Intelligent pet management system based on artificial intelligence

Abstract

The invention relates to the technical field of management systems, and particularly discloses an intelligent pet management system based on artificial intelligence, which comprises a data acquisition module, a data storage module, a data analysis module, a management control module and a man-machine interaction module, the data acquisition module synchronously acquires multi-dimensional data through a multi-type sensor group, wherein the multi-dimensional data comprises pet behavior data, physiological data and environmental data; after triple processing is executed by the data preprocessing unit, a structured data frame is generated and transmitted to the data storage module; the data storage module adopts a MongoDB distributed database architecture and stores data according to a timestamp, a data type and a pet ID three-dimensional index structure. Through the closed-loop design of comprehensive perception, intelligent analysis, personalized management, efficient interaction and continuous optimization, the refined and intelligent level of pet management is remarkably improved, full-period guarantee is provided for pet health, and meanwhile the management cost and operation complexity of a user are reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of management systems, in particular to an intelligent pet management system based on artificial intelligence. BACKGROUND

[0002] With the increase of pet-keeping rate, the demand for fine and intelligent pet management is increasingly prominent. The existing pet management methods have many limitations: Single and one-sided data collection: Traditional management relies on manual observation and recording, or only collects data through a single sensor (such as a simple activity pedometer), which cannot fully capture the pet's behavior patterns (such as movement trajectory, posture change), physiological state (such as heart rate fluctuation, abnormal body temperature) and environmental influence (such as temperature and humidity, air quality), resulting in blind spots in the assessment of the pet's health status; Weak analysis capability and lack of predictability: Existing systems mostly use simple statistical methods to process data, which can only reflect the real-time state and cannot mine the pet's behavior patterns and health trends from historical data. For example, it cannot early warn of the risk of abnormal heart rate in senior pets or predict changes in the activity needs of puppies, making it difficult to achieve preventive management; Homogenization of management solutions and poor adaptability: Existing solutions are mostly based on general rules (such as fixed feeding amount and unified activity suggestions), without considering the species differences (such as different physiological needs of dogs and cats), age stages (such as different exercise intensity needs of puppies and senior dogs) and individual habits (such as some pets being sensitive to environmental temperature), resulting in poor management effectiveness; Inefficient human-computer interaction and delayed response: The interaction mode of traditional systems is single (such as only through indicator lights), making it difficult for users to intuitively obtain detailed pet status; the alarm mechanism for abnormal conditions (such as sudden temperature rise) is not perfect, often resulting in delayed notification or unclear information, which may delay the intervention opportunity; Insufficient system iteration capability: There is a lack of data storage and model optimization mechanism, which cannot continuously improve the analysis accuracy based on massive user data, and is prone to evaluation deviation after long-term use, making it difficult to adapt to the changes in the pet's state during its growth. SUMMARY

[0003] To overcome the deficiencies of the prior art, the present application provides an intelligent pet management system based on artificial intelligence, which solves the problems in the background art.

[0004] To achieve the above purpose, the present application realizes the following technical scheme: an intelligent pet management system based on artificial intelligence, comprising a data collection module, a data storage module, a data analysis module, a management control module and a human-computer interaction module. The data acquisition module synchronously acquires multi-dimensional data including pet behavior data, physiological data and environmental data through a plurality of types of sensor groups; after three times of processing by a data preprocessing unit, a structured data frame is generated and transmitted to a data storage module; The data storage module adopts a MongoDB distributed database architecture, stores data according to a three-dimensional index structure of time stamp, data type and pet ID, supports millisecond-level parallel retrieval, the single-node data storage capacity is ≥10 TB, and has a remote disaster recovery backup function; The data analysis module calls an artificial intelligence algorithm to deeply analyze the stored data, first performs dimensionality reduction processing through principal component analysis, then uses an attention mechanism to weight and fuse behavior and physiological characteristics, inputs a pre-trained CNN-LSTM hybrid model, and outputs a health status score and a future 24-hour behavior trend probability distribution; The management control module generates a multi-dimensional management scheme based on the health status score and behavior trend prediction, and executes the scheme in real time through an RS485 bus driving execution unit; The human-computer interaction module constructs a bidirectional interaction channel of hardware and software, integrates a pickup and a speaker on the hardware side, and deploys a mobile APP on the software side to realize visual display of the management scheme, real-time response of user instructions and hierarchical push of abnormal information.

