Cat food feeding amount prediction method and system based on deep residual shrinkage memory network

Abstract

The application discloses a cat food feeding amount prediction method and system based on a deep residual shrinkage memory network, and the prediction method comprises the following steps: S1, a controller collects data of a plurality of sensors to establish a historical database; S2, data processing is performed on the data in the historical database; S3, the processed historical database is input into the deep residual shrinkage memory network for network training; S4, the trained model is deployed to the controller, real-time data is collected by the controller to predict the cat food feeding amount; and when the cat food feeding time arrives, the predicted cat food amount is fed. According to the application, pet attribute information data, pet food feeding data, defecation information data and other information are collected by a plurality of sensors as training data of a model, then a deep residual shrinkage memory network model is trained to learn the excretion habit and eating habit of a cat, so that the accurate and reasonable cat food feeding amount that meets the current pet healthy eating state is predicted, and the cat is fed in a healthy and scientific manner.

Description

Technical Field

[0001] This invention belongs to the field of pet food feeding technology, and in particular relates to a method and system for predicting cat food feeding amount based on deep residual shrinkage memory network. Background Technology

[0002] Current cat feeding methods primarily involve timed and measured feedings. However, this method fails to accurately control the amount of cat food given based on the cat's individual health condition, leading to problems such as obesity. Furthermore, inconsistent food intake results in either insufficient food intake (leaving the cat uneaten) or excessive food intake (leading to waste). All of these factors negatively impact the cat's health. Some researchers have undertaken technological improvements to address these issues.

[0003] Chinese patent application CN116034897A discloses a pet food dispensing control method, including: collecting pet attribute information data, pet food data, and pet environment data; determining the daily pet food dispensing amount and food collection interval based on the pet attribute information data, pet food data, and pet environment data; determining the single-time pet food dispensing amount and pet food feeding time based on the set daily pet feeding duration, the pet's first feeding time, the daily dispensing amount, and the food collection interval; and controlling the dispensing of food into the pet feeding container in a feeding device based on the single-time pet food dispensing amount and pet food feeding time. This method improves feeding accuracy to some extent through certain calculations. However, the accuracy of the data calculated by this method is not high; it relies on habitual experience and is not updated in real time, making it unable to provide reasonable cat food dispensing based on the cat's habits. Summary of the Invention

[0004] The purpose of this invention is to provide a cat food feeding amount prediction method based on deep residual shrinkage memory network, so as to solve the problem of low accuracy in existing pet feeding amount prediction.

[0005] To address the above problems, this invention discloses a method for predicting cat food feeding amounts based on a deep residual shrinking memory network, comprising the following steps:

[0006] S1. The controller collects data from several sensors to establish a historical database. The data from these sensors includes the cat's weight, temperature and humidity data in the litter box, and the cat's toileting data. The toileting data includes the type of feces and urine, the number of times of feces and urine, the time of feces and urine, and the interval between toileting sessions. The controller organizes the total number of feces and urine before feeding the cat, the interval between two consecutive feces or urine sessions, the weight data, and the amount of cat food fed, and establishes a historical database to store the above data.

[0007] S2. Data processing is performed on the historical database. Since the dimensions and numerical ranges of the data vary, directly using the raw data to train the network will cause noise to mask smaller features in the sequence, and will also increase the difficulty of network convergence. Therefore, normalization is used to linearly transform the input data to [0,1] to standardize the data and improve the convergence speed and learning effect of the neural network layers, defined as follows:

[0008] ;

[0009] Where X represents the original data. This represents the normalized data. and These represent the maximum and minimum values ​​in the sequence, respectively.

[0010] S3. Input the processed historical database into the deep residual shrinking memory network for network training; including:

[0011] A wide convolutional layer is used as the first layer of the deep residual shrinking memory network;

[0012] A deep residual shrinking module is added after the wide convolutional layer. The deep residual shrinking module finds the interference characteristics of the input samples based on the attention mechanism and sets them to zero using the soft thresholding function, thereby reducing the impact of noise interference on the prediction effect. The soft thresholding operation relationship is as follows:

[0013] ;

[0014] Where x is the input feature, y is the output feature, and τ is the threshold;

[0015] The attention mechanism introduces a lightweight sub-network to automatically learn the importance of each feature channel and assign a weight to each channel accordingly. By using different weights, important features can be strengthened and redundant features weakened.

