Refrigeration equipment interaction method, device and system and refrigeration equipment

Abstract

The invention discloses an interaction method, device and system of refrigeration equipment and the refrigeration equipment, and belongs to the technical field of refrigeration equipment. The method comprises the following steps: receiving first input of a user; in response to the first input, determining target interaction data of the user and refrigeration equipment; querying a target interaction intention corresponding to the target interaction data through a routing mechanism; based on the target interaction intention and execution information of the refrigeration equipment, an interaction task corresponding to the target interaction intention is determined, and the execution information comprises at least one of an execution behavior, an execution plan and an execution memory of the refrigeration equipment; and executing the interaction task through a target execution module corresponding to the interaction task. According to the method, the accuracy of the interaction information fed back to the user can be improved, and the interaction effect is improved.

Description

Technical Field

[0001] This application belongs to the field of refrigeration equipment technology, and particularly relates to an interaction method, device, system, and refrigeration equipment. Background Technology

[0002] Smart refrigerators not only have the preservation function of traditional refrigerators, but also integrate technologies such as the Internet and artificial intelligence to interact with users. Through remote control by mobile phone and voice interaction, smart refrigerators can provide users with services such as diet advice and health management, bringing users a more convenient and intelligent life experience.

[0003] Currently, during the interaction between users and smart refrigerators, the accuracy of downstream task matching is low, and smart refrigerators often cannot accurately provide the interactive information that users want, resulting in poor interaction effects. Summary of the Invention

[0004] This application aims to address at least one of the technical problems existing in the prior art. To this end, this application proposes an interaction method, apparatus, system, and refrigeration equipment for cooling devices, which can improve the accuracy of interactive information fed back to the user and enhance the interaction effect.

[0005] In a first aspect, this application provides an interaction method for a refrigeration device, the method comprising:

[0006] Receive the user's first input;

[0007] In response to the first input, the target interaction data between the user and the refrigeration device is determined;

[0008] The target interaction intent corresponding to the target interaction data is queried through the routing mechanism;

[0009] Based on the target interaction intent and the execution information of the refrigeration device, the interaction task corresponding to the target interaction intent is determined. The execution information includes at least one of the execution behavior, execution plan, and execution memory of the refrigeration device.

[0010] The interactive task is executed by the target execution module corresponding to the interactive task.

[0011] According to the interaction method of the refrigeration equipment in this application, the target interaction intent corresponding to the target interaction data is matched by the rules and algorithms preset by the routing mechanism. The target interaction intent is accurate. Based on the target interaction intent and the execution behavior, execution plan and execution memory of the refrigeration equipment, the interaction task is reasonably determined and the interaction task is executed by the target execution module. This can improve the accuracy of the interaction information fed back to the user and enhance the interaction effect.

[0012] According to one embodiment of this application, querying the target interaction intent corresponding to the target interaction data through a routing mechanism includes:

[0013] By using the routing mechanism and the attention mechanism, the target interaction intent corresponding to the target interaction data is queried.

[0014] According to one embodiment of this application, determining the interaction task corresponding to the target interaction intent based on the target interaction intent and the execution information of the cooling device includes:

[0015] Based on the target interaction intent and the execution information, the interaction task corresponding to the target interaction intent is determined through the thought chain.

[0016] According to one embodiment of this application, the execution memory is formed through the following steps:

[0017] Obtain the interaction content between the user and the refrigeration equipment;

[0018] Based on the time information, spatial information and frequency information of the interactive content, calculate the memory value corresponding to the interactive content;

[0019] The execution memory is formed based on the interactive content and the memory value corresponding to the interactive content.

[0020] According to one embodiment of this application, executing the interactive task through the target execution module corresponding to the interactive task includes:

[0021] The interactive task is input into the encoding sub-model of the encoding / decoding model, and the interactive task is encoded by the encoding sub-model to obtain the encoded interactive task.

[0022] The encoding interaction task is input into the decoding sub-model of the encoding / decoding model, and the encoding interaction task is decoded by the decoding sub-model to obtain the decoding interaction task;

[0023] Based on the decoding interaction task, the target execution module is determined, and the interaction task is executed through the target execution module.

[0024] According to one embodiment of this application, encoding the interaction task through the encoding sub-model to obtain an encoded interaction task includes:

[0025] In the encoder network of the encoding sub-model, the target network layer and the pruning conditions of the target network layer are determined;

[0026] Based on the pruning conditions, the target network layer is pruned, and the interaction task is encoded through the pruned encoder network to obtain the encoded interaction task.

[0027] According to one embodiment of this application, the encoding / decoding model is trained through the following steps:

[0028] Obtain the supervised dataset for the encoding / decoding model;

[0029] Based on the embedded encoding of the encoder output in the encoding sub-model, the encoding and decoding model is fine-tuned to determine the score data;

[0030] Based on the score data and the supervised dataset, the encoding / decoding model is trained through reinforcement learning.

[0031] Secondly, this application provides an interaction device for a refrigeration device, the device comprising:

[0032] The first receiving module is used to receive the user's first input;

[0033] A first processing module is configured to, in response to the first input, determine the target interaction data between the user and the refrigeration device;

[0034] The second processing module is used to query the target interaction intent corresponding to the target interaction data through a routing mechanism;

[0035] The third processing module is used to determine the interaction task corresponding to the target interaction intent based on the target interaction intent and the execution information of the refrigeration device. The execution information includes at least one of the execution behavior, execution plan and execution memory of the refrigeration device.

[0036] The fourth processing module is used to execute the interactive task through the target execution module corresponding to the interactive task.

[0037] According to the interaction device of the refrigeration equipment of this application, the target interaction intent corresponding to the target interaction data is matched by the rules and algorithms preset by the routing mechanism. The target interaction intent is accurate. Based on the target interaction intent and the execution behavior, execution plan and execution memory of the refrigeration equipment, the interaction task is reasonably determined and the interaction task is executed by the target execution module. This can improve the accuracy of the interaction information fed back to the user and enhance the interaction effect.

[0038] Thirdly, this application provides a refrigeration device, comprising:

[0039] The interactive device of the refrigeration equipment as described in the second aspect above;

[0040] A data acquisition device is connected to the interaction device and is used to collect interaction data between the user and the refrigeration equipment.

[0041] According to the refrigeration device of this application, the target interaction intent corresponding to the target interaction data is matched by the rules and algorithms preset by the routing mechanism. The target interaction intent obtained is accurate. Based on the target interaction intent and the execution behavior, execution plan and execution memory of the refrigeration device, the interaction task is reasonably determined and the interaction task is executed by the target execution module. This can improve the accuracy of the interaction information fed back to the user and enhance the interaction effect.