[0005] Preferably, the data acquisition module comprises: The behavior sensing unit combines a UWB positioning sensor and a three-axis acceleration sensor to collect pet motion trajectory coordinates and motion acceleration values at a sampling rate of 10 Hz and motion acceleration values After Kalman filtering optimization, a 128×128 behavior pattern feature matrix is generated, with rows representing time steps and columns representing feature dimensions. The physiological sensing unit integrates a MAX30102 heart rate sensor and a SHT30 temperature and humidity sensor to transmit a continuous physiological parameter sequence through Bluetooth BLE5.0; The environmental sensing unit is equipped with an MQ-135 air quality sensor and a DS18B20 temperature sensor to collect environmental parameters of the pet's activity area at a sampling rate of 1 Hz and generate an environmental quality index.

[0006] Preferably, the deep learning algorithm of the data analysis module performs the following operations: Behavior data processing: sliding window is used to divide time series data to extract 12-dimensional behavior features including motion frequency, stay area proportion, posture conversion times, motion trajectory complexity, average motion speed, active period proportion, high-frequency motion duration, vertical direction motion proportion, specific area stay duration, motion interval mean, posture holding duration and sudden turning times, wherein motion frequency = number of motion state frames in the window / total number of frames × 60; Physiological data processing: time series alignment of heart rate and body temperature data, calculation of 8 physiological characteristic indexes including heart rate variability, body temperature fluctuation amplitude, average heart rate, heart rate fluctuation range, body temperature mean value, heart rate abnormality times, body temperature peak occurrence time and respiratory frequency variability, where HRV≤50ms is determined as heart rate abnormality; Feature fusion and model inference: 12-dimensional behavior features and 8 physiological features are spliced into a 20-dimensional fusion feature vector, which is input into a CNN-LSTM hybrid model, and the health state score and future 24-hour behavior trend probability distribution are output through a Softmax activation function, each time slice containing the probability values of active / quiet / feeding / sleeping 4 types of behavior.

[0007] Preferably, the model training of the deep learning algorithm uses a loss function: ; Wherein, is the total loss function; is the number of training samples; is the true health score of the sample; is the model prediction score; is the model weight parameter matrix; is the L2 regularization coefficient.

[0008] Preferably, the management control module calculates the pet health score The formula is: ; Wherein, is the behavior score, 0-100 points: calculated based on exercise frequency, activity duration compliance rate, and posture diversity, with exercise frequency≥5 times / minute counted as full score; is the physiological score, 0-100 points: calculated based on heart rate and body temperature, with heart rate within the normal range of the species and body temperature fluctuation amplitude≤0.5℃ counted as full score; is the environmental adaptation score, 0-100 points: calculated based on environmental temperature, humidity, and air quality, with temperature in the range of 20-26℃, humidity in the range of 40%-60%, and air quality index≤100 counted as full score; is the weight coefficient , with initial values of 0.3, 0.5, and 0.2, respectively, and user-defined adjustment supported.

[0009] Preferably, the weight coefficient is dynamically adjusted according to the pet species and age, satisfying: ; Wherein, is the behavior score weight function: Puppy ≤1 year, Adult dog 1-8 years, Senior dog >8 years, ; Kitten ≤1 year, Adult cat 1-7 years, Senior cat >7 years, ; The physiological score weight function: Puppy Adult dog Senior dog ; Kitten Adult cat Senior cat ; The environmental score weight function: all species in the juvenile stage Adult stage Old age stage .

[0010] Preferably, the management control module generates an optimization objective function of the management scheme: ; Wherein: is the utility function, , is the management cycle, default 24 hours, , ; The health benefit function: , Every 1 point, 2 utility values / hour; The management cost function: , wherein is the feeding cost, 2 utility values per gram of food , is the equipment energy consumption cost, 2 utility values per kilowatt hour .

[0011] Preferably, the human-computer interaction module comprises: Voice interaction unit, using the ASR speech recognition engine of Kedou Xunfei, receiving user voice instructions, instruction response delay ≤1s; Text interaction unit, through the APP interface to display structured data: health score curve, diet suggestion list, activity plan, support user text input, input response delay ≤500ms; The feedback unit feeds back the system execution result in a double-channel feedback mode: voice feedback and APP push notification, and the priority is divided into three levels: common, important and urgent.

