An intelligent sensing system and control method for water dispenser

Abstract

The present invention proposes an intelligent perception system and a control method for a water dispenser. The system reads perception data from each perception data channel, integrates and identifies the perception data in a perception time window, and determines whether the perception data in the perception time window is learning data or decision data. The learning data is used to train a decision model, and the decision data is used to input the decision model so that the decision model outputs interactive feature data. When the perception data in the perception time window is learning data, the learning data is persisted in a learning sample database. When the perception data in the perception time window is decision data, the decision data is input into the decision model so that the decision model outputs interactive feature data. User interactive actions are performed according to the interactive feature data, so that the water dispenser truly has the ability of autonomous learning and autonomous decision-making.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control technology, and in particular to an intelligent sensing system applied to a water dispenser and a control method thereof. Background Art

[0002] Water dispensers are one of the most widely used household appliances in people's daily lives. Water dispensers are widely loved for their ability to provide drinking water conveniently and safely, especially for their ability to provide cold and hot water instantly. With the emergence of smart home technology and artificial intelligence technology, water dispensers are gradually developing in the direction of intelligence, such as water quality detection, user behavior analysis, automatic sensing of water, intelligent temperature adjustment, personalized drinking water suggestions, remote control, and smart home linkage. However, the intelligent technology used by existing water dispensers still remains at the level of performing operations based on preset programs or rules. Its degree of intelligence relies on the feasibility, versatility and richness of preset programs or preset rules, and cannot truly achieve autonomous learning and autonomous decision-making. Its intelligent capabilities are limited, and there are still very large limitations in practical applications. Summary of the invention

[0003] Based on the above problems, the present invention proposes an intelligent sensing system and a control method for a water dispenser, so that the water dispenser can truly have the ability of autonomous learning and autonomous decision-making.

[0004] In view of this, the first aspect of the present invention proposes an intelligent perception system applied to a water dispenser, comprising a perception layer for acquiring perception data, a data layer for processing, identifying and storing the perception data, a decision layer for using the perception data for autonomous learning or interactive decision-making, and an interaction layer for interacting with a user, the perception layer comprising a human perception module for perceiving human behavior and human state, an environmental perception module for perceiving environmental state, a water supply perception module for perceiving water supply parameters, and an event perception module for perceiving interactive events, the data layer comprising a data processing module for performing data standardization, normalization and numerical filling, a data feature extraction module for performing data feature identification, and a learning sample database for storing learning data, the decision layer comprising a decision model for outputting interactive feature data according to decision data, the interactive layer comprising an interactive feature data parsing module for parsing interactive control instructions and interactive control parameters of the water dispenser from the interactive feature data, and an interactive control instruction execution module for executing the interactive control instructions according to the interactive control parameters, and the intelligent perception system is configured as follows:

[0005] Configuring the size of the corresponding perception time window of each perception type, wherein the perception types include human behavior perception, human state perception, environmental state perception, water supply parameter perception, and interactive event perception;

[0006] Reading perception data from each perception data channel, the perception data including at least one of visual perception data, ultrasonic perception data, sound perception data, infrared perception data, flow perception data, temperature perception data, and touch perception data;

[0007] Integrating and identifying the perception data in the perception time window;

[0008] Determining whether the perception data in the perception time window is learning data or decision data, wherein the learning data is used to train a decision model, and the decision data is used to input the decision model so that the decision model outputs interactive feature data;

[0009] When the perception data in the perception time window is learning data, persisting the learning data in a learning sample database;

[0010] When the perception data in the perception time window is decision data, inputting the decision data into the decision model so that the decision model outputs interaction feature data;

[0011] A user interaction action is performed according to the interaction feature data.

[0012] A second aspect of the present invention provides a control method for an intelligent sensing system applied to a water dispenser, comprising:

[0013] Specifically, the perception time window is a specific length of time. The intelligent perception system of the smart water dispenser perceives and identifies human behavior, human state, environmental state, water supply parameters and interactive events by integrating the perception data of each perception data channel in the perception time window. Since the perception data required for perceiving and identifying human behavior, human state, environmental state, water supply parameters and interactive events are different, and the perception and identification methods are different, the size of the corresponding perception time window will also be different accordingly. It is necessary to configure the size of the corresponding perception time window for each perception type.

[0014] For example, human behavior is usually manifested as one or more actions that last for a certain period of time. The recognition of human behavior requires the support of perception data of a perception time window that lasts for a longer period of time. However, some specific environmental states, such as ambient temperature and humidity, are instantaneous. The recognition of these environmental states only requires one instantaneous perception data. Therefore, the perception time window corresponding to environmental state perception can be configured as a smaller value.

[0015] In the technical solution of the present invention, the learning data is the perception data including the interactive events, and the decision data is the perception data not including the interactive events. The intelligent perception system uses the learning data in the learning sample database as the training sample data of the decision model to continuously train and optimize the decision model, so that the decision model can use the perception data of the perception layer to accurately output interactive decisions.

[0016] Furthermore, the step of integrating and identifying the perception data in the perception time window specifically includes:

[0017] Configure the associated sensing data channel of each sensing type and the feature type corresponding to each associated sensing data channel;

[0018] Extracting feature data of corresponding perception types from the perception data of each associated perception data channel in the perception time window;

[0019] When there is no corresponding feature data in the perception data of any perception data channel, the feature data of the corresponding perception data channel is configured as a null value or configured as zero;

[0020] Otherwise, the feature data of the corresponding perception data channel is standardized and normalized;

[0021] Construct a perceptual feature data matrix of the perceptual time window.

[0022] Furthermore, after the step of integrating and identifying the perception data in the perception time window, the method further includes:

[0023] Determining whether there is a human body feature data dimension reflecting the existence characteristics of a human body in the perception feature data matrix, wherein the human body feature data dimension corresponds to one or more perception data channels;

[0024] When there is no human feature data dimension reflecting the existence characteristics of the human body in the perception feature data matrix, discard the perception feature data matrix in the current perception time window, and return to the step of reading perception data from each perception data channel;

[0025] When there is a human feature data dimension reflecting the existence characteristics of the human body in the perception feature data matrix, it is executed to determine whether the perception data in the perception time window is learning data or decision data and its subsequent steps.

[0026] Furthermore, the step of determining whether the perception data in the perception time window is learning data or decision data specifically includes:

[0027] Determining an interactive perception data channel, wherein the interactive perception data channel includes a sound perception data channel and a touch perception data channel;

[0028] Determine whether the data of the interactive perception data channel in the perception time window is null or zero;

[0029] When the data of the interactive perception data channel in the perception time window are all null values ​​or all zero, determining that the perception data in the perception time window is decision data;

[0030] Otherwise, it is determined that the perception data in the perception time window is learning data.

[0031] Furthermore, after the step of persisting the learning data in the learning sample database, the method further includes:

[0032] Configure the model deviation test cycle;

[0033] Extracting test data from the learning sample database in each model deviation test cycle to perform a deviation test on the decision model, wherein the test data is learning data whose generation time is less than a preset time threshold;

[0034] When the deviation test result of the decision model is greater than a preset deviation threshold, the decision model is retrained using the learning data in the learning sample database.