[0042] Fourthly, this application provides an interactive system for a refrigeration device, comprising:

[0043] The refrigeration equipment described in the third aspect above;

[0044] The server is connected to the cooling device and is used to output the interaction results obtained by the interactive device in performing the interactive task to the user.

[0045] According to the interaction system of the refrigeration equipment of this application, the target interaction intent corresponding to the target interaction data is matched by the rules and algorithms preset by the routing mechanism. The target interaction intent is accurate. Based on the target interaction intent and the execution behavior, execution plan and execution memory of the refrigeration equipment, the interaction task is reasonably determined and the interaction task is executed by the target execution module. This can improve the accuracy of the interaction information fed back to the user and enhance the interaction effect.

[0046] Fifthly, this application provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the interaction method of the cooling device as described in the first aspect above.

[0047] In a sixth aspect, this application provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the interaction method of the refrigeration device as described in the first aspect above.

[0048] In a seventh aspect, this application provides a computer program product, including a computer program that, when executed by a processor, implements the interaction method of the refrigeration device as described in the first aspect above.

[0049] 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

[0050] 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:

[0051] Figure 1This is a flowchart illustrating the interaction method of the refrigeration equipment provided in the embodiments of this application;

[0052] Figure 2 This is a schematic diagram of the structure of a large model of the refrigeration equipment provided in the embodiments of this application;

[0053] Figure 3 This is a schematic diagram of the structure of the intelligent agent application module provided in the embodiments of this application;

[0054] Figure 4 This is a schematic diagram of the structure of the codec network provided in the embodiments of this application;

[0055] Figure 5 This is a schematic diagram of the structure of the model service module provided in the embodiments of this application;

[0056] Figure 6 This is a schematic diagram of the structure of the refrigeration equipment provided in the embodiments of this application;

[0057] Figure 7 This is a schematic diagram of the structure of the interactive device of the refrigeration equipment provided in the embodiments of this application;

[0058] Figure 8 This is a signaling flowchart of the interaction system of the refrigeration equipment provided in the embodiments of this application;

[0059] Figure 9 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0060] 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.

[0061] 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.

[0062] The following description, in conjunction with the accompanying drawings, details the interaction method, interaction device, refrigeration equipment, interaction system, electronic device, and readable storage medium of the refrigeration equipment provided in this application, through specific embodiments and application scenarios.

[0063] The interaction method of the refrigeration equipment can be applied to the terminal, and can be executed by the hardware or software in the terminal.

[0064] The terminal includes, but is not limited to, portable communication devices such as mobile phones or tablets with touch-sensitive surfaces (e.g., touchscreen displays and / or touchpads). It should also be understood that, in some embodiments, the terminal may not be a portable communication device, but rather a desktop computer with touch-sensitive surfaces (e.g., touchscreen displays and / or touchpads).

[0065] 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.

[0066] The interaction method of the refrigeration device provided in this application embodiment can be executed by an electronic device or a functional module or functional entity in an electronic device that can implement the interaction method of the refrigeration device. The electronic devices mentioned in this application embodiment include, but are not limited to, mobile phones, tablets, computers, cameras, touch devices and wearable devices. The interaction method of the refrigeration device provided in this application embodiment will be described below with an electronic device as the execution subject as an example.

[0067] It should be noted that the refrigeration equipment in the embodiments of this application can be understood as a broad refrigeration storage device, including but not limited to refrigerators, freezers, display cases, beverage cabinets, wine cabinets, refrigerated display cases, and refrigerated vending machines, etc. Refrigeration equipment has diverse structural forms and a wide range of applications.

[0068] The interaction method of the refrigeration device in the embodiments of this application can be achieved through, for example... Figure 2 The large-scale model of the refrigeration equipment shown is implemented.

[0069] like Figure 1 As shown, the interaction method of the refrigeration equipment includes steps 110-150.

[0070] Step 110: Receive the user's first input.

[0071] The first input is used for user interaction with the refrigeration equipment, and the first input can be in at least one of the following ways:

[0072] Firstly, the first input can be a touch operation, including but not limited to click, swipe, and press operations.

[0073] In this embodiment, receiving the user's first input can be receiving the user's touch operation on the display area of ​​the terminal display screen of the refrigeration device.

[0074] To reduce the rate of user error, the effective area of ​​the first input can be limited to a specific area, such as the upper middle area of ​​the touch screen interface of a refrigeration device; or, while the interactive interface is displayed, the target control can be displayed on the current interface, and touching the target control will enable the first input; or the first input can be set to a series of taps on the display area within a target time interval.

[0075] Secondly, the first input can be a physical button input.

[0076] In this embodiment, the terminal of the refrigeration device is provided with corresponding physical buttons on its body to receive the user's first input, which can be receiving the user's operation of pressing the corresponding physical button; the first input can also be a combination operation of pressing multiple physical buttons at the same time.

[0077] Thirdly, the first input can be voice input.

[0078] In this embodiment, the terminal of the cooling device can receive the user's voice input through a voice acquisition device such as a microphone.

[0079] Of course, in other embodiments, the first input may also be in other forms, including but not limited to character input, image input, audio and video input, etc., which can be determined according to actual needs, and this application embodiment does not limit it.

[0080] Step 120: In response to the first input, determine the target interaction data between the user and the refrigeration equipment.

[0081] In this embodiment, after receiving the first input, the terminal of the refrigeration device can respond to the first input and determine the target interaction data between the user and the refrigeration device.

[0082] Among them, the target interaction data is the data generated during the interaction between the user and the refrigeration equipment based on the user's first input in order to achieve the interaction goal. The target interaction data corresponds to the first input.

[0083] For example, if the first input is voice input, the corresponding target interaction data is voice interaction data; if the first input is text input, the corresponding target interaction data is text interaction data; if the first input is image input, the corresponding target interaction data is image interaction data; if the first input is audio / video input, the corresponding target interaction data is audio / video interaction data.

[0084] In this embodiment, the content of the target interactive data can be the same as the content of the first input, or the content of the target interactive data can be the content after processing the first input.

[0085] Step 130: Query the target interaction intent corresponding to the target interaction data through the routing mechanism.

[0086] Among them, the routing mechanism is the mechanism for allocating information flow in deep learning to ensure that data can reach the appropriate processing unit.

[0087] Target interaction intent is used to represent the user's interaction goal, such as asking how many days until the food in the refrigeration equipment expires, expecting the refrigeration equipment to provide recipe suggestions, or instructing the refrigeration equipment to turn off the lights.