[0012] Preferably, the intelligent pet management system based on artificial intelligence further comprises a cloud server, which is used for: Storing historical pet data: structured data is stored by using a MySQL database, unstructured data is stored by using a MongoDB, nearly 3 years of data are retained, incremental backup is supported, and the data compression rate is greater than or equal to 50%; Based on a federal learning framework, the artificial intelligence model of the data analysis module is continuously optimized, the number of participating nodes is greater than or equal to 100, the model aggregation period is 24 hours, and the optimization process follows a parameter updating rule: ; Wherein: is the model parameter of the i-th iteration; is the learning rate, and the initial value is 0.001, and exponential decay is adopted. ; is the gradient of the loss function at the i-th iteration.

[0013] Preferably, the management control module has a real-time alarm function, when the abnormal state judgment condition is met, the man-machine interaction module is triggered to send an alarm to the user: ; Wherein: is an abnormal state indicator, indicates an abnormality; is a health score threshold: 60 points for dogs, 55 points for cats, and 50 points for small mammals, which can be manually adjusted by plus or minus 10 points; Physiological parameter safety range: Heart rate: 60-140 times / minute for dogs, 120-140 times / minute for cats, and 200-300 times / minute for rabbits; Body temperature: 38.0-39.0 DEG C for dogs, 38.5-39.5 DEG C for cats, and 38.5-40.0 DEG C for rabbits; Alarm form: When the health score is less than or equal to the threshold, the "urgent" level push and telephone notification are triggered, and the APP interface flashes a warning light.

[0014] The application provides an intelligent pet management system based on artificial intelligence, which has the following beneficial effects: ​1. Multi-dimensional data acquisition, realizing comprehensive perception: Through multi-type sensor group synchronous acquisition of pet behavior data (motion trajectory, posture conversion, etc.), physiological data (heart rate, body temperature, etc.) and environmental data (temperature and humidity, air quality, etc.), and pre-processing (denoising, standardization) to ensure data quality, complete input is provided for accurate analysis, solving the problem of traditional data acquisition.

[0015] 2. AI deep analysis, improving evaluation and prediction ability: Using a CNN-LSTM hybrid model to fuse behavior and physiological features, not only can output real-time health score, but also can predict future 24-hour behavior trend (such as active / feeding period probability); Through principal component analysis dimension reduction and attention mechanism weighting, the pertinence of feature extraction is enhanced, solving the limitations of traditional analysis methods that cannot mine data correlation and lack of predictability.

[0016] 3. Dynamic adaptive management, realizing personalized solution: Based on the species and age of the pet, the health score weight is dynamically adjusted (such as higher weight for physiological parameters of senior dogs), and through optimization of the objective function (maximizing health benefits and cost balance), a customized management plan is generated (such as high-frequency feeding plan for puppies, low-humidity environment adjustment for cats), solving the problem of traditional homogeneous solutions.

[0017] 4. Efficient human-computer interaction, strengthening real-time response: Build a "hardware and software" two-way interaction channel, through voice recognition (supporting Chinese and English instructions), APP visual display (health curve, diet list) and hierarchical push (ordinary / urgent alarm), so that users can intuitively grasp the pet state; Trigger "urgent push + telephone notification" in abnormal situation (such as heart rate out of safe range), ensure the timeliness of intervention, solve the problem of low efficiency and slow response of traditional interaction.

[0018] 5. Cloud collaborative optimization, guaranteeing system adaptability: Through federated learning framework to continuously optimize AI model, combined with distributed database to store historical data (retain 3 years of records), so that the system can continuously improve analysis accuracy as the pet grows and user needs change; Off-site disaster recovery backup function ensures data security, solves the problem of insufficient iteration ability and poor adaptability of traditional systems.

