtain the at least one sub-task text feature corresponding to the weight information.

3. The control method of a refrigerating appliance according to claim 2, characterized in that, The planning agent performs text conversion on the task information to obtain first task text, comprising: In the case that the task information comprises the voice information, the voice information is converted into first text information; In the case that the task information comprises the image information, image feature extraction is performed on the image information to obtain image features corresponding to the image information, and the image features are converted into a text description; The first task text is obtained based on at least one of the first text information and the text description.

4. The control method of a refrigeration appliance according to any one of claims 1-3, characterized in that, The reflection agent outputs a task result corresponding to the task information based on the at least one sub-task text feature, comprising: The reflection agent outputs the task result corresponding to the task information based on context corresponding to each sub-task text feature, context relationship corresponding to each sub-task text feature and multi-step reasoning information corresponding to each sub-task text feature.

5. The control method of a refrigerating appliance according to claim 4, characterized in that, The refrigeration equipment further comprises a memory agent; the reflection agent outputs a task result corresponding to the task information based on context corresponding to each sub-task text feature, context relationship corresponding to each sub-task text feature and multi-step reasoning information corresponding to each sub-task text feature, comprising: The memory agent constructs target relationships between each sub-task text feature based on context corresponding to each sub-task text feature, context relationship corresponding to each sub-task text feature and multi-step reasoning information corresponding to each sub-task text feature; The reflection agent outputs the task result corresponding to the task information based on the target relationships.

6. The control method of a refrigerating appliance according to claim 5, characterized in that, The reflection agent outputs the task result corresponding to the task information based on the target relationships, comprising: Based on the target relationships, a current sub-task and a previous sub-task corresponding to the current sub-task are determined; obtaining an execution result corresponding to the current subtask based on the execution result corresponding to the previous subtask and the execution content corresponding to the current subtask; in a case where the current subtask is the last subtask in all subtasks included in the target relationship, outputting a task result corresponding to the task information.

7. The control method of a refrigerating appliance according to any one of claims 1-3, characterized in that, The reflective agent outputs the task result corresponding to the task information based on the at least one subtask text feature, including: determining an output mode corresponding to a task scenario based on the task information; outputting the task result corresponding to the task information based on the output mode corresponding to the task scenario.

8. The control method of a refrigerating appliance according to claim 7, characterized in that, The method further includes: training a generative network model based on the task information; determining the output mode corresponding to the task scenario based on the inference task. After obtaining the task information input by the user, the method further includes:

9. The control method of a refrigerating appliance according to any one of claims 1-3, characterized in that, The reflective agent obtains historical task information; the historical task information includes: historical tasks and historical task results corresponding to the historical tasks; in a case where a first similarity between the task information and the historical task is greater than a target similarity threshold, retrieving a task result corresponding to the task information from the historical task information to output the task result corresponding to the task information. The first similarity is calculated based on a context similarity between the task information and the historical task and / or a semantic similarity between the task information and the historical task. The refrigeration device includes a planning agent and a reflective agent; the device includes:

10. The control method of a refrigerating appliance according to claim 9, characterized in that, The first processing module is configured to cause the planning agent to obtain task information input by a user; the task information includes at least one of text information, voice information, image information, and audio / video information; 11. A control device of a refrigerating apparatus, characterized by comprising: The second processing module is configured to cause the planning agent to split the task information to obtain at least one subtask text feature, and send the at least one subtask text feature to the reflective agent; each subtask text feature corresponds to execution order information and weight information; The third processing module is configured to cause the reflective agent to output a task result corresponding to the task information based on the at least one subtask text feature. including: at least one agent involved in the control method of the refrigeration device according to any one of claims 1-10.

12. A refrigeration appliance characterized by, 13. The refrigeration device of claim 12, the at least one agent, including: a planning agent configured to split task information to obtain at least one subtask text feature; a reflective agent configured to output a task result corresponding to the task information; the planning agent and the reflective agent are electrically connected.

14. The refrigeration device of claim 13, the reflective agent, further including: a memory agent configured to construct a target relationship between each subtask based on the at least one subtask. ​ ​ 15. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, which the processor executes, implements the method of the refrigeration appliance as claimed in any of claims 1-10.

16. A computer program product comprising a computer program, characterized in that, The computer program, which the processor executes, implements the control method of the refrigeration appliance as claimed in any of claims 1-10.

Images