[0035] Furthermore, in each model deviation test cycle, the step of extracting test data from the learning sample database to perform deviation test on the decision model specifically includes:

[0036] Determining an input data dimension and an output data dimension in the perception feature data matrix of the test data, wherein the output data dimension is a data dimension corresponding to the interactive perception data channel, and the input data dimension is other data dimensions in the perception feature data matrix except the output data dimension;

[0037] Constructing an input data matrix based on the data of the input data dimension;

[0038] Merging the perception data of the output data dimension into first output data, wherein the first output data is a one-dimensional data sequence;

[0039] Inputting the input data matrix into the decision model to obtain second output data, wherein the second output data is a one-dimensional data sequence, and the first output data and the second output data are interactive feature data;

[0040] calculating the similarity between the first output data and the second output data;

[0041] A deviation test result of the decision model is determined according to the similarity between the first output data and the second output data.

[0042] Furthermore, the step of inputting the decision data into the decision model so that the decision model outputs the interactive feature data specifically includes:

[0043] Determining input data dimensions in a perceptual feature data matrix of the decision data;

[0044] Constructing an input data matrix based on the data of the input data dimension;

[0045] The input data matrix is ​​input into the decision model to obtain the interaction feature data, where the interaction feature data is a one-dimensional data sequence.

[0046] Furthermore, the data dimension corresponding to the interactive perception data channel in the perception feature data matrix includes interactive control instruction execution condition identification features, and before the step of persisting the learning data in the learning sample database, it also includes:

[0047] Acquire water supply parameter sensing data from a water supply parameter sensing data channel;

[0048] Determining, according to the water supply parameter sensing data, whether the water dispenser has the conditions to execute the interactive control instruction based on the interactive control parameter;

[0049] The interactive control instruction execution condition identification feature is configured according to a judgment result of whether the water dispenser has the conditions for executing the interactive control instruction based on the interactive control parameter.

[0050] Furthermore, the step of executing the user interaction action according to the interaction feature data specifically includes:

[0051] Parsing the interactive feature data to obtain interactive control instructions and interactive control parameters of the water dispenser;

[0052] Acquire water supply parameter sensing data from a water supply parameter sensing data channel;

[0053] Determining, according to the water supply parameter sensing data, whether the water dispenser has the conditions to execute the interactive control instruction based on the interactive control parameter;

[0054] When the water dispenser has the condition to execute the interactive control instruction based on the interactive control parameter, the interactive control instruction is executed according to the interactive control parameter.

[0055] The present invention proposes an intelligent perception system and a control method for a water dispenser. The system reads perception data from each perception data channel, integrates and identifies the perception data in a perception time window, and determines whether the perception data in the perception time window is learning data or decision data. The learning data is used to train a decision model, and the decision data is used to input the decision model so that the decision model outputs interactive feature data. When the perception data in the perception time window is learning data, the learning data is persisted in a learning sample database. When the perception data in the perception time window is decision data, the decision data is input into the decision model so that the decision model outputs interactive feature data. User interactive actions are performed according to the interactive feature data, so that the water dispenser truly has the ability of autonomous learning and autonomous decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 is a schematic diagram of an intelligent sensing system applied to a water dispenser provided by an embodiment of the present invention;

[0057] Figure 2 It is a flow chart of a control method of an intelligent sensing system applied to a water dispenser provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0058] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0059] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0060] In the description of the present invention, the term "multiple" refers to two or more. Unless otherwise clearly defined, the orientation or positional relationship indicated by the terms "upper", "lower", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation of the present invention. The terms "connection", "installation", "fixation", etc. should be understood in a broad sense. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be directly connected or indirectly connected through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances. In addition, the terms "first", "second", etc. are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first", "second", etc. can explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, the meaning of "multiple" is two or more.

[0061] In the description of this specification, the description of the terms "one embodiment", "some implementations", "specific embodiments", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0062] The following describes an intelligent sensing system and control method for a water dispenser according to some embodiments of the present invention with reference to the accompanying drawings.

[0063] like Figure 1As shown, the first aspect of the present invention proposes an intelligent perception system applied to a water dispenser, comprising a perception layer for acquiring perception data, a data layer for processing, identifying and storing the perception data, a decision layer for using the perception data for autonomous learning or interactive decision-making, and an interaction layer for interacting with a user, the perception layer comprising a human perception module for perceiving human behavior and human state, an environmental perception module for perceiving environmental state, a water supply perception module for perceiving water supply parameters, and an event perception module for perceiving interactive events, the data layer comprising a data processing module for performing data standardization, normalization and numerical filling, a data feature extraction module for performing data feature identification, and a learning sample database for storing learning data, the decision layer comprising a decision model for outputting interactive feature data according to the decision data, the interactive layer comprising an interactive feature data parsing module for parsing the interactive feature data to obtain interactive control instructions and interactive control parameters of the water dispenser, and an interactive control instruction execution module for executing the interactive control instructions according to the interactive control parameters.

[0064] The human perception module and the environmental perception module perform multi-dimensional comprehensive perception of human behavior, human state and environmental state through multiple types of sensors. The perception layer includes multiple perception data channels that obtain perception data from various types of sensors, including but not limited to a visual perception data channel for obtaining visual perception data from a camera, an ultrasonic perception data channel for obtaining ultrasonic perception data from an ultrasonic sensor, a sound perception data channel for obtaining sound perception data from a microphone, an infrared perception data channel for obtaining infrared perception data from an infrared sensor, a flow perception data channel for obtaining flow perception data from a flow sensor, a temperature perception data channel for obtaining temperature perception data from a temperature sensor, and a touch perception data channel for obtaining touch perception data from a touch panel.

[0065] Exemplarily, the human perception module senses and identifies human behavior and human status based on the perception data provided by the visual data channel, the sound perception data channel and / or the infrared perception data channel, and the environmental perception module senses and identifies environmental status based on the perception data provided by the visual data channel, the temperature perception data channel and / or the infrared perception data channel.

[0066] Furthermore, the perception layer also includes an Internet data channel for obtaining environmental data from the Internet. Since water dispensers are usually placed in indoor environments and are affected by indoor air-conditioning equipment, the indoor temperature, humidity, wind speed, etc. are usually different from those outdoors. In order to improve the comprehensive perception capability, the intelligent perception system obtains the environmental status data of the outdoor environment through the Internet data channel. The outdoor environmental status data includes but is not limited to outdoor data such as outdoor temperature, outdoor humidity, and outdoor wind speed.

[0067] Further, the water dispenser includes a first temperature sensor for obtaining ambient temperature and a second temperature sensor for obtaining drinking water temperature, and accordingly, the sensing data channel includes a first temperature sensing data channel for obtaining ambient temperature sensing data from the first temperature sensor, and a second temperature sensing data channel for obtaining water temperature sensing data from the second temperature sensor. The water supply sensing module senses and identifies the water supply status based on the sensing data provided by the flow sensing data channel and / or the second temperature sensing data channel.