[0088] In this embodiment, a routing mechanism is used to parse and understand the target interaction intent corresponding to the target interaction data. The routing mechanism can determine the target interaction intent based on the target interaction data through predefined rules, algorithms, or architectures.

[0089] For example, the routing mechanism can extract feature information from the target interaction data, identify the user's initial intent corresponding to the target interaction data based on the feature information, match the user's initial intent with the preset interaction intent corresponding to the cooling device, and determine the preset interaction intent that is closest to the user's initial intent as the target interaction intent.

[0090] For example, routing mechanisms can also be combined with natural language processing techniques to perform semantic analysis, sentiment analysis, etc., to obtain the target interaction intent. Routing mechanisms can also be combined with pattern matching, machine learning, etc., to determine the target interaction intent based on the target interaction data.

[0091] In this embodiment, the routing mechanism can be a semantic routing mechanism.

[0092] Step 140: Based on the target interaction intent and the execution information of the refrigeration device, determine the interaction task corresponding to the target interaction intent.

[0093] Among them, the execution information of the refrigeration equipment is the information corresponding to the refrigeration equipment when performing interactive tasks. The execution information includes at least one of the refrigeration equipment's execution behavior, execution plan, and execution memory.

[0094] The execution behavior can include the interactive tasks that the refrigeration equipment is currently performing and the interactive tasks that the refrigeration equipment can perform. The execution plan can include the order and execution time of the interactive tasks that the refrigeration equipment will perform. The execution memory can include information corresponding to the interactive tasks performed by the refrigeration equipment at historical moments.

[0095] In this embodiment, the interaction task corresponding to the target interaction intent is the downstream task executed by the refrigeration device in the process of realizing the target interaction intent.

[0096] In this step, the execution information can be used to coordinate with other interactive tasks, and the interactive task corresponding to the target interactive intent can be determined based on the experience information in historical interactive tasks.

[0097] Step 150: Execute the interactive task through the target execution module corresponding to the interactive task.

[0098] Among them, the target execution module is the module that executes the interaction task corresponding to the target interaction intent. There can be one or more target execution modules.

[0099] In this embodiment, after determining the interaction task corresponding to the target interaction intent, the target execution module can be determined according to the correspondence between the interaction task and the execution module. The target execution module then executes the interaction task corresponding to the target interaction intent to realize the interaction with the user.

[0100] According to the interaction method of the refrigeration equipment provided in the embodiments of this application, the target interaction intent corresponding to the target interaction data is matched by the rules and algorithms preset by the routing mechanism. The target interaction intent is accurate. Based on the target interaction intent and the execution behavior, execution plan and execution memory of the refrigeration equipment, the interaction task is reasonably determined and the interaction task is executed by the target execution module. This can improve the accuracy of the interaction information fed back to the user and enhance the interaction effect.

[0101] In some embodiments, querying the target interaction intent corresponding to the target interaction data through a routing mechanism includes:

[0102] By using a routing mechanism combined with an attention mechanism, the target interaction intent corresponding to the target interaction data can be queried.

[0103] The attention mechanism is a structure embedded in machine learning models, used to automatically learn and calculate the contribution of input data to output data.

[0104] In this embodiment, an attention mechanism is incorporated into the routing mechanism, that is, the attention mechanism is applied between the various nodes and modules of the routing mechanism, so that the various nodes and modules of the routing mechanism can interact and merge information, thereby better understanding the overall semantic structure of the target interactive data.

[0105] In practice, attention mechanisms can be incorporated into the routing mechanism through methods such as dot product operations.

[0106] In some embodiments, the interaction task corresponding to the target interaction intent is determined based on the target interaction intent and the execution information of the cooling device, including:

[0107] Based on the target interaction intent and execution information, the interaction task corresponding to the target interaction intent is determined through the thought chain.

[0108] Among them, the Mind Chain is an improved prompting technology that allows large models to progressively participate in the process of breaking down a complex problem into subproblems and solving them sequentially.

[0109] In this embodiment, based on the target interaction intent and the execution information of the cooling device, the thought chain can deduce the interaction task corresponding to the target interaction intent through a series of logical reasoning steps.

[0110] In some embodiments, execution memory is formed through the following steps:

[0111] Obtain the interaction content between the user and the refrigeration equipment;

[0112] Calculate the memory value corresponding to the interactive content based on the time, space and frequency information of the interactive content;

[0113] Execution memory is formed based on the interactive content and the corresponding memory value.

[0114] The user in the interaction content between the user and the refrigeration equipment can be the same user or different users as the user in the first input received by the user.

[0115] The interaction content between the user and the refrigeration equipment can be the data content obtained by the refrigeration equipment from the user's interaction behavior.

[0116] In this embodiment, the interactive content may include voice data, image data, text data, and audio / video data.

[0117] In this embodiment, the cooling device can be equipped with a data acquisition device, such as an image acquisition device, a voice acquisition device, or a text input device, which is used to collect corresponding interactive data.

[0118] In this embodiment, the time information of the interaction content can represent the time and duration of the interaction between the user and the cooling device, the spatial information of the interaction content can represent the relative spatial position between the user and the cooling device, and the frequency information of the interaction content can represent the frequency of the user's interaction with the cooling device within a certain period of time, such as within a day.

[0119] In this embodiment, the memory value can be calculated based on the time, space, and frequency information of the interactive content using a preset memory value calculation formula.

[0120] For example, the time, space and frequency information of the interactive content can be normalized, and then a weighted sum can be obtained to obtain the memory value, which can then form a sequence.

[0121] In this embodiment, the obtained memory value can be used to form an average memory value, i.e., to form an execution memory. The memory value can be compared with the maximum and minimum values ​​of a preset memory value range, and the execution memory can be formed based on whether the memory value is within the preset memory value range. Alternatively, the memory value can be compared with a preset threshold, and the execution memory can be formed based on the comparison result with the preset threshold. Or, the memory value can be compared with a residual memory value, and the execution memory can be formed based on the comparison result with the residual memory value.

[0122] For example, the memory value corresponding to the interactive content can be compared with a preset threshold. When the memory value is less than the preset threshold, the interactive content is formed into short-term execution memory, and when the memory value is greater than the preset threshold, the interactive content is formed into long-term execution memory.

[0123] Short-term execution memory is used by large models to store and process immediate information in a short period of time, while long-term execution memory is used by large models to store and retrieve information over a long period of time.