[0019] In summary, through the closed-loop design of comprehensive perception, intelligent analysis, personalized management, efficient interaction and continuous optimization, the invention significantly improves the fine and intelligent level of pet management, provides whole-cycle guarantee for pet health, and reduces the management cost and operation complexity of users. BRIEF DESCRIPTION OF DRAWINGS

[0020] Fig. 1 A principle block diagram of an intelligent pet management system based on artificial intelligence is described in the invention; Fig. 2A principle block diagram of a data acquisition module in an intelligent pet management system based on artificial intelligence according to the present application; Fig. 3 A principle block diagram of a man-machine interaction module in an intelligent pet management system based on artificial intelligence according to the present application. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0022] As Figs. 1-3 shown, the present application provides a technical solution: an intelligent pet management system based on artificial intelligence, comprising: a data acquisition module, a data storage module, a data analysis module, a management control module and a man-machine interaction module. The data acquisition module synchronously acquires multi-dimensional data through multiple types of sensor groups, including pet behavior data (motion trajectory coordinates, activity duration, posture change frequency), physiological data (heart rate, body surface temperature, respiratory rate) and environmental data (activity area temperature and humidity, air quality index, light intensity); after 3-fold processing (Gaussian filter denoising, Z-score standardization, time sequence alignment and completion) performed by the data preprocessing unit, a structured data frame (each frame containing 64-dimensional feature values) is generated and transmitted to the data storage module; the data storage module adopts a MongoDB distributed database architecture, stores data according to a three-dimensional index structure of time stamp, data type (behavior / physiology / environment) and pet ID, supports millisecond-level parallel retrieval, the single-node data storage capacity is ≥10TB, and has the function of off-site disaster recovery backup; the data analysis module calls artificial intelligence algorithms to deeply analyze the stored data, first reduces the dimension through principal component analysis (retaining 95% of the characteristic variance), then uses the attention mechanism to weight and fuse behavior and physiological characteristics, inputs a pre-trained CNN-LSTM hybrid model (containing 3 convolution layers and 2 LSTM layers), and outputs a health status score (0-100 points, with an interval of 1 point) and a future 24-hour behavior trend probability distribution (15-minute granularity, a total of 96 time slices); the management control module generates a multi-dimensional management scheme (diet plan containing calorie ratio, activity suggestion containing intensity / duration, environmental regulation parameters) based on the health status score and behavior trend prediction, and drives the execution unit (automatic feeder precision ±5g, environmental regulator temperature control precision ±0.5℃) to execute in real time through the RS485 bus; the human-computer interaction module constructs a bidirectional interaction channel of hardware and software, the hardware end integrates a pickup and a speaker, and the software end deploys a mobile APP, realizing visual display of the management scheme (dynamic charts and text instructions), real-time response of user instructions and hierarchical push of abnormal information.

[0023] More specifically, as shown in Fig. 2 The data acquisition module includes a behavior sensing unit, a physiological sensing unit and an environmental sensing unit; the behavior sensing unit combines a UWB positioning sensor and a three-axis acceleration sensor to acquire pet motion trajectory coordinates and motion acceleration values , after Kalman filter optimization, generate a 128x128 behavior pattern feature matrix (rows represent time steps, columns represent feature dimensions); The physiological sensing unit integrates MAX30102 heart rate sensor (measurement range 30-250 times / minute, accuracy ±2 times / minute, sampling rate 100Hz) and SHT30 temperature and humidity sensor (synchronously collect body temperature, range 35-42℃, accuracy ±0.3℃, humidity measurement range 0-100%RH, accuracy ±2%RH), through Bluetooth BLE5.0 transmission to transmit continuous physiological parameter sequence (1 record generated every 500ms); The environmental sensing unit is equipped with MQ-135 air quality sensor (detects harmful gas concentration, resolution 0.1ppm, detection range 0-1000ppm) and DS18B20 temperature sensor (measurement range -55℃ to 125℃, accuracy ±0.5℃), to collect environmental parameters of the pet activity area at a sampling rate of 1Hz, and generate an environmental quality index (0-500, corresponding to excellent / good / medium / poor / severe).

[0024] The behavior sensing unit uses a combination of UWB positioning sensor (model DWM1000, positioning accuracy ±10cm) and three-axis acceleration sensor (model MPU6050, range ±16g), both connected through SPI bus, and synchronously collect data; collect at a sampling rate of 10Hz (i.e. 10 groups of data generated every second): motion trajectory coordinates (x, y, z, unit m): reflect the position change of the pet in three-dimensional space; motion acceleration values (a x ,aᵧ,a_z, unit m / s 2 ): represent the intensity and direction of the pet's motion; after Kalman filter algorithm processing (suppressing positioning drift and acceleration noise), generate a 128x128 behavior pattern feature matrix; the rows of the matrix correspond to time steps (every 0.1s), and the columns correspond to feature dimensions (including trajectory change rate, acceleration peak value, etc. derived features), which record the behavior dynamics of the pet.