[0068] The event perception module perceives and identifies user interaction events based on the perception data provided by the sound perception data channel and / or the touch perception data channel. For example, voice interaction instructions input by the user are obtained through the sound perception data channel, or touch interaction instructions input by the user are obtained through the touch perception data channel.

[0069] like Figure 2 As shown, the intelligent sensing system is configured as follows:

[0070] Configuring the size of the corresponding perception time window of each perception type, wherein the perception types include human behavior perception, human state perception, environmental state perception, water supply parameter perception, and interactive event perception;

[0071] Reading perception data from each perception data channel, the perception data including at least one of visual perception data, ultrasonic perception data, sound perception data, infrared perception data, flow perception data, temperature perception data, and touch perception data;

[0072] Integrating and identifying the perception data in the perception time window;

[0073] Determining whether the perception data in the perception time window is learning data or decision data, wherein the learning data is used to train a decision model, and the decision data is used to input the decision model so that the decision model outputs interactive feature data;

[0074] When the perception data in the perception time window is learning data, persisting the learning data in a learning sample database;

[0075] When the perception data in the perception time window is decision data, inputting the decision data into the decision model so that the decision model outputs interaction feature data;

[0076] A user interaction action is performed according to the interaction feature data.

[0077] Specifically, the perception time window is a specific length of time. The intelligent perception system of the smart water dispenser perceives and identifies human behavior, human state, environmental state, water supply parameters and interactive events by integrating the perception data of each perception data channel in the perception time window. Since the perception data required for perceiving and identifying human behavior, human state, environmental state, water supply parameters and interactive events are different, and the perception and identification methods are different, the size of the corresponding perception time window will also be different accordingly. It is necessary to configure the size of the corresponding perception time window for each perception type.

[0078] For example, human behavior is usually manifested as one or more actions that last for a certain period of time. The recognition of human behavior requires the support of perception data of a perception time window that lasts for a longer period of time. However, some specific environmental states, such as ambient temperature and humidity, are instantaneous. The recognition of these environmental states only requires one instantaneous perception data. Therefore, the perception time window corresponding to environmental state perception can be configured as a smaller value.

[0079] In the technical solution of the present invention, the learning data is the perception data including the interactive events, and the decision data is the perception data not including the interactive events. The intelligent perception system uses the learning data in the learning sample database as the training sample data of the decision model to continuously train and optimize the decision model, so that the decision model can use the perception data of the perception layer to accurately output interactive decisions.

[0080] Furthermore, in the step of integrating and identifying the perception data in the perception time window, the intelligent perception system is configured as follows:

[0081] Configure the associated sensing data channel of each sensing type and the feature type corresponding to each associated sensing data channel;

[0082] Extracting feature data of corresponding perception types from the perception data of each associated perception data channel in the perception time window;

[0083] When there is no corresponding feature data in the perception data of any perception data channel, the feature data of the corresponding perception data channel is configured as a null value or configured as zero;

[0084] Otherwise, the feature data of the corresponding perception data channel is standardized and normalized;

[0085] Construct a perceptual feature data matrix of the perceptual time window.

[0086] In the technical solution of the present invention, the associated perception data channels corresponding to each perception type are pre-configured, for example, the associated perception data channels of human behavior perception include one or more of visual perception data channels, sound perception data channels, and infrared perception data channels, and the associated perception data channels of environmental state perception include one or more of visual perception data channels, temperature perception data channels, and Internet data channels, etc. It should be known that the associated perception data channels corresponding to the above perception types are only examples, and in specific implementations, different associated perception data channels corresponding to each perception type can be configured as needed.

[0087] Furthermore, for different perception types, they can be associated with the same perception data channel, but the corresponding feature types in the same perception data channel may be different. For example, for human behavior perception, the data features extracted from the visual data channel may include head feature data, hand feature data and other limb feature data. For human state perception, the data features extracted from the visual data channel may include face feature data. For environmental state perception, the data features extracted from the visual data channel may include environmental brightness feature data.

[0088] In the step of integrating and identifying the perception data in the perception time window, a perception feature data matrix is ​​constructed for each perception type. The perception feature data matrices of all perception types are n-dimensional matrices, where n is the number of perception data channels in the perception layer. In the perception feature data matrix, data of non-associated perception data channels of the current perception type are all configured as null values ​​or zeros.

[0089] In the step of persisting the learning data in the learning sample database, the intelligent perception system is configured as follows:

[0090] Obtaining timestamp information of the perception time window, the timestamp information including one or more of the upper boundary time, the lower boundary time, and any time point therebetween, that is, the timestamp information may include one or more time point information, which may be the upper boundary time of the perception time window, the upper and lower boundary times of the perception time window, or any time point between the upper and lower boundary times of the perception time window;

[0091] The perception feature data matrix in the perception time window is associated with the timestamp information and saved in a learning sample database.

[0092] Furthermore, after the step of integrating and identifying the perception data in the perception time window, the intelligent perception system is configured as follows:

[0093] Determining whether there is a human body feature data dimension reflecting the existence characteristics of a human body in the perception feature data matrix, wherein the human body feature data dimension corresponds to one or more perception data channels;

[0094] When there is no human feature data dimension reflecting the existence characteristics of the human body in the perception feature data matrix, discard the perception feature data matrix in the current perception time window, and return to the step of reading perception data from each perception data channel;

[0095] When there is a human feature data dimension reflecting the existence characteristics of the human body in the perception feature data matrix, it is executed to determine whether the perception data in the perception time window is learning data or decision data and its subsequent steps.

[0096] In the technical scheme of the above-mentioned implementation mode, when there is no human feature data dimension reflecting the characteristics of the human body in the perception feature data matrix, it can be determined that there is no human body around the water dispenser, and therefore there are no people with a tendency to approach water around the water dispenser. In this case, the perception data read from each perception data channel of the perception layer of the intelligent perception system has no learning or decision-making value, and there is no need to perform storage or analysis operations on it. After discarding the perception feature data matrix in the current perception time window, the step of reading perception data from each perception data channel is returned without executing the step of determining whether the perception data in the perception time window is learning data or decision data and its subsequent steps.

[0097] Specifically, the perception feature data matrix is ​​composed of feature data extracted from each perception data channel, and each data dimension in the perception feature data matrix corresponds to a perception data channel. Therefore, when the feature data in a perception data channel is human feature data, the perception data channel is a perception data channel that can reflect the existence characteristics of the human body. Correspondingly, the data dimension corresponding to the perception data channel in the perception feature data matrix is ​​the human feature data dimension that reflects the existence characteristics of the human body.

[0098] Taking the visual perception data channel as an example, the visual perception data in the perception time window is composed of several frames of image data. In the step of extracting feature data of corresponding perception types from the perception data of each associated perception data channel in the perception time window, it specifically includes extracting human limb feature data and / or facial feature data from the image data. When human limb feature data and / or facial feature data exist in these several frames of image data, the data dimension corresponding to the visual perception data channel in the perception feature data matrix is ​​determined to be the human body feature data dimension.