[0124] In some embodiments, the interactive task is executed through the target execution module corresponding to the interactive task, including:

[0125] The interactive task is input into the encoding sub-model of the encoding / decoding model, and the interactive task is encoded by the encoding sub-model to obtain the encoded interactive task.

[0126] The encoding interaction task is input into the decoding sub-model of the encoding / decoding model. The decoding sub-model decodes the encoding interaction task to obtain the decoding interaction task.

[0127] Based on the decoding interaction task, the target execution module is determined, and the interaction task is executed through the target execution module.

[0128] Among them, the encoding and decoding model is a model within the larger model of the refrigeration equipment, the encoding and decoding model is a model that can encode and decode data, the encoding sub-model is a model within the encoding and decoding model that can encode data, and the decoding sub-model is a model within the encoding and decoding model that can decode data.

[0129] In this embodiment, the encoding sub-model encodes the interaction task to obtain the encoded interaction task, which can improve the interpretability of the data corresponding to the interaction task and reduce data redundancy. The decoding sub-model decodes the encoded interaction task to obtain the decoded interaction task. Based on the decoded interaction task, the target execution module is determined, which can improve the understanding of data in the process of determining the target execution module, increase the speed of determining the target execution module, and make the obtained target execution module more accurate.

[0130] In some embodiments, the interaction task is encoded through an encoding sub-model to obtain an encoded interaction task, including:

[0131] In the encoder network of the encoding sub-model, determine the target network layer and the pruning conditions of the target network layer;

[0132] Based on the pruning conditions, the target network layer is pruned, and the interactive task is encoded through the pruned encoder network to obtain the encoded interactive task.

[0133] The encoder network can be one of the networks in the aforementioned encoding sub-model. The encoder network can perform operations such as data encoding, data fusion, large model structure optimization, and large model context pruning.

[0134] like Figure 4 As shown, the encoder network is obtained by combining and optimizing the attention block (Att Block) and the feedforward neural block (FFN Block).

[0135] In this embodiment, for different data types, decision tasks, decision indicators, downstream task categories, application scenario categories, and functional requirements, the fusion block corresponding to the attention block and feedforward neural block with the best effect is selected. This includes using a residual network and a normalization layer for normalization calculation and a multi-head attention mechanism to form an Att Block for encoding, which is beneficial to improving the performance of the encoder network.

[0136] The normalization layer can be a root mean square layer normalization (RMSNorm) layer.

[0137] The pruning condition can be that the normalized value of the input / output of the target network layer is less than the set pruning threshold.

[0138] like Figure 4As shown, the input layer embeds pruning at the input position of the data to be processed. The Multi-Head Self-Attention (MHSA) pruning layer, i.e., the MHSA-pruning layer, is used as the embedding layer. Based on the input and output of the embedding layer, pruning is performed inside the neurons in the encoder network. The normalization of each neuron can be calculated by the RMSnorm layer on the input and output data, and it is determined whether the obtained normalized value is less than the set pruning threshold. If it is less than the pruning threshold, the pruning operation is performed, deleting the row or column of the transpose matrix corresponding to the pruning. This can retain useful parameters in the transpose matrix, delete network parameters that have little impact on the pre-training and fine-tuning of the encoder network, and obtain the embedding layer with important weights. These weights can improve the pre-training efficiency and generalization ability of the encoder network. No residual network is set in the encoder network, which can reduce the number of network layers, improve computational efficiency, and make the performance of the embedding layer better.

[0139] The target network layer can be an MHSA-pruning layer.

[0140] In this embodiment, RMSNorm normalization is performed to obtain better data representation features. The RMSNorm normalization calculation module, after the multi-head attention mechanism (post-Norm), identifies unimportant neurons in the output of the activation layer in the encoder network and performs pruning, deleting these unimportant neurons, such as GeLU or ReLU layers, and outputs the normalization calculation for the next layer. This reduces the parameters of the encoder network, which is equivalent to performing compression and pruning of the activation layer with the multi-head attention mechanism without affecting the performance and pre-training quality of the encoder network. By changing the internal structure of the encoder network, performance optimization is achieved, reflecting the parallel computation of the internal structure of the encoder network and further improving efficiency.

[0141] In this embodiment, pruning the target network layer in the encoder network according to the pruning conditions can reduce the complexity of the encoder network and improve its generalization ability.

[0142] In some embodiments, the encoding / decoding model is trained through the following steps:

[0143] Obtain the supervised dataset for the encoding / decoding model;

[0144] Based on the embedded encoding of the encoder output in the encoding sub-model, the encoding and decoding model is fine-tuned to determine the score data;

[0145] Based on the score data and supervised dataset, the encoding and decoding model is trained through reinforcement learning.

[0146] The supervised dataset is a dataset used for supervised training of the encoding and decoding model. The data in the supervised dataset can be generated by a large model of the refrigeration equipment or obtained based on existing data.

[0147] In this embodiment, the data is processed by embedding an encoder in the encoding sub-model to obtain embedded encoding. The embedded encoding may include key features and contextual information of the data. The embedded encoding may be in the form of a vector. The pre-trained encoding and decoding model is further trained through the embedded encoding to achieve fine-tuning of the encoding and decoding model.

[0148] Fine-tuning the codec model allows you to verify its performance on the corresponding validation set and obtain the corresponding score data, which can characterize the performance of the codec model.

[0149] In this embodiment, the encoding / decoding model uses reinforcement learning to adjust parameters based on feedback from the score data and the supervised dataset, thereby obtaining higher score data and completing the training.

[0150] The interaction method for a refrigeration device provided in this application can be executed by an interaction device of the refrigeration device. This application uses the example of an interaction device of the refrigeration device executing the interaction method to illustrate the interaction device of the refrigeration device provided in this application.

[0151] This application also provides an interactive device for a refrigeration equipment.

[0152] like Figure 7 As shown, the interaction device of the refrigeration equipment includes:

[0153] The first receiving module 710 is used to receive the user's first input;

[0154] The first processing module 720 is used to determine the target interaction data between the user and the refrigeration equipment in response to the first input.

[0155] The second processing module 730 is used to query the target interaction intent corresponding to the target interaction data through a routing mechanism;

[0156] The third processing module 740 is used to determine the interaction task corresponding to the target interaction intent based on the target interaction intent and the execution information of the refrigeration device. The execution information includes at least one of the execution behavior, execution plan and execution memory of the refrigeration device.

[0157] The fourth processing module 750 is used to execute the interactive task through the target execution module corresponding to the interactive task.