[0025] The physiological sensing unit integrates MAX30102 heart rate sensor and SHT30 temperature and humidity sensor, both connected through I 2 C bus, sharing timing reference. Collect parameters: heart rate measurement range 30-250 times / minute, accuracy ±2 times / minute, sampling rate 100Hz (1 data point generated every 10ms), detected by infrared photoelectric volume method; synchronously collect body temperature, range 35-42℃, accuracy ±0.3℃; synchronously collect environmental humidity (range 0-100%RH, accuracy ±2%RH), to assist in determining whether the abnormal body temperature is related to the environment; package and send continuous physiological parameter sequence through Bluetooth BLE5.0 protocol (transmission rate 2Mbps), 1 summary record (including heart rate mean, body temperature value, etc. in this period) generated every 500ms, to reduce transmission redundancy.

[0026] The environmental sensing unit is equipped with an MQ-135 air quality sensor and a DS18B20 temperature sensor, which are independently powered to avoid interference. The collected parameters are: the MQ-135 sensor detects a range of 0-1000 ppm with a resolution of 0.1 ppm and can identify gases such as CO and formaldehyde; the DS18B20 sensor measures a range of -55°C to 125°C with an accuracy of ±0.5°C, covering extreme temperature scenarios in the pet activity area; the gas concentration and temperature data are mapped to an environmental quality index (0-500 points), which corresponds to the classification rules of "excellent (0-100), good (101-200), medium (201-300), poor (301-400), and severe (401-500)", and intuitively reflects environmental suitability.

[0027] The data preprocessing unit performs three-step processing on the raw data collected by the three types of sensing units: Gaussian filter denoising: smooth high-frequency interference (such as abnormal values caused by sensor jitter); Z-score standardization: convert different dimension data (such as acceleration m / s 2 and temperature °C) to standard normal distribution for subsequent fusion analysis; time series alignment and completion: linear interpolation is used to complete the missing data based on the system unified timestamp (accurate to 10 ms), and finally generate a structured data frame containing 64-dimensional feature values for each frame, which is transmitted to the data storage module through Ethernet.

[0028] More specifically, the deep learning algorithm of the data analysis module performs the following operations: Behavior data processing: sliding window is used to segment time series data, and 12-dimensional behavior features such as motion frequency, stay area proportion, posture conversion times, motion trajectory complexity, average motion speed, active period proportion, high-frequency motion duration, vertical direction motion proportion, specific area stay duration, motion interval mean, posture retention duration, and sudden turning times are extracted, where motion frequency = number of motion state frames in the window / total number of frames x 60; Physiological data processing: time series alignment is performed on heart rate and body temperature data based on the system timestamp, and 8 physiological feature indexes such as heart rate variability, body temperature fluctuation amplitude, average heart rate, heart rate fluctuation range, body temperature mean, heart rate abnormality times, body temperature peak occurrence time, and respiratory frequency variability are calculated, where HRV≤50 ms is determined as heart rate abnormality; Feature fusion and model inference: The 12-dimensional behavioral features and 8 physiological features are concatenated into a 20-dimensional fused feature vector, which is then input into a CNN-LSTM hybrid model (the convolutional layer uses 32 3×3 convolutional kernels, and the LSTM layer has 64 hidden units). The model outputs a health status score (0-100 points) and a probability distribution of behavioral trends for the next 24 hours after the softmax activation function (each time slice contains the probability values ​​of four types of behaviors: "active / quiet / eating / sleeping").

[0029] The data analysis module uses deep learning algorithms to process, fuse, and infer the collected behavioral and physiological data, outputting health status scores and behavioral trend prediction results. The specific operations are as follows: The specific operations for behavioral data processing are as follows: Sliding window segmentation: A sliding window with a size of 5 minutes and a step size of 1 minute is used to segment time-series behavioral data (motion trajectory coordinates, acceleration values, etc.). Each window corresponds to a continuous segment of behavioral records, ensuring that the temporal locality characteristics of the data are fully captured.