[0016] A neural network layer is added to the deep residual shrinkage module.

[0017] Next, evaluate the model performance:

[0018] MSE: It measures the average of the squared errors and is given by the following equation:

[0019] ;

[0020] MAPE: It is the average or mean of the absolute percentage error of the predictions.

[0021] ;

[0022] R 2The coefficient of determination measures how much one variable can explain the change in another variable when predicting the outcome of an event. The system calculates it using the following formula:

[0023] ;

[0024] in For predicted values, This is the actual value. It is the average of the predicted values.

[0025] S4. Deploy the trained model to the controller, which collects real-time data to predict the amount of cat food to be fed; when the cat food feeding time arrives, feed the predicted amount of cat food.

[0026] This invention also discloses a cat food feeding prediction system based on a deep residual shrinkage memory network, including a data acquisition module and a data processing module. The data acquisition module includes a cat weight data acquisition unit, a litter box temperature and humidity data acquisition unit, and a cat littering data acquisition unit.

[0027] The data processing module includes a historical database, a primary data processing unit, and a network training unit. The data from the data acquisition module is stored in the historical database. The primary data processing unit performs primary processing on the data in the historical database. The network training unit performs network training on the primary-processed data in the historical database.

[0028] The primary data processing unit performs primary processing on the data in the historical database, including: normalization processing, linearly transforming the input data to [0,1] to standardize the data and improve the convergence speed and learning effect of the neural network layer, as defined in the following formula:

[0029] ;

[0030] Where X represents the original data. This represents the normalized data. and These represent the maximum and minimum values ​​in the sequence, respectively.

[0031] The network training unit performs network training on data that has undergone preliminary processing in the historical database, including:

[0032] A wide convolutional layer is used as the first layer of the deep residual shrinking memory network;

[0033] A deep residual shrinking module is added after the wide convolutional layer. The deep residual shrinking module finds the interference characteristics of the input samples based on the attention mechanism and sets them to zero using the soft thresholding function, thereby reducing the impact of noise interference on the prediction effect. The soft thresholding operation relationship is as follows:

[0034] ;

[0035] Where x is the input feature, y is the output feature, and τ is the threshold;

[0036] The attention mechanism introduces a lightweight sub-network to automatically learn the importance of each feature channel and assign a weight to each channel accordingly. By using different weights, important features can be strengthened and redundant features weakened.

[0037] A neural network layer is added to the deep residual shrinkage module.

[0038] Compared with the prior art, the beneficial effects of the present invention are:

[0039] This invention uses data collected from multiple sensors, including pet attribute information, pet food feeding data, and toilet information, as training data for a model. Then, it trains a deep residual shrinking memory network model to learn the cat's excretion and eating habits, thereby predicting an accurate and reasonable amount of cat food to feed the cat under the current healthy eating conditions, so that the cat can receive healthy and scientific feeding. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the prediction method of the present invention;

[0041] Figure 2 This is a basic module diagram of the present invention;

[0042] Figure 3 This is a diagram of the neural network layers of the present invention;

[0043] Figure 4 This is a network architecture diagram of the deep residual shrinkage memory network of the present invention. Detailed Implementation

[0044] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0045] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," "outer," "front end," "rear end," "both ends," "one end," and "the other end," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0046] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installed," "equipped with," "connected," etc., should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] This invention discloses a method for predicting cat food feeding amount based on a deep residual shrinking memory network, comprising the following steps:

[0049] S1. The controller collects data from several sensors to establish a historical database. The data from these sensors includes the cat's weight, temperature and humidity data in the litter box, and the cat's toileting data. The toileting data includes the type of feces and urine, the number of times of feces and urine, the time of feces and urine, and the interval between toileting sessions. The controller organizes the total number of feces and urine before feeding the cat, the interval between two consecutive feces or urine sessions, the weight data, and the amount of cat food fed, and establishes a historical database to store the above data.