[0158] According to the interactive device of the refrigeration equipment provided in the embodiments of this application, the target interactive intent corresponding to the target interactive data is matched by the rules and algorithms preset by the routing mechanism. The target interactive intent is accurate. Based on the target interactive intent and the execution behavior, execution plan and execution memory of the refrigeration equipment, the interactive task is reasonably determined and the interactive task is executed by the target execution module. This can improve the accuracy of the interactive information fed back to the user and enhance the interactive effect.

[0159] In some embodiments, the second processing module 730 is used to query the target interaction intent corresponding to the target interaction data through a routing mechanism combined with an attention mechanism.

[0160] In some embodiments, the third processing module 740 is used to determine the interaction task corresponding to the target interaction intent based on the target interaction intent and execution information, through a thought chain.

[0161] In some embodiments, execution memory is formed through the following steps:

[0162] Obtain the interaction content between the user and the refrigeration equipment;

[0163] Calculate the memory value corresponding to the interactive content based on the time, space and frequency information of the interactive content;

[0164] Execution memory is formed based on the interactive content and the corresponding memory value.

[0165] In some embodiments, the third processing module 740 is used to input the interaction task into the encoding sub-model of the encoding and decoding model, and encode the interaction task through the encoding sub-model to obtain the encoded interaction task.

[0166] The encoding interaction task is input into the decoding sub-model of the encoding / decoding model. The decoding sub-model decodes the encoding interaction task to obtain the decoding interaction task.

[0167] Based on the decoding interaction task, the target execution module is determined, and the interaction task is executed through the target execution module.

[0168] In some embodiments, the third processing module 740 is used to determine the target network layer and the pruning conditions of the target network layer in the encoder network of the encoding sub-model;

[0169] Based on the pruning conditions, the target network layer is pruned, and the interactive task is encoded through the pruned encoder network to obtain the encoded interactive task.

[0170] In some embodiments, the encoding / decoding model is trained through the following steps:

[0171] Obtain the supervised dataset for the encoding / decoding model;

[0172] Based on the embedded encoding of the encoder output in the encoding sub-model, the encoding and decoding model is fine-tuned to determine the score data;

[0173] Based on the score data and supervised dataset, the encoding and decoding model is trained through reinforcement learning.

[0174] The interactive device of the cooling device in this application embodiment can be an electronic device or a component of 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 laptop, a mobile internet device (MID), an ultra-mobile personal computer (UMPC), a netbook, or a personal digital assistant (PDA), etc., and can also be a server, network attached storage (NAS), a personal computer (PC), etc., and this application embodiment does not specifically limit it.

[0175] The interactive device of 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.

[0176] The interactive device for the refrigeration equipment provided in this application embodiment can achieve... Figure 1 The various processes implemented in the method implementation examples will not be described again here to avoid repetition.

[0177] This application also provides a refrigeration device.

[0178] The refrigeration equipment includes the interactive device and data acquisition device of the aforementioned refrigeration equipment.

[0179] The data acquisition device is connected to the interaction device, and the data acquisition device is used to collect interaction data between the user and the refrigeration equipment.

[0180] According to the refrigeration device provided in the embodiments of this application, the target interaction intent corresponding to the target interaction data is matched by the rules and algorithms preset by the routing mechanism. The target interaction intent obtained is accurate. Based on the target interaction intent and the execution behavior, execution plan and execution memory of the refrigeration device, the interaction task is reasonably determined and the interaction task is executed by the target execution module. This can improve the accuracy of the interaction information fed back to the user and enhance the interaction effect.

[0181] This application also provides an interactive system for a refrigeration device.

[0182] The interactive system of refrigeration equipment includes the aforementioned refrigeration equipment and server.

[0183] The server is connected to the cooling equipment and is used to output the interaction results obtained by the interactive device from performing the interaction task to the user.

[0184] According to the refrigeration equipment interaction system provided in the embodiments of this application, the target interaction intent corresponding to the target interaction data is matched by the rules and algorithms preset by the routing mechanism. The target interaction intent obtained is accurate. Based on the target interaction intent and the execution behavior, execution plan and execution memory of the refrigeration equipment, the interaction task is reasonably determined and the interaction task is executed by the target execution module. This can improve the accuracy of the interaction information fed back to the user and enhance the interaction effect.

[0185] The following introduces a method such as... Figure 2 The large model of the refrigeration equipment shown is combined with a data acquisition device and a server to collect data, realize the interaction method of the refrigeration equipment, and provide feedback of the interaction results to the user in a specific embodiment.

[0186] Step 1: The data channel module is initialized to acquire data from various sources, including the refrigeration equipment, its own information, internal and external environmental information, human-computer interaction, multi-turn dialogue, intelligent agents, multiple intelligent agents, personal intelligent assistants, data transmission, data generation tools, and data upload. It also initializes or restores data sources and terminals. The focus is on acquiring data from the refrigerator itself, including information on diet, ingredients, health and nutrition, refrigerator knowledge, casual conversation, songs, music, and recipes. This involves acquiring real-time and offline voice, text, and video data from various input methods of the refrigeration equipment, including 5G / 6G, WiFi, APP, microphone, mobile phone, Bluetooth, touchscreen, fingerprint device, image acquisition device, and camera. If the hardware acquisition device is an independent peripheral, the data acquisition is completed by the external device.

[0187] Step 2 is achieved through the data acquisition module, which completes data acquisition tasks such as interaction between the refrigeration equipment and various data information, user information, 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. It mainly completes the acquisition of text data, voice data, audio data, and audio-visual data, and the generation of multimodal data.

[0188] Step 3 is achieved through the data processing module, which completes the preprocessing tasks for various data types of information, including information about the refrigeration equipment itself, the internal and external environment, and the user. First, this data is cleaned and converted into different formats, such as parsing and storing the data. For audio and video data, tools or scripts are used to separate the audio and video to obtain valid audio and video data. Text data undergoes preprocessing, such as stop word removal and deduplication. Next, image frame data is processed. Through these basic data preprocessing tasks, valid data of the corresponding data types are obtained.

[0189] It should be noted that if the audio and image data themselves are valid, you can proceed directly to step six.

[0190] Step 4: Through the data fusion module, the acquired information data, such as voice data, image data, and text data, are transformed and fused for model pre-training, model prediction, and downstream tasks.

[0191] First, for converting speech data obtained from microphones and pickups into text, based on the spatiotemporal characteristics and contextual information features of the speech data from the cooling device, an end-to-end transformer block speech recognition model architecture based on convolutional enhancement is adopted to obtain the contextual information of the speech data. Based on the local feature information of the speech data, the convolutional enhancement model is used to obtain efficient local speech semantic feature information. The speech-to-text model can be implemented based on a Gaussian neural network model or an end-to-end deep neural network model with an encoding and decoding structure, for example, using the Kaldi open-source speech recognition architecture.