[0030] 12-dimensional behavioral feature extraction: Motion frequency: Calculated using the formula "number of motion frames within the window / total number of frames × 60" (unit: times / minute), where "motion frames" refers to frames with an acceleration value exceeding 0.5 m / s². 2 The sampling frames reflect the frequency of the pet's activities per unit time; Percentage of dwelling area: the percentage of time the pet spends in a preset fixed area (such as within 1m of the nest or food bowl) within the window (unit: %); Number of posture transitions: the cumulative number of posture transitions such as "standing-lying down-jumping" identified by changes in acceleration vectors; Trajectory complexity: the path tortuosity calculated based on trajectory coordinates (unit: none, the larger the value, the more complex the trajectory); Average movement speed: the ratio of the total displacement distance to time within the window (unit: m / s); Percentage of active periods: the percentage of time during which the acceleration value exceeds the threshold within the total window duration (unit: %); High-frequency movement duration: acceleration value ≥ 2m / s 2 Cumulative duration (in seconds); Vertical movement percentage: the proportion of movement distance along the z-axis to the total movement distance (in percent), reflecting the intensity of vertical activities such as jumping and climbing; Duration in specific areas: the cumulative time spent in key areas such as food bowls and litter boxes (in seconds); Average movement interval: the average time spent between two consecutive movements (acceleration > 0.5 m / s²). 2 Average time interval between (unit: s); Posture holding time: average duration of the same posture (e.g., lying down) (unit: s); Number of sudden turns: number of times the direction of movement changes by an angle ≥90° per unit time (unit: times / minute).

[0031] The specific operations for physiological data processing are as follows: Time alignment: Interpolate and align the heart rate (sampling rate 100 Hz) and body temperature (sampling rate 1 Hz) data based on the system unified timestamp (accurate to 10 ms), ensuring one-to-one correspondence in the time dimension.

[0032] 8 physiological feature indicators: Heart rate variability (HRV): calculated by the standard deviation of RR interval (the time interval between two consecutive heartbeats) (unit: ms), where HRV≤50 ms is determined as abnormal heart rate; Body temperature fluctuation amplitude: the difference between the maximum and minimum body temperature in the window (unit: °C); Average heart rate: the arithmetic mean of heart rate in the window (unit: beats / minute); Heart rate fluctuation range: the difference between the maximum and minimum heart rate in the window (unit: beats / minute); Body temperature mean: the arithmetic mean of body temperature in the window (unit: °C); Heart rate abnormality times: the cumulative number of heart rates exceeding the normal range of the species (e.g. 60-140 beats / minute for dogs) in the window; Body temperature peak occurrence time: the specific time point (accurate to seconds) when the body temperature reaches the maximum value in the window; Respiratory rate variability: the standard deviation of respiratory rate (synchronized with heart rate sensor) (unit: beats / minute).

[0033] The specific operation of feature fusion and model inference is as follows: Feature fusion: concatenate the 12-dimensional behavior features and the 8 physiological features by dimension to form a 20-dimensional fusion feature vector, realizing the correlation analysis of behavior and physiological state (e.g. "high frequency exercise duration + heart rate fluctuation range" jointly reflects the reasonableness of exercise intensity).

[0034] CNN-LSTM hybrid model structure: Convolutional layer uses 32 3x3 convolutional kernels to extract local features from the fusion feature vector (e.g. short-term heart rate and exercise correlation), outputting a 32x18 feature map; LSTM layer sets 64 hidden units to process the time series features output by the convolutional layer, capturing long-term dependencies (e.g. behavior-physiological change trend over several hours); the output layer outputs two classes of results through the Softmax activation function: Health status score: 0-100 points (interval 1 point), comprehensively reflecting the current health level of the pet; Future 24-hour behavior trend probability distribution: 15-minute time slices (96 in total), each time slice contains the probability values of "active / quiet / feeding / sleeping" 4 types of behavior (sum = 1), supporting forward-looking management.

[0035] More specifically, the model training of the deep learning algorithm uses the loss function: ; where, is the total loss function; The number of training samples (value range 10000-50000, single sample contains continuous 60 minutes of data, time interval 1s); The true health score of the sample (labeled by a veterinarian according to the "Pet Health Evaluation Standard", interval 1 minute); The model prediction score; The model weight parameter matrix (dimension 512x256); The L2 regularization coefficient (value 0.001, used to suppress overfitting).