[0050] S2. Data processing is performed on the historical database. Because the data collected by sensors contains noise and the dimensions and numerical ranges of different data types vary, directly using the raw data to train the network will mask smaller features in the sequence, and the convergence difficulty will increase during network training. Normalization is used to linearly transform the input data to [0,1] to standardize the data and improve the convergence speed and learning effect of the neural network layers, defined as follows:

[0051] ;

[0052] Where X represents the original data. This represents the normalized data. and These represent the maximum and minimum values ​​in the sequence, respectively.

[0053] S3. Input the processed historical database into the deep residual shrinking memory network for network training; including:

[0054] A wide convolutional layer is used as the first layer of the deep residual shrinking memory network. The core idea of ​​the wide convolutional layer is to improve the model's robustness to noise by using larger convolutional kernels. Large convolutional kernels have a larger receptive field, thus ignoring noise interference in the data.

[0055] A deep residual shrinking module is added after the wide convolutional layer. The deep residual shrinking module finds the interference characteristics of the input samples based on the attention mechanism and sets them to zero using the soft thresholding function, thereby reducing the impact of noise interference on the prediction effect. The soft thresholding operation relationship is as follows:

[0056] ;

[0057] Where x is the input feature, y is the output feature, and τ is the threshold;

[0058] The attention mechanism introduces a lightweight sub-network to automatically learn the importance of each feature channel and assign a weight to each channel accordingly. Different weights can be used to strengthen important features and weaken redundant features; basic modules include... Figure 2 As shown in the figure. Where C, W, and H represent the number of channels, width, and height of the input data, respectively.

[0059] A neural network layer is added to the deep residual shrinkage module to enhance the model's tracking system's dynamic characteristics, effectively learn long-term dependencies, and improve the network's temporal feature extraction. The core design of the neural network layer includes three gates: an input gate, a forget gate, and an output gate. Figure 3 As shown.

[0060] Neural network layers include input gates Forgotten Gate and output gate Its structure includes:

[0061] ;

[0062] ;

[0063] ;

[0064] ;

[0065] ;

[0066] ;

[0067] , ; Represents a neural network layer cell; Input and output at any given moment; Cell state at time t; Cell state at time t-1; ; ; Indicates input gate Weight matrix; This represents the input gate weight matrix; Indicates the input gate bias term; Represents the cell weight matrix; Represents cells Weight matrix; Indicates cell bias term; Represents the forget gate weight matrix; Represents the Gate of Oblivion matrix; This represents the cell weight matrix at time t-1, representing the forgetting gate. Indicates the forget gate bias term; This represents the output gate weight matrix; Indicates output gate Weight matrix; This represents the output gate cell weight matrix; Indicates the output gate bias term; Activation function, hyperbolic tangent function.

[0068] After the model is built, historical data is input into the model for training. The historical data table is as follows:

[0069] Table 1 Network Input / Output Data

[0070]

[0071] To comprehensively evaluate model performance, MSE, R2, and MAPE were used as three performance metrics during model training to assess the predictive power of the proposed wide convolutional depth residual shrinkage memory network.

[0072] MSE: It measures the average of the squared errors and is given by the following equation:

[0073] ;

[0074] MAPE: It is the average or mean of the absolute percentage error of the predictions.

[0075] ;

[0076] R 2 The coefficient of determination measures how much one variable can explain the change in another variable when predicting the outcome of an event. The system calculates it using the following formula:

[0077] ;

[0078] in, Predicted value Actual value The average of the predicted values.

[0079] S4. Deploy the trained model to the controller, which collects real-time data to predict the amount of cat food to be fed; when the cat food feeding time arrives, feed the predicted amount of cat food.

[0080] This invention uses data collected from multiple sensors, including pet attribute information, pet food feeding data, and toilet information, as training data for a model. Then, it trains a deep residual shrinking memory network model to learn the cat's excretion and eating habits, thereby predicting an accurate and reasonable amount of cat food to feed the cat under the current healthy eating conditions, so that the cat can receive healthy and scientific feeding.