[0099] Similarly, the sound perception data in the perception time window is sound data having a time length equal to the size of the perception time window. The step of extracting feature data of the corresponding perception type from the perception data of each associated perception data channel in the perception time window specifically includes extracting speech feature data from the sound data. When speech feature data exists in the sound data, it is determined that the data dimension corresponding to the sound perception data channel in the perception feature data matrix is ​​the human body feature data dimension.

[0100] The above only gives an example of extracting human feature data from the perception data in a perception data channel to determine whether the data dimension corresponding to the perception data channel in the perception feature data matrix is ​​a human feature data dimension. In the actual implementation process, the data features of the perception data in multiple perception data channels can be combined to perform human body recognition. When the data features of multiple perception data channels can reflect the existence characteristics of the human body, the data dimensions corresponding to the multiple perception data channels in the perception feature data matrix are all determined as human feature data dimensions. For example, human body recognition is performed by combining ultrasonic perception data features with infrared perception data features, or by combining visual perception data features with infrared perception data features.

[0101] Further, in the step of determining whether the perception data in the perception time window is learning data or decision data, the intelligent perception system is configured as follows:

[0102] Determining an interactive perception data channel, wherein the interactive perception data channel includes a sound perception data channel and a touch perception data channel;

[0103] Determine whether the data of the interactive perception data channel in the perception time window is null or zero;

[0104] When the data of the interactive perception data channel in the perception time window are all null values ​​or all zero, determining that the perception data in the perception time window is decision data;

[0105] Otherwise, it is determined that the perception data in the perception time window is learning data.

[0106] It should be noted that the interactive perception data channels in the above embodiments are only examples, and the interactive perception data channels are different depending on the type of interactive sensor used by the water dispenser. However, for each specific water dispenser, the interactive perception data channels contained in its perception layer are determined. Therefore, in the control method of the intelligent perception system provided by the present invention, before the step of reading perception data from each perception data channel, it also includes configuring the interactive perception data channel of the water dispenser.

[0107] Furthermore, after the step of persisting the learning data in the learning sample database, the intelligent perception system is configured as follows:

[0108] Configure the model deviation test cycle;

[0109] Extracting test data from the learning sample database in each model deviation test cycle to perform a deviation test on the decision model, wherein the test data is learning data whose generation time is less than a preset time threshold;

[0110] When the deviation test result of the decision model is greater than a preset deviation threshold, the decision model is retrained using the learning data in the learning sample database.

[0111] People's drinking habits are usually relatively stable, especially in a home or office environment, where personnel changes are usually not very frequent, and most people maintain relatively regular drinking habits in their daily lives and work. In this case, the model deviation test cycle can be configured to be a relatively long time. Of course, when the system has sufficient computing resources and the amount of newly generated learning data is large, configuring a smaller model deviation test cycle can enable the decision model to have more accurate decision-making capabilities.

[0112] In the technical solution of the above-mentioned implementation mode, the generation time of the learning data is specifically the difference between the current time and the time when the learning data is written into the learning sample database. The current time referred to here refers to the time when the intelligent perception system executes the step of extracting test data from the learning sample database to perform deviation testing on the decision model in each model deviation test cycle.

[0113] Furthermore, in the step of extracting test data from the learning sample database to perform deviation testing on the decision model in each model deviation testing cycle, the intelligent perception system is configured as follows:

[0114] Determining an input data dimension and an output data dimension in the perception feature data matrix of the test data, wherein the output data dimension is a data dimension corresponding to the interactive perception data channel, and the input data dimension is other data dimensions in the perception feature data matrix except the output data dimension;

[0115] Constructing an input data matrix based on the data of the input data dimension;

[0116] Merging the perception data of the output data dimension into first output data, wherein the first output data is a one-dimensional data sequence;

[0117] Inputting the input data matrix into the decision model to obtain second output data, wherein the second output data is a one-dimensional data sequence, and the first output data and the second output data are interactive feature data;

[0118] calculating the similarity between the first output data and the second output data;

[0119] A deviation test result of the decision model is determined according to the similarity between the first output data and the second output data.

[0120] In the technical solutions of some embodiments of the present invention, the step of constructing an input data matrix based on the data of the input data dimension is specifically to generate the input data matrix using the mapping relationship between the feature data of the input data dimension and the input data matrix. More specifically, after the feature data in the input data dimension is standardized and normalized, multiple equal-length data sequences are obtained by numerical filling, and these equal-length data sequences are merged into the input data matrix.

[0121] The decision model is a deep learning model for realizing feature data classification, and the input data matrix is ​​composed of feature data extracted from the perception data obtained from multiple perception data channels within the perception time window. The first output data and the second output data are interactive feature data, that is, the first output data and the second output data have the same data format, and the same data format described here specifically means that the two data sequences have the same data length, and the data elements at corresponding positions in the two data sequences represent the same interactive features. More specifically, each data element in the data sequence of the first output data and the second output data is a classification label of an interactive feature, and each interactive feature is used to represent a specific interactive control instruction or a specific interactive control parameter. In the interactive feature data, different classification labels of each interactive feature correspond to different instruction states of the interactive control instruction, or different parameter values ​​or parameter ranges of the interactive control parameter. In the step of merging the perception data of the output data dimension into the first output data, the discrete interactive feature data of multiple output data dimensions are mapped to a one-dimensional data sequence according to a pre-configured mapping rule to obtain the first output data.

[0122] In the technical solutions of some embodiments of the present invention, the step of calculating the similarity between the first output data and the second output data is specifically to calculate the Euclidean distance or the Manhattan distance between the two discrete data sequences. Of course, other algorithms for calculating the similarity of discrete data sequences may also be used, such as calculating the cosine similarity of the two data sequences after vectorizing them.

[0123] In the technical solutions of some embodiments of the present invention, the step of determining the deviation test result of the decision model according to the similarity between the first output data and the second output data is specifically to directly determine the inverse of the similarity between the first output data and the second output data as the deviation test result data of the decision model, or to map the inverse of the similarity between the first output data and the second output data to a specific numerical range, for example, within the range of [0,1], and determine the mapped numerical value as the deviation test result data of the decision model.

[0124] Further, in the step of inputting the decision data into the decision model so that the decision model outputs the interactive feature data, the intelligent perception system is configured as follows:

[0125] Determining input data dimensions in a perceptual feature data matrix of the decision data;

[0126] Constructing an input data matrix based on the data of the input data dimension;

[0127] The input data matrix is ​​input into the decision model to obtain the interaction feature data, where the interaction feature data is a one-dimensional data sequence.

[0128] Specifically, since the data dimensions of the corresponding interactive perception data channels in the perception feature data matrix of the decision data are all null values ​​or zero, in the step of determining the input data dimensions in the perception feature data matrix of the decision data, the input data dimensions can be determined very intuitively and conveniently.

[0129] Furthermore, the data dimension corresponding to the interactive perception data channel in the perception feature data matrix includes interactive control instruction execution condition identification features. Before the step of persisting the learning data in the learning sample database, the intelligent perception system is configured as follows:

[0130] Acquire water supply parameter sensing data from a water supply parameter sensing data channel;

[0131] Determining, according to the water supply parameter sensing data, whether the water dispenser has the conditions to execute the interactive control instruction based on the interactive control parameter;

[0132] The interactive control instruction execution condition identification feature is configured according to a judgment result of whether the water dispenser has the conditions for executing the interactive control instruction based on the interactive control parameter.