[0192] This paper proposes a method for extracting text from images and audio / video data obtained by acquisition devices such as gesture and eye trackers. The method extracts and identifies text content from image features in audio data. First, the audio data is separated. The sentences identified based on image features are relatively complex, such as inconsistent sentence lengths, inconsistent word structures, and contextual relationships. The image recognition method is based on a deep fusion network model of transfer and distillation diffusion based on spatiotemporal and long-distance dependent features to identify text content.

[0193] Among them, the key steps of the distillation-diffusion deep fusion model are to improve and utilize the transfer and teacher model output results in the distillation-diffusion deep fusion model to mine rich semantic feature information of sentence sequences. Alternatively, a deep neural network fusion model structure based on the diffusion network model can be used to obtain semantic feature information.

[0194] After the deep fusion model of distillation, migration, or quantization diffusion is trained, the teacher model is introduced into a temporally and spatially continuous student model, and the resulting student model has the same parameters as the teacher model.

[0195] The transfer and distillation model is transformed into a student model with discrete time steps and a short number of steps. This process is repeated for N steps until the teacher model is distilled to half the number of steps of the student model, i.e., N / 2.

[0196] Next, we will perform effective text data fusion and filtering. This will facilitate the training and information generation based on text data after the large model is built. Filtering refers to cleaning the text data and generating high-quality data.

[0197] Step 5 is achieved through the historical data accumulation module, which involves converting the human-computer interaction or intelligent interaction data collected by the refrigeration equipment into text data, and then vectorizing and mapping this text data to the same vector space through transfer learning. This is beneficial for the unification of data representation calculation and measurement, and data features can be extracted using deep convolutional networks to obtain data representation.

[0198] In this process, the conversion of historical non-text data, such as voice and images, into text data can be done by referring to step four. Obtaining historical data further ensures the comprehensiveness and richness of the generated high-quality dataset.

[0199] Step Six: Implemented through the intelligent agent application module, this involves completing tasks such as intelligent agent autonomous systems, actions, planning, hybrid routing attention mechanisms, integrated business workflows combining memory and toolsets, and automation. Specifically, for example... Figure 3 As shown, on the one hand, the large model is used as a control or service. When there is a downstream task, the hybrid routing attention mechanism is combined with routing, intelligent agents, tools, services, and model APIs. It utilizes the established or deployed memory, action, observation, and other business processes and workflow development capabilities. On the other hand, intelligent agents are combined with one / few-shot and prompting engineering to form a downstream task generator, which effectively adapts to downstream tasks and meets more business needs and development efficiency. For example, if there are eggs with an expiration date or no eggs with an expiration date, the generator can efficiently obtain the generation result through the memory management module and intelligent agents.

[0200] Based on the query, the task of the hybrid routing attention mechanism for agents is completed, that is, based on the query sentence, such as asking about preservation, food, etc., the corresponding agents are assigned through weights and hybrid routing.

[0201] First, the text content is obtained. To avoid issues such as illusions in the generated returned information, an attention mechanism is added to the routing, that is, mutual attention between each route. This is beneficial for obtaining semantic information between text content. Therefore, the dot product operation of query and key is performed between routes.

[0202] Then, after obtaining the dot product result, the dot product operation of the value is performed.

[0203] Next, the results are added together to obtain the route weights and summation, thus determining the hybrid route attention mechanism.

[0204] In this embodiment, when a large model is used as a control or service, downstream tasks are achieved by using a hybrid routing attention mechanism, intelligent agents, tools, services, and model API capabilities, and leveraging existing or deployed business processes and workflows such as memory, action, and observation.

[0205] Intelligent agents or multiple intelligent agents complete optimizer / evolver / hint generator tasks. For the received query, the intelligent agent forms the corresponding optimizer through its own functions such as action, planning, and memory, including hint words. By combining the intelligent agent with one / few-shot and hint engineering, a downstream task generation scheme is formed, which effectively adapts to downstream tasks and meets more business needs and development efficiency.

[0206] Beforehand, one / few-shot and prompt engineering inputs are fed into the large model. The large model generates a solution generator for the current task through the thinking chain algorithm. When the toolset is called again, the information results are returned through the intelligent agent of the workflow. Among them, the memory manager or memory module calculates the average memory value based on the obtained task content. If the average value is less than the threshold, short memory is selected; if the average value is less than the threshold, long memory is selected; otherwise, it is skipped.

[0207] In this embodiment, the toolset is formed by encapsulating the API of a large model. When the corresponding toolset is called, the query response is completed and the result is returned. The toolset may include knowledge graphs, external knowledge content, knowledge services, etc.

[0208] Step 7: This is achieved through the Retrieval-Augmented Generation (RAG) module. By embedding vector models and vector libraries, when faced with a query, the module calculates or extracts results based on semantic similarity and returns the required results to the refrigeration equipment.

[0209] Step 8: Implemented through the large model encoder module, this step completes tasks such as text data encoding and optimization of its model fusion blocks, optimization of the model's internal structure, and context learning. For different data types, different downstream task categories, and application scenario categories, the best-performing modules are selected. This includes encoding attention blocks (Att Blocks) composed of residual networks, RMS Norm normalization, and multi-head attention mechanisms, which can improve model performance, information accuracy, and intent understanding capabilities.

[0210] It is understandable that the large model encoder module may include the aforementioned encoding sub-models, which may include, for example... Figure 4 The encoder network in the codec network shown.

[0211] First, the embedding layer encoder is pruning. The embedding layer encoder can be divided into several sub-blocks to obtain more fine-grained coding units, which is beneficial for extracting more detailed granular semantic features in the future.

[0212] For example, pruning is performed on the input and output data of the embedding layer within the neurons. Specifically, RMSOrm normalization is calculated for the input and output results. Normalization is performed on each neuron, and a pruning threshold is set for each RMSOrm. If the RMSOrm is less than the set pruning threshold, a pruning operation is performed, deleting the corresponding row or column of the transpose matrix. That is, 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, resulting in an embedding layer with important weights. The efficient computation and generalization ability of the weights for large model pre-training are fully demonstrated and verified. The residual network is removed, the number of network layers is reduced, and the computational efficiency is improved, resulting in better performance.