[0036] The loss function used in the training of the deep learning model (CNN-LSTM hybrid model) of the data analysis module is the mean square error function with L2 regularization. Its core function is to measure the deviation of the model prediction value from the true value, and to suppress overfitting through regularization to ensure the generalization ability of the model on different pet samples. The specific implementation is as follows: The loss function is composed of the following formula: ; The first part (mean square error term): ; The function is to quantify the deviation of the model prediction value from the true value, forcing the model to learn the true law of the sample; wherein, is the number of training samples, with a value range of 10000-50000, and a single sample containing continuous 60 minutes of behavior, physiological and environmental data (time interval 1s, each sample containing 3600 data points); is the true health score of the i-th sample, manually labeled by a veterinarian according to the "Pet Health Evaluation Standard" (score 0-100, interval 1, such as "85" and "72"); is the predicted health score of the i-th sample by the model (output format consistent with ).

[0037] The second part (L2 regularization term): ; The function is to limit the absolute value of the parameter by penalizing the square sum of the model weight parameters, to avoid overfitting of the model to the noise or local features in the training data; wherein, is the model weight parameter matrix, with a dimension of 512x256 (covering CNN layer convolution kernel weight, LSTM layer hidden unit weight and full connection layer weight); is the L2 norm square of , i.e. the square sum of all parameters; is the regularization coefficient, with a value of 0.001 (experimentally verified that this value can achieve the optimal balance between bias and variance).

[0038] ​The computational logic during training is as follows: For each training batch (batch size set to 32), first calculate the mean squared error of all samples within that batch (part one), then add the regularization term (part two) to obtain the total loss L; calculate the loss function with respect to each parameter using the backpropagation algorithm. The gradient, according to the update rule Adjust the parameters until L converges to a stable minimum value (set the number of training rounds to 100, and stop training when the change in L is <0.001 for 10 consecutive rounds).

[0039] The mean squared error term ensures the model's prediction accuracy (making...). as close as possible Meanwhile, the model complexity is controlled by L2 regularization to avoid overfitting problems such as "high accuracy on the training set but low accuracy on the test set", ensuring that the model is reliable in predicting the health scores of pets of different species (dogs / cats / small mammals) and different ages.

[0040] More specifically, the management and control module calculates the pet's health score. The formula is: ; Rate your pet's health (0-100 points), with higher scores indicating better health. in, Behavioral scoring (0-100 points): Calculated based on movement frequency (weight 0.3), activity duration achievement rate (weight 0.4), and posture diversity (weight 0.3). A movement frequency of ≥5 times / minute is considered full marks. Physiological score (0-100 points): calculated based on heart rate (weight 0.5) and body temperature (weight 0.5). A score of 0.5 is given if the heart rate is within the normal range for the species and the body temperature fluctuation is ≤0.5℃. Environmental adaptability score (0-100 points): Calculated based on ambient temperature (weight 0.4), humidity (weight 0.3), and air quality (weight 0.3). A score of 100 is given for a temperature of 20-26℃, humidity of 40%-60%, and air quality index ≤100. Weighting coefficients The initial values ​​are 0.3, 0.5, and 0.2, respectively, and can be customized by the user.

[0041] Behavioral rating ( The calculation is based on a weighted average of three core indicators from the 12-dimensional behavioral characteristics: Exercise frequency (weight 0.3): The target is ≥5 times / minute (100 points for meeting the target, points are proportionally reduced for not meeting the target, e.g., 3 times / minute = 60 points); Activity duration target rate (weight 0.4): Calculate the actual activity duration percentage based on the recommended activity duration for pet species / age (e.g., 1 hour per day for adult dogs) (100% percentage = 100 points); Posture diversity (weight 0.3): Count the number of posture types such as "standing / lying down / walking / jumping" in the statistical window (≥5 types = 100 points, 20 points deducted for each missing type); Calculation formula: 0.3 Movement frequency score +0.4 Activity duration target achievement rate score +0.3 Posture diversity score.

[0042] Physiological score ( The calculation is based on a weighted average of two key physiological indicators: Heart rate (weight 0.5): 100 points are awarded if within the normal range for the species (e.g., 60-140 beats / minute for dogs); points are deducted for deviations beyond this range (e.g., 80 points for a dog's heart rate of 150 beats / minute); Body temperature (weight 0.5): 100 points are awarded for temperature fluctuations ≤0.5℃; 20 points are deducted for every 0.1℃ exceeding this range (e.g., 60 points for a fluctuation of 0.7℃). Calculation formula: .