[0133] Specifically, the interactive control instruction execution condition identification feature is used to identify feature data of whether the water dispenser has the conditions to execute the interactive control instruction based on the interactive control parameters, which is specifically a classification label of having the interactive control instruction execution condition or not having the interactive control instruction execution condition. In the step of configuring the interactive control instruction execution condition identification feature according to the judgment result of whether the water dispenser has the conditions to execute the interactive control instruction based on the interactive control parameters, when the water dispenser has the conditions to execute the interactive control instruction based on the interactive control parameters, the interactive control instruction execution condition identification feature is configured as a classification label corresponding to having the interactive control instruction execution condition. Conversely, when the water dispenser does not have the conditions to execute the interactive control instruction based on the interactive control parameters, the interactive control instruction execution condition identification feature is configured as a classification label corresponding to not having the interactive control instruction execution condition.

[0134] In the technical solution of the above implementation, in the step of performing the user interaction action according to the interaction feature data, the intelligent perception system is configured as follows:

[0135] Parsing the interactive control instruction execution condition identification feature from the interactive feature data;

[0136] Determine whether to execute the interactive control instruction based on the interactive control parameter according to the interactive control instruction execution condition identification feature.

[0137] In the technical solutions of some embodiments of the present invention, the water supply parameter sensing data channel includes a water level sensing data channel for determining the amount of water remaining in the water storage container, a pressure sensing data channel for determining whether a water cup exists, etc. In the technical solutions of these embodiments, the step of judging whether the water dispenser has the conditions for executing the interactive control instruction based on the interactive control parameter according to the environmental state sensing data specifically includes judging whether the amount of water remaining in the water storage container is sufficient and whether a water cup exists below the water outlet, etc.

[0138] In the technical solutions of other embodiments of the present invention, in the step of performing the user interaction action according to the interaction feature data, the intelligent perception system is configured as follows:

[0139] Parsing the interactive feature data to obtain interactive control instructions and interactive control parameters of the water dispenser;

[0140] Acquire water supply parameter sensing data from a water supply parameter sensing data channel;

[0141] Determining, according to the water supply parameter sensing data, whether the water dispenser has the conditions to execute the interactive control instruction based on the interactive control parameter;

[0142] When the water dispenser has the condition to execute the interactive control instruction based on the interactive control parameter, the interactive control instruction is executed according to the interactive control parameter.

[0143] In the technical scheme of the above-mentioned implementation mode, the data dimension corresponding to the interactive perception data channel in the perception feature data matrix does not include the interactive control instruction execution condition identification feature. After the decision model outputs the interactive feature data, the step of judging whether the water dispenser has the conditions to execute the interactive control instruction based on the interactive control parameters is performed according to the water supply parameter perception data, so as to determine whether it is necessary to execute the interactive control instruction according to the interactive control parameters.

[0144] like Figure 2 As shown, the second aspect of the present invention proposes a control method for an intelligent sensing system applied to a water dispenser, comprising:

[0145] Specifically, the perception time window is a specific length of time. The intelligent perception system of the smart water dispenser perceives and identifies human behavior, human state, environmental state, water supply parameters and interactive events by integrating the perception data of each perception data channel in the perception time window. Since the perception data required for perceiving and identifying human behavior, human state, environmental state, water supply parameters and interactive events are different, and the perception and identification methods are different, the size of the corresponding perception time window will also be different accordingly. It is necessary to configure the size of the corresponding perception time window for each perception type.

[0146] For example, human behavior is usually manifested as one or more actions that last for a certain period of time. The recognition of human behavior requires the support of perception data of a perception time window that lasts for a longer period of time. However, some specific environmental states, such as ambient temperature and humidity, are instantaneous. The recognition of these environmental states only requires one instantaneous perception data. Therefore, the perception time window corresponding to environmental state perception can be configured as a smaller value.

[0147] In the technical solution of the present invention, the learning data is the perception data including the interactive events, and the decision data is the perception data not including the interactive events. The intelligent perception system uses the learning data in the learning sample database as the training sample data of the decision model to continuously train and optimize the decision model, so that the decision model can use the perception data of the perception layer to accurately output interactive decisions.

[0148] Specifically, the perception time window is a specific length of time. The intelligent perception system of the smart water dispenser perceives and identifies human behavior, human state, environmental state, water supply parameters and interactive events by integrating the perception data of each perception data channel in the perception time window. Since the perception data required for perceiving and identifying human behavior, human state, environmental state, water supply parameters and interactive events are different, and the perception and identification methods are different, the size of the corresponding perception time window will also be different accordingly. It is necessary to configure the size of the corresponding perception time window for each perception type.

[0149] For example, human behavior is usually manifested as one or more actions that last for a certain period of time. The recognition of human behavior requires the support of perception data of a perception time window that lasts for a longer period of time. However, some specific environmental states, such as ambient temperature and humidity, are instantaneous. The recognition of these environmental states only requires one instantaneous perception data. Therefore, the perception time window corresponding to environmental state perception can be configured as a smaller value.

[0150] In the technical solution of the present invention, the learning data is the perception data including the interactive events, and the decision data is the perception data not including the interactive events. The intelligent perception system uses the learning data in the learning sample database as the training sample data of the decision model to continuously train and optimize the decision model, so that the decision model can use the perception data of the perception layer to accurately output interactive decisions.

[0151] Furthermore, the step of integrating and identifying the perception data in the perception time window specifically includes:

[0152] Configure the associated sensing data channel of each sensing type and the feature type corresponding to each associated sensing data channel;

[0153] Extracting feature data of corresponding perception types from the perception data of each associated perception data channel in the perception time window;

[0154] When there is no corresponding feature data in the perception data of any perception data channel, the feature data of the corresponding perception data channel is configured as a null value or configured as zero;

[0155] Otherwise, the feature data of the corresponding perception data channel is standardized and normalized;

[0156] Construct a perceptual feature data matrix of the perceptual time window.

[0157] In the technical solution of the present invention, the associated perception data channels corresponding to each perception type are pre-configured, for example, the associated perception data channels of human behavior perception include one or more of visual perception data channels, sound perception data channels, and infrared perception data channels, and the associated perception data channels of environmental state perception include one or more of visual perception data channels, temperature perception data channels, and Internet data channels, etc. It should be known that the associated perception data channels corresponding to the above perception types are only examples, and in specific implementations, different associated perception data channels corresponding to each perception type can be configured as needed.

[0158] Furthermore, for different perception types, they can be associated with the same perception data channel, but the corresponding feature types in the same perception data channel may be different. For example, for human behavior perception, the data features extracted from the visual data channel may include head feature data, hand feature data and other limb feature data. For human state perception, the data features extracted from the visual data channel may include face feature data. For environmental state perception, the data features extracted from the visual data channel may include environmental brightness feature data.

[0159] In the step of integrating and identifying the perception data in the perception time window, a perception feature data matrix is ​​constructed for each perception type. The perception feature data matrices of all perception types are n-dimensional matrices, where n is the number of perception data channels in the perception layer. In the perception feature data matrix, data of non-associated perception data channels of the current perception type are all configured as null values ​​or zeros.