[0213] Next, the activation layer pruning task is completed, which involves activation layer pruning with multi-head attention mechanism, residual and RMSnorm normalization calculation. The RMSnorm normalization calculation module is after the multi-head attention mechanism, i.e., post-Norm. It identifies unimportant neurons in the output of the activation layer and prunes them, i.e., deletes unimportant neurons, such as GeLU or ReLU layers. Reducing the model parameters is equivalent to performing compression pruning of activation layers with multi-head attention mechanism without affecting model performance and pre-training quality. This can improve the optimization performance of the internal structure of the deep network model and improve the efficiency of parallel computation of the internal structure of the model.

[0214] Then, a fully connected feed-forward neural network (FFN) is constructed. First, N FFN network layers, residuals, and RMSNorm are constructed, which can improve network performance, reduce the resources occupied by RMSNorm normalization computation, and facilitate the acquisition of high-value contextual semantic feature information.

[0215] Complete the pruning of activation layers with a full multi-head self-attention mechanism, and the normalization of residual layers and RMSNorm. The specific process is the same as completing the activation layer punting task, i.e., the calculation of activation layer pruning, residual and RMSNorm normalization with a multi-head attention mechanism.

[0216] This process involves performing RMSNorm normalization to obtain better data representation features, identifying unimportant neurons in the activation layer output and pruning them (e.g., deleting these unimportant neurons from GeLU or ReLU layers), and then outputting the normalization calculation for the next layer.

[0217] Finally, the feedforward FFN layer operation is completed, and on this basis, the residual network layer and RMSNorm normalization operation are completed. Finally, the encoded output result is output. The residual and RMSNorm normalization operations and feedforward network operations are completed on the data input and output results. It can be fully connected. Finally, the encoded result is output.

[0218] Step 9: Implement the image recognition classifier module. Through distillation techniques such as teacher-student models or quantization, classify food images, including images of ingredients, food, beverages, soy sauce, vinegar, etc. Complete tasks such as building and pre-training a large visual model based on Vi-transformer, i.e., a visual transformer structure model. For example, extract contextual image information from food images to improve the efficiency and accuracy of the large model visual classifier.

[0219] First, the image data is segmented into patches, which means that the image data is reduced to a 2D data format by word segmentation method and unified into the same 2D data vector space.

[0220] Then, the 2D vector space is mapped to a 1D vector space of the same length, which generates the patchembedding.

[0221] Next, for learnable predictive classification, we obtain position embedding vectors, which can acquire multi-layer sequence information such as position, part of speech, and word order, providing more information for the classification embedding vectors.

[0222] Next, the obtained vectors are encoded using a transformer encoder block, which includes operations such as normalization, multi-head attention mechanism, and multi-layer perception.

[0223] Finally, a multilayer perceptron (MLP) classification is performed to identify the category of ingredients, food, or fruits and vegetables.

[0224] Step 10: Implement autonomous learning tasks through the reinforcement learning model pre-training module, and complete tasks such as fine-tuning the model and reinforcement learning from human feedback (RLHF).

[0225] First, based on the generated or collected high-quality supervised dataset, and using the data with embedded encoding, the score distribution or score sequence of the reward model is obtained by fine-tuning the large model. On the basis of the reward model, combined with the high-quality dataset and human feedback behaviors, habits, or even hobbies and interests, such as combining the names of dishes made by users or the history of question and answer, a large language model (LLM) is generated through reinforcement learning. The generated LLM has good robustness and stability in terms of general network parameters, professional domain data, and general data. In addition, reinforcement learning of proximal policy optimization (PPO) / distributed proximal optimization (DPO) strategies is carried out.

[0226] Step 11: Implemented through the large model decoder module, which completes tasks such as decoding the encoded data, model compression, and downstream multi-task adaptation. This can improve the performance of the model itself, as well as the accuracy and timeliness of information. It can also optimize the model's own structure and enhance the complementarity, dependence, and correlation of semantics in the pre-training resource data.

[0227] It is understandable that the large model decoder module may include the aforementioned decoding sub-models, which may include, for example... Figure 4 The decoder network in the codec network shown.

[0228] For speech and video output data, a deep network model consisting of a fully attention module (Att conVBlock) and a transformer (conVand ReLU & LayerNorm) is used to complete the decoding task. First, a unified feature vector or expression is formed between features through the cross-attention mechanism. Then, convolution operation, ReLU calculation, and LayerNorm calculation are performed to obtain semantic features of high-level, long-dependency distance relationships, thereby completing the decoding task and generating the corresponding speech and video data.

[0229] For the text output data, the encoded text features are first semantically compressed using a word segmentation-level network model, and then the text format data is output through an autoregressive decoder.

[0230] Step 12: Implemented through the model service module, which involves deploying the pre-trained large model project, publishing the application service to downstream multi-tasks, providing service data types including voice, image, text and video, including tasks such as application, invocation, authorization, logging and monitoring, and executing downstream tasks based on the classification and recognition results.

[0231] like Figure 5As shown, a visual, user-friendly, and quantifiable multi-task joint application service interface is implemented, following the steps of model evaluation, model encapsulation, model plugins, model service activation, API call, API results, and matching downstream tasks.

[0232] like Figure 6 As shown, the refrigeration device with a large model provided in this application embodiment includes several layers, including constant temperature layers such as freezing and refrigeration layers, and also includes a human-computer interaction module. The data acquisition module may include gesture acquisition devices, voice acquisition devices, text acquisition devices, audio and video acquisition devices, image acquisition devices, microphone arrays, radar signal acquisition, etc. It also includes optical character recognition (OCR) devices, file modules, virtual reality (VR), augmented reality (AR), etc., and also includes a display interface. The presentation form of the display interface may include APP interface, PC interface, web interface, mini-program interface, and large screen, etc. The communication module supports protocols such as TCP / IP, Wi-Fi, Bluetooth, etc., and supports networks such as 5G, Wi-Fi, wired network, 6G, etc. The application fields include home, hotel, industrial, indoor and outdoor occasions, military and aerospace, etc., and the embedding methods include flat embedding, vehicle embedding, integration, etc.

[0233] Table 1

[0234] Serial Number Scene Name Specific intelligent agent Task Examples 1 Food identification Vegetable Name Image Recognition Eggs, cabbage... 2 Food preservation Smart Preservation System Apple's expiration date may still be a few days away. 3 Food Management Food Management Output the quantity of various ingredients 4 Refrigerator unit Local intelligent agent How to turn off the lights 5 Health knowledge Nutritional and Health Intelligent Body Can people with high blood pressure eat tomatoes? 6 Health monitoring Health Intelligent Body What is my blood pressure today?