[0081] The above descriptions are merely embodiments of the present invention, and common knowledge regarding specific structures and characteristics of the solutions is not described in detail here. It will be apparent to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the present invention is defined by the appended claims rather than the foregoing description. Therefore, it is intended that all variations falling within the meaning and scope of equivalents of the claims be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for predicting cat food feeding amount based on deep residual shrinking memory network, characterized in that, Includes the following steps: S1. The controller collects data from several sensors to establish a historical database; S2. Perform data processing on the historical database, including: normalization to linearly transform the input data to [0,1] to standardize the data and improve the convergence speed and learning effect of the neural network, as defined in the following formula: ； Where X represents the original data. This represents the normalized data. and These represent the maximum and minimum values ​​in the sequence, respectively. S3. Input the processed historical database into the deep residual shrinking memory network for network training; including: A wide convolutional layer is used as the first layer of the deep residual shrinking memory network; A deep residual shrinking module is added after the wide convolutional layer. The deep residual shrinking module finds the interference characteristics of the input samples based on the attention mechanism and sets them to zero using the soft thresholding function, thereby reducing the impact of noise interference on the prediction effect. The soft thresholding operation relationship is as follows: ； Where x is the input feature, y is the output feature, and τ is the threshold; The attention mechanism introduces a lightweight sub-network to automatically learn the importance of each feature channel and assign a weight to each channel accordingly. By using different weights, important features can be strengthened and redundant features weakened. A neural network layer is added to the deep residual shrinkage module; the neural network layer includes an input gate. Forgotten Gate and output gate Its structure includes: ； , ; Represents a neural network layer cell; Input and output at any given moment; Cell at time t; Cells at time t-1; ; ; Indicates input gate Weight matrix; Represents the input gate weight matrix; Indicates the input gate bias term; Represents the cell weight matrix; Represents cells Weight matrix; Indicates cell bias term; Represents the forget gate weight matrix; Represents the Gate of Oblivion matrix; This represents the cell weight matrix at time t-1, representing the forgetting gate. Indicates the forget gate bias term; This represents the output gate weight matrix; Indicates output gate Weight matrix; This represents the output gate cell weight matrix; Indicates the output gate bias term; Activation function, hyperbolic tangent function; S4. Deploy the trained model to the controller, which collects real-time data to predict the amount of cat food to be fed; when the cat food feeding time arrives, feed the predicted amount of cat food.

2. The cat food feeding amount prediction method based on deep residual shrinking memory network according to claim 1, characterized in that, The data from the various sensors include cat weight data, litter box temperature and humidity data, and cat toileting data. The toileting data includes the type of feces and urine, the number of feces and urine, the time of feces and urine, and the interval between toileting sessions. The controller organizes the total number of feces and urine before feeding the cat food, the interval between two consecutive feces or urine sessions, the weight data, and the amount of cat food fed, and establishes a historical database to store the above data.

3. The cat food feeding amount prediction method based on deep residual shrinking memory network according to claim 2, characterized in that, The step of inputting the processed data into a deep residual shrinking memory network for network training also includes: evaluating model performance. MSE: The average of the squares of the measurement error, given by the following equation: ; MAPE: is the average or mean of the absolute percentage error of the predictions. ; R 2 The coefficient of determination measures how much one variable can explain the change in another variable when predicting the outcome of an event. The system calculates it using the following formula: ; in For predicted values, This is the actual value. It is the average of the predicted values.

4. A prediction system based on the cat food feeding amount prediction method using the deep residual shrinking memory network as described in claim 1, characterized in that: It includes a data acquisition module and a data processing module. The data acquisition module includes a cat weight data acquisition unit, a litter box temperature and humidity data acquisition unit, and a cat littering data acquisition unit. The data processing module includes a historical database, a primary data processing unit, and a network training unit. The data from the data acquisition module is stored in the historical database. The primary data processing unit performs primary processing on the data in the historical database. The network training unit performs network training on the primary-processed data in the historical database.

Images