[0160] The step of persisting the learning data in the learning sample database specifically includes:

[0161] Obtaining timestamp information of the perception time window, the timestamp information including one or more of the upper boundary time, the lower boundary time, and any time point therebetween, that is, the timestamp information may include one or more time point information, which may be the upper boundary time of the perception time window, the upper and lower boundary times of the perception time window, or any time point between the upper and lower boundary times of the perception time window;

[0162] The perception feature data matrix in the perception time window is associated with the timestamp information and saved in a learning sample database.

[0163] Furthermore, after the step of integrating and identifying the perception data in the perception time window, the method further includes:

[0164] Determining whether there is a human body feature data dimension reflecting the existence characteristics of a human body in the perception feature data matrix, wherein the human body feature data dimension corresponds to one or more perception data channels;

[0165] When there is no human feature data dimension reflecting the existence characteristics of the human body in the perception feature data matrix, discard the perception feature data matrix in the current perception time window, and return to the step of reading perception data from each perception data channel;

[0166] When there is a human feature data dimension reflecting the existence characteristics of the human body in the perception feature data matrix, it is executed to determine whether the perception data in the perception time window is learning data or decision data and its subsequent steps.

[0167] In the technical scheme of the above-mentioned implementation mode, when there is no human feature data dimension reflecting the characteristics of the human body in the perception feature data matrix, it can be determined that there is no human body around the water dispenser, and therefore there are no people with a tendency to approach water around the water dispenser. In this case, the perception data read from each perception data channel of the perception layer of the intelligent perception system has no learning or decision-making value, and there is no need to perform storage or analysis operations on it. After discarding the perception feature data matrix in the current perception time window, the step of reading perception data from each perception data channel is returned without executing the step of determining whether the perception data in the perception time window is learning data or decision data and its subsequent steps.

[0168] Specifically, the perception feature data matrix is ​​composed of feature data extracted from each perception data channel, and each data dimension in the perception feature data matrix corresponds to a perception data channel. Therefore, when the feature data in a perception data channel is human feature data, the perception data channel is a perception data channel that can reflect the existence characteristics of the human body. Correspondingly, the data dimension corresponding to the perception data channel in the perception feature data matrix is ​​the human feature data dimension that reflects the existence characteristics of the human body.

[0169] Taking the visual perception data channel as an example, the visual perception data in the perception time window is composed of several frames of image data. In the step of extracting feature data of corresponding perception types from the perception data of each associated perception data channel in the perception time window, it specifically includes extracting human limb feature data and / or facial feature data from the image data. When human limb feature data and / or facial feature data exist in these several frames of image data, the data dimension corresponding to the visual perception data channel in the perception feature data matrix is ​​determined to be the human body feature data dimension.

[0170] Similarly, the sound perception data in the perception time window is sound data having a time length equal to the size of the perception time window. The step of extracting feature data of the corresponding perception type from the perception data of each associated perception data channel in the perception time window specifically includes extracting speech feature data from the sound data. When speech feature data exists in the sound data, it is determined that the data dimension corresponding to the sound perception data channel in the perception feature data matrix is ​​the human body feature data dimension.

[0171] The above only gives an example of extracting human feature data from the perception data in a perception data channel to determine whether the data dimension corresponding to the perception data channel in the perception feature data matrix is ​​a human feature data dimension. In the actual implementation process, the data features of the perception data in multiple perception data channels can be combined to perform human body recognition. When the data features of multiple perception data channels can reflect the existence characteristics of the human body, the data dimensions corresponding to the multiple perception data channels in the perception feature data matrix are all determined as human feature data dimensions. For example, human body recognition is performed by combining ultrasonic perception data features with infrared perception data features, or by combining visual perception data features with infrared perception data features.

[0172] Furthermore, the step of determining whether the perception data in the perception time window is learning data or decision data specifically includes:

[0173] Determining an interactive perception data channel, wherein the interactive perception data channel includes a sound perception data channel and a touch perception data channel;

[0174] Determine whether the data of the interactive perception data channel in the perception time window is null or zero;

[0175] When the data of the interactive perception data channel in the perception time window are all null values ​​or all zero, determining that the perception data in the perception time window is decision data;

[0176] Otherwise, it is determined that the perception data in the perception time window is learning data.

[0177] It should be noted that the interactive perception data channels in the above embodiments are only examples, and the interactive perception data channels are different depending on the type of interactive sensor used by the water dispenser. However, for each specific water dispenser, the interactive perception data channels contained in its perception layer are determined. Therefore, in the control method of the intelligent perception system provided by the present invention, before the step of reading perception data from each perception data channel, it also includes configuring the interactive perception data channel of the water dispenser.

[0178] Furthermore, after the step of persisting the learning data in the learning sample database, the method further includes:

[0179] Configure the model deviation test cycle;

[0180] Extracting test data from the learning sample database in each model deviation test cycle to perform a deviation test on the decision model, wherein the test data is learning data whose generation time is less than a preset time threshold;

[0181] When the deviation test result of the decision model is greater than a preset deviation threshold, the decision model is retrained using the learning data in the learning sample database.

[0182] People's drinking habits are usually relatively stable, especially in a home or office environment, where personnel changes are usually not very frequent, and most people maintain relatively regular drinking habits in their daily lives and work. In this case, the model deviation test cycle can be configured to be a relatively long time. Of course, when the system has sufficient computing resources and the amount of newly generated learning data is large, configuring a smaller model deviation test cycle can enable the decision model to have more accurate decision-making capabilities.

[0183] In the technical solution of the above-mentioned implementation mode, the generation time of the learning data is specifically the difference between the current time and the time when the learning data is written into the learning sample database. The current time referred to here refers to the time when the intelligent perception system executes the step of extracting test data from the learning sample database to perform deviation testing on the decision model in each model deviation test cycle.

[0184] Furthermore, in each model deviation test cycle, the step of extracting test data from the learning sample database to perform deviation test on the decision model specifically includes:

[0185] Determining an input data dimension and an output data dimension in the perception feature data matrix of the test data, wherein the output data dimension is a data dimension corresponding to the interactive perception data channel, and the input data dimension is other data dimensions in the perception feature data matrix except the output data dimension;

[0186] Constructing an input data matrix based on the data of the input data dimension;

[0187] Merging the perception data of the output data dimension into first output data, wherein the first output data is a one-dimensional data sequence;

[0188] Inputting the input data matrix into the decision model to obtain second output data, wherein the second output data is a one-dimensional data sequence, and the first output data and the second output data are interactive feature data;

[0189] calculating the similarity between the first output data and the second output data;

[0190] A deviation test result of the decision model is determined according to the similarity between the first output data and the second output data.

[0191] In the technical solutions of some embodiments of the present invention, the step of constructing an input data matrix based on the data of the input data dimension is specifically to generate the input data matrix using the mapping relationship between the feature data of the input data dimension and the input data matrix. More specifically, after the feature data in the input data dimension is standardized and normalized, multiple equal-length data sequences are obtained by numerical filling, and these equal-length data sequences are merged into the input data matrix.