[0235] like Figure 8 As shown, the acquisition module or acquisition device collects data, transmitting text, images, audio, video, and other data to the database and toolset. The data is then uploaded to the LLM for fine-tuning, pre-training, and other purposes. The LLM, database, and toolset interact and exchange information, which is then displayed on the local end of the cooling device or on the large screen or corresponding area of ​​the embedded system. The prediction results are presented on the output end or intelligent presentation device via API or HTTP.

[0236] Table 1 shows the task instances, i.e., the interaction results, output by the specific intelligent agents in different task scenarios.

[0237] The interaction method for refrigeration equipment provided in this application integrates a large model agent of refrigeration equipment with N-shot, prompt words, hybrid routing attention mechanism, network containing mutual attention, self-attention and cross-attention, etc. In this method, the construction of a deep neural network model or LLM with an end-to-end encoder-decoder structure can be based on a transformer with multi-head attention only-decoder, encoder transformer, full cross-attention, group attention and full self-attention block, thereby constructing a deep neural network large language model.

[0238] The encoding and decoding large model construction involved in the embodiments of this application can adopt semantic compression large language models such as decoder-only architecture, encoder-only architecture, and encoder-decoder architecture. Models (LLMs) include, but are not limited to, deep network models such as LLMs with graph neural networks, attention mechanisms, transformer models and their variants or improvements, distillation network LLMs, latent / diffusion model LLMs, U-net networks, etc., as well as fusion deep network model LLMs such as transformer-based bidirectional long short-term memory networks, multimodal convolutional neural networks and recurrent neural network combined models (BiLSTM+MMCNN+RNN), bidirectional long short-term memory networks, multimodal convolutional neural networks and attention mechanisms combined models (BiLSTM+MMCNN+Attention), multimodal convolutional neural networks, bidirectional long short-term memory networks and attention mechanisms combined models (MMCNN+BiLSTM+Attention), etc., as well as models combining gated recurrent units and convolutional neural networks (GRU+CNN), gated recurrent unit convolutional neural networks and attention mechanisms (GRU+CNN+Attention), deep reinforcement learning and reward models, retrieval-augmented generation (RAG) LLMs, etc., and Gaussian mixture deep neural network models.

[0239] In some embodiments, such as Figure 9 As shown, this application embodiment also provides an electronic device 900, including a processor 901, a memory 902, and a computer program stored in the memory 902 and executable on the processor 901. When the program is executed by the processor 901, it implements the various processes of the above-described interaction method embodiment of the cooling device and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0240] 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.

[0241] 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 above-described interaction method embodiment of the refrigeration device and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0242] 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.

[0243] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the interaction method of the above-mentioned refrigeration device.

[0244] 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.

[0245] 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 interaction method embodiment of the refrigeration device, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0246] 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.

[0247] 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.

[0248] 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 prior art, 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.

[0249] 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.

[0250] 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.

[0251] 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. An interaction method for a refrigeration device, characterized in that, include: Receive the user's first input; In response to the first input, the target interaction data between the user and the refrigeration device is determined; The target interaction intent corresponding to the target interaction data is queried through the routing mechanism; Based on the target interaction intent and the execution information of the refrigeration device, the interaction task corresponding to the target interaction intent is determined. The execution information includes at least one of the execution behavior, execution plan, and execution memory of the refrigeration device. The interactive task is executed by the target execution module corresponding to the interactive task.

2. The interaction method of the refrigeration equipment according to claim 1, characterized in that, The step of querying the target interaction intent corresponding to the target interaction data through a routing mechanism includes: By using the routing mechanism and the attention mechanism, the target interaction intent corresponding to the target interaction data is queried.

3. The interaction method of the refrigeration equipment according to claim 1, characterized in that, The step of determining the interaction task corresponding to the target interaction intent based on the target interaction intent and the execution information of the refrigeration device includes: Based on the target interaction intent and the execution information, the interaction task corresponding to the target interaction intent is determined through the thought chain.

4. The interaction method of the refrigeration equipment according to claim 1, characterized in that, The execution memory is formed through the following steps: Obtain the interaction content between the user and the refrigeration equipment; Based on the time information, spatial information and frequency information of the interactive content, calculate the memory value corresponding to the interactive content; The execution memory is formed based on the interactive content and the memory value corresponding to the interactive content.

5. The interaction method of the refrigeration equipment according to any one of claims 1-4, characterized in that, The step of executing the interactive task through the target execution module corresponding to the interactive task includes: The interactive task is input into the encoding sub-model of the encoding / decoding model, and the interactive task is encoded by the encoding sub-model to obtain the encoded interactive task. The encoding interaction task is input into the decoding sub-model of the encoding / decoding model, and the encoding interaction task is decoded by the decoding sub-model to obtain the decoding interaction task; Based on the decoding interaction task, the target execution module is determined, and the interaction task is executed through the target execution module.

6. The interaction method of the refrigeration equipment according to claim 5, characterized in that, The step of encoding the interaction task through the encoding sub-model to obtain the encoded interaction task includes: In the encoder network of the encoding sub-model, the target network layer and the pruning conditions of the target network layer are determined; Based on the pruning conditions, the target network layer is pruned, and the interaction task is encoded through the pruned encoder network to obtain the encoded interaction task.

7. The interaction method of the refrigeration equipment according to claim 5, characterized in that, The encoding / decoding model is trained through the following steps: Obtain the supervised dataset for the encoding / decoding model; Based on the embedded encoding of the encoder output in the encoding sub-model, the encoding and decoding model is fine-tuned to determine the score data; Based on the score data and the supervised dataset, the encoding / decoding model is trained through reinforcement learning.

8. An interactive device for a refrigeration equipment, characterized in that, include: The first receiving module is used to receive the user's first input; A first processing module is configured to, in response to the first input, determine the target interaction data between the user and the refrigeration device; The second processing module is used to query the target interaction intent corresponding to the target interaction data through a routing mechanism; The third processing module is used to determine the interaction task corresponding to the target interaction intent based on the target interaction intent and the execution information of the refrigeration device. The execution information includes at least one of the execution behavior, execution plan and execution memory of the refrigeration device. The fourth processing module is used to execute the interactive task through the target execution module corresponding to the interactive task.

9. A refrigeration device, characterized in that, include: The interactive device for the refrigeration equipment as described in claim 8; A data acquisition device is connected to the interaction device and is used to collect interaction data between the user and the refrigeration equipment.

10. An interactive system for a refrigeration device, characterized in that, include: The refrigeration equipment as described in claim 9; The server is connected to the cooling device and is used to output the interaction results obtained by the interactive device in performing the interactive task to the user.

Images