[0192] The decision model is a deep learning model for realizing feature data classification, and the input data matrix is ​​composed of feature data extracted from the perception data obtained from multiple perception data channels within the perception time window. The first output data and the second output data are interactive feature data, that is, the first output data and the second output data have the same data format, and the same data format described here specifically means that the two data sequences have the same data length, and the data elements at corresponding positions in the two data sequences represent the same interactive features. More specifically, each data element in the data sequence of the first output data and the second output data is a classification label of an interactive feature, and each interactive feature is used to represent a specific interactive control instruction or a specific interactive control parameter. In the interactive feature data, different classification labels of each interactive feature correspond to different instruction states of the interactive control instruction, or different parameter values ​​or parameter ranges of the interactive control parameter. In the step of merging the perception data of the output data dimension into the first output data, the discrete interactive feature data of multiple output data dimensions are mapped to a one-dimensional data sequence according to a pre-configured mapping rule to obtain the first output data.

[0193] In the technical solutions of some embodiments of the present invention, the step of calculating the similarity between the first output data and the second output data is specifically to calculate the Euclidean distance or the Manhattan distance between the two discrete data sequences. Of course, other algorithms for calculating the similarity of discrete data sequences may also be used, such as calculating the cosine similarity of the two data sequences after vectorizing them.

[0194] In the technical solutions of some embodiments of the present invention, the step of determining the deviation test result of the decision model according to the similarity between the first output data and the second output data is specifically to directly determine the inverse of the similarity between the first output data and the second output data as the deviation test result data of the decision model, or to map the inverse of the similarity between the first output data and the second output data to a specific numerical range, for example, within the range of [0,1], and determine the mapped numerical value as the deviation test result data of the decision model.

[0195] Furthermore, the step of inputting the decision data into the decision model so that the decision model outputs the interactive feature data specifically includes:

[0196] Determining input data dimensions in a perceptual feature data matrix of the decision data;

[0197] Constructing an input data matrix based on the data of the input data dimension;

[0198] The input data matrix is ​​input into the decision model to obtain the interaction feature data, where the interaction feature data is a one-dimensional data sequence.

[0199] Specifically, since the data dimensions of the corresponding interactive perception data channels in the perception feature data matrix of the decision data are all null values ​​or zero, in the step of determining the input data dimensions in the perception feature data matrix of the decision data, the input data dimensions can be determined very intuitively and conveniently.

[0200] Furthermore, the data dimension corresponding to the interactive perception data channel in the perception feature data matrix includes interactive control instruction execution condition identification features, and before the step of persisting the learning data in the learning sample database, it also includes:

[0201] Acquire water supply parameter sensing data from a water supply parameter sensing data channel;

[0202] Determining, according to the water supply parameter sensing data, whether the water dispenser has the conditions to execute the interactive control instruction based on the interactive control parameter;

[0203] The interactive control instruction execution condition identification feature is configured according to a judgment result of whether the water dispenser has the conditions for executing the interactive control instruction based on the interactive control parameter.

[0204] Specifically, the interactive control instruction execution condition identification feature is used to identify feature data of whether the water dispenser has the conditions to execute the interactive control instruction based on the interactive control parameters, which is specifically a classification label of having the interactive control instruction execution condition or not having the interactive control instruction execution condition. In the step of configuring the interactive control instruction execution condition identification feature according to the judgment result of whether the water dispenser has the conditions to execute the interactive control instruction based on the interactive control parameters, when the water dispenser has the conditions to execute the interactive control instruction based on the interactive control parameters, the interactive control instruction execution condition identification feature is configured as a classification label corresponding to having the interactive control instruction execution condition. Conversely, when the water dispenser does not have the conditions to execute the interactive control instruction based on the interactive control parameters, the interactive control instruction execution condition identification feature is configured as a classification label corresponding to not having the interactive control instruction execution condition.

[0205] In the technical solution of the above implementation, the step of executing the user interaction action according to the interaction feature data specifically includes:

[0206] Parsing the interactive feature data to obtain the interactive control instruction execution condition identification feature and the interactive control instruction and interactive control parameter of the water dispenser;

[0207] Determine whether to execute the interactive control instruction based on the interactive control parameter according to the interactive control instruction execution condition identification feature.

[0208] In the technical solutions of some embodiments of the present invention, the water supply parameter sensing data channel includes a water level sensing data channel for determining the amount of water remaining in the water storage container, a pressure sensing data channel for determining whether a water cup exists, etc. In the technical solutions of these embodiments, the step of judging whether the water dispenser has the conditions for executing the interactive control instruction based on the interactive control parameter according to the environmental state sensing data specifically includes judging whether the amount of water remaining in the water storage container is sufficient and whether a water cup exists below the water outlet, etc.

[0209] In the technical solutions of other embodiments of the present invention, the step of executing the user interaction action according to the interaction feature data specifically includes:

[0210] Parsing the interactive feature data to obtain interactive control instructions and interactive control parameters of the water dispenser;

[0211] Acquire water supply parameter sensing data from a water supply parameter sensing data channel;

[0212] Determining, according to the water supply parameter sensing data, whether the water dispenser has the conditions to execute the interactive control instruction based on the interactive control parameter;

[0213] When the water dispenser has the condition to execute the interactive control instruction based on the interactive control parameter, the interactive control instruction is executed according to the interactive control parameter.

[0214] In the technical scheme of the above-mentioned implementation mode, the data dimension corresponding to the interactive perception data channel in the perception feature data matrix does not include the interactive control instruction execution condition identification feature. After the decision model outputs the interactive feature data, the step of judging whether the water dispenser has the conditions to execute the interactive control instruction based on the interactive control parameters is performed according to the water supply parameter perception data, so as to determine whether it is necessary to execute the interactive control instruction according to the interactive control parameters.

[0215] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0216] According to the embodiments of the present invention as described above, these embodiments do not describe all the details in detail, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and changes can be made based on the above description. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can make good use of the present invention and the modified use based on the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. An intelligent sensing system applied to a water dispenser, characterized in that: The intelligent sensing system comprises a sensing layer for acquiring sensing data, a data layer for processing, identifying and storing sensing data, a decision layer for autonomous learning or interactive decision-making using sensing data, and an interactive layer for interacting with users. The sensing layer comprises a human sensing module for sensing human behavior and human state, an environmental sensing module for sensing environmental state, a water supply sensing module for sensing water supply parameters, and an event sensing module for sensing interactive events. The data layer comprises a data processing module for performing data standardization, normalization and numerical filling, a data feature extraction module for performing data feature identification, and a learning sample database for storing learning data. The decision layer comprises a decision model for outputting interactive feature data according to decision data. The interactive layer comprises an interactive feature data parsing module for parsing interactive control instructions and interactive control parameters of a water dispenser from interactive feature data, and an interactive control instruction execution module for executing the interactive control instructions according to the interactive control parameters. The intelligent sensing system is configured as follows: Configuring the size of the corresponding perception time window of each perception type, wherein the perception types include human behavior perception, human state perception, environmental state perception, water supply parameter perception, and interactive event perception; Reading perception data from each perception data channel, the perception data including at least one of visual perception data, ultrasonic perception data, sound perception data, infrared perception data, flow perception data, temperature perception data, and touch perception data; Integrating and identifying the perception data in the perception time window; Determining whether the perception data in the perception time window is learning data or decision data, wherein the learning data is used to train a decision model, and the decision data is used to input the decision model so that the decision model outputs interactive feature data; When the perception data in the perception time window is learning data, persisting the learning data in a learning sample database; When the perception data in the perception time window is decision data, inputting the decision data into the decision model so that the decision model outputs interaction feature data; Execute a user interaction action according to the interaction feature data; The step of integrating and identifying the perception data in the perception time window includes constructing a perception feature data matrix of the perception time window; After the steps of integrating and identifying the perception data in the perception time window, the intelligent perception system is configured to: Determining whether there is a human body feature data dimension reflecting the existence characteristics of a human body in the perception feature data matrix, wherein the human body feature data dimension corresponds to one or more perception data channels; When there is no human feature data dimension reflecting the existence characteristics of the human body in the perception feature data matrix, discard the perception feature data matrix in the current perception time window, and return to the step of reading perception data from each perception data channel; When there is a human feature data dimension reflecting the existence characteristics of the human body in the perception feature data matrix, it is executed to determine whether the perception data in the perception time window is learning data or decision data and its subsequent steps.

2. A control method for an intelligent sensing system applied to a water dispenser, characterized in that: include: Configuring the size of the corresponding perception time window of each perception type, wherein the perception types include human behavior perception, human state perception, environmental state perception, water supply parameter perception, and interactive event perception; Reading perception data from each perception data channel, the perception data including at least one of visual perception data, ultrasonic perception data, sound perception data, infrared perception data, flow perception data, temperature perception data, and touch perception data; Integrating and identifying the perception data in the perception time window; Determining whether the perception data in the perception time window is learning data or decision data, wherein the learning data is used to train a decision model, and the decision data is used to input the decision model so that the decision model outputs interactive feature data; When the perception data in the perception time window is learning data, persisting the learning data in a learning sample database; When the perception data in the perception time window is decision data, inputting the decision data into the decision model so that the decision model outputs interaction feature data; Execute a user interaction action according to the interaction feature data; The step of integrating and identifying the perception data in the perception time window includes constructing a perception feature data matrix of the perception time window; After the step of integrating and identifying the perception data in the perception time window, the method further includes: Determining whether there is a human body feature data dimension reflecting the existence characteristics of a human body in the perception feature data matrix, wherein the human body feature data dimension corresponds to one or more perception data channels; When there is no human feature data dimension reflecting the existence characteristics of the human body in the perception feature data matrix, discard the perception feature data matrix in the current perception time window, and return to the step of reading perception data from each perception data channel; When there is a human feature data dimension reflecting the existence characteristics of the human body in the perception feature data matrix, it is executed to determine whether the perception data in the perception time window is learning data or decision data and its subsequent steps.

3. The control method of the intelligent sensing system applied to a water dispenser according to claim 2 is characterized in that: The steps of integrating and identifying the perception data in the perception time window specifically include: Configure the associated sensing data channel of each sensing type and the feature type corresponding to each associated sensing data channel; Extracting feature data of corresponding perception types from the perception data of each associated perception data channel in the perception time window; When there is no corresponding feature data in the perception data of any perception data channel, the feature data of the corresponding perception data channel is configured as a null value or configured as zero; Otherwise, the feature data of the corresponding perception data channel is standardized and normalized.

4. The control method of the intelligent sensing system applied to a water dispenser according to claim 3 is characterized in that: The step of determining whether the perception data in the perception time window is learning data or decision data specifically includes: Determining an interactive perception data channel, wherein the interactive perception data channel includes a sound perception data channel and a touch perception data channel; Determine whether the data of the interactive perception data channel in the perception time window is null or zero; When the data of the interactive perception data channel in the perception time window are all null values ​​or all zero, determining that the perception data in the perception time window is decision data; Otherwise, it is determined that the perception data in the perception time window is learning data.

5. The control method of the intelligent sensing system applied to a water dispenser according to claim 4, characterized in that: After the step of persisting the learning data in the learning sample database, the method further includes: Configure the model deviation test cycle; Extracting test data from the learning sample database in each model deviation test cycle to perform a deviation test on the decision model, wherein the test data is learning data whose generation time is less than a preset time threshold; When the deviation test result of the decision model is greater than a preset deviation threshold, the decision model is retrained using the learning data in the learning sample database.

6. The control method of the intelligent sensing system applied to a water dispenser according to claim 5, characterized in that: The step of extracting test data from the learning sample database to perform deviation testing on the decision model in each model deviation testing cycle specifically includes: Determining an input data dimension and an output data dimension in the perception feature data matrix of the test data, wherein the output data dimension is a data dimension corresponding to the interactive perception data channel, and the input data dimension is other data dimensions in the perception feature data matrix except the output data dimension; Constructing an input data matrix based on the data of the input data dimension; Merging the perception data of the output data dimension into first output data, wherein the first output data is a one-dimensional data sequence; Inputting the input data matrix into the decision model to obtain second output data, wherein the second output data is a one-dimensional data sequence, and the first output data and the second output data are interactive feature data; calculating the similarity between the first output data and the second output data; A deviation test result of the decision model is determined according to the similarity between the first output data and the second output data.

7. The control method of the intelligent sensing system applied to a water dispenser according to claim 3, characterized in that: The step of inputting the decision data into the decision model so that the decision model outputs the interactive feature data specifically includes: Determining input data dimensions in a perceptual feature data matrix of the decision data; Constructing an input data matrix based on the data of the input data dimension; The input data matrix is ​​input into the decision model to obtain the interaction feature data, where the interaction feature data is a one-dimensional data sequence.

8. The control method of the intelligent sensing system applied to a water dispenser according to claim 7, characterized in that: The data dimension corresponding to the interactive perception data channel in the perception feature data matrix includes the interactive control instruction execution condition identification feature, and before the step of persisting the learning data in the learning sample database, it also includes: Acquire water supply parameter sensing data from a water supply parameter sensing data channel; Determining, according to the water supply parameter sensing data, whether the water dispenser has the conditions to execute the interactive control instruction based on the interactive control parameter; The interactive control instruction execution condition identification feature is configured according to a judgment result of whether the water dispenser has the conditions for executing the interactive control instruction based on the interactive control parameter.

9. The control method of the intelligent sensing system applied to a water dispenser according to claim 7, characterized in that: The step of performing a user interaction action according to the interaction feature data specifically includes: Parsing the interactive feature data to obtain interactive control instructions and interactive control parameters of the water dispenser; Acquire water supply parameter sensing data from a water supply parameter sensing data channel; Determining, according to the water supply parameter sensing data, whether the water dispenser has the conditions to execute the interactive control instruction based on the interactive control parameter; When the water dispenser has the condition to execute the interactive control instruction based on the interactive control parameter, the interactive control instruction is executed according to the interactive control parameter.

Images