[0043] Environment adaptability score ( The calculation is based on a weighted average of three core environmental parameters: Ambient temperature (weight 0.4): 100 points for a temperature range of 20-26℃, deducting 10 points for every 1℃ deviation (e.g., 18℃ = 80 points); Humidity (weight 0.3): 100 points for a humidity range of 40%-60%, deducting 10 points for every 5% deviation (e.g., 30% = 80 points); Air quality (weight 0.3): 100 points for an Air Quality Index (AQI) ≤ 100, 80 points for 101-200, 60 points for 201-300, and so on. Calculation formula: .

[0044] Weight adjustment mechanism, initial weights: This reflects the assessment logic of "prioritizing physiological state"; users can modify the weights through the human-computer interaction module (e.g., for post-operative pets, adjust the weights accordingly). Adjusted to 0.7 (strengthening the weighting of physiological states), the system automatically ensures... .

[0045] Through weighted fusion of sub-scores, It can accurately reflect the pet's real-time health level (such as...) The weight adjustment can adapt to different scenes (such as focusing on behavior score for young pets and focusing on physiological score for old pets), and provide quantitative basis for personalized generation of subsequent management solutions.

[0046] More specifically, the weight coefficient is dynamically adjusted according to the pet species and age, meeting the following conditions: ; Among them, The behavior score weight function is: Puppies ≤1 year old, Adult dogs 1-8 years old, Old dogs >8 years old, ; Kittens ≤1 year old, Adult cats 1-7 years old, Old cats >7 years old, ; Logic: Young pets have more active behavior, and the weight is higher than that of old pets.

[0047] The physiological score weight function is: Puppies Adult dogs Old dogs ; Kittens Adult cats Old cats ; Logic: Old pets have physiological function recession, and the weight of physiological indicators on health is higher.

[0048] The environment score weight function is: all species in the young stage Adult stage Old stage (The environment has equal influence on pets of different ages).

[0049] Logic: The environment has equal influence on the health of pets of different ages, and the weight remains stable.

[0050] More specifically, the management control module generates an optimization objective function for the management solution as follows: ; Among them: Utility function (unit: utility value), the higher the value, the better the solution, , Management cycle (default 24 hours, , ); Health benefit function: , Every 1 point of improvement, 2 utility values per hour; (e.g. hourly income of 160 utility values) is increased; The management cost function is: wherein is the feeding cost (160 utility values per gram of food ), is the equipment energy consumption cost (160 utility values per kilowatt hour ).

[0051] The system maximizes the 24-hour by adjusting the diet plan (feeding amount, frequency), activity recommendations (duration, intensity), and environmental parameters (temperature, humidity). For example, for a puppy , the activity duration is increased to improve , while the feeding amount is controlled to avoid excessive cost; for an elderly cat , the environmental temperature is optimized (improve and low-intensity activities (balance ) to reduce the risk of physiological index abnormalities.

[0052] More specifically, as shown in Fig. 3 , the human-computer interaction module includes: a voice interaction unit, a text interaction unit, and a feedback unit; the voice interaction unit uses the ASR speech recognition engine of Kedouxunfei (supports Chinese / English, recognition accuracy ≥ 96%, accuracy ≥ 90% in a noise environment (≤ 60dB)), receives user voice instructions (such as "query today's health report"), and the instruction response delay is ≤ 1s; the text interaction unit displays structured data through the APP interface: health score curve (horizontal axis time, vertical axis score), diet suggestion list (including food materials / amount / time), activity plan (including time period / intensity), supports user text input (such as "adjust feeding time to 18:00"), and the input response delay is ≤ 500ms; the feedback unit feeds back the system execution results (such as "feeding time has been adjusted to 18:00") in two channels: voice feedback (uses TTS speech synthesis technology, tone can be selected), APP push notification (priority is divided into "ordinary / important / emergency" three levels).

[0053] More specifically, the intelligent pet management system based on artificial intelligence further includes a cloud server, which is used for: storing historical pet data: structured data (health score, behavior record) is stored using MySQL database, unstructured data (sensor raw waveform) is stored using MongoDB, nearly 3 years of data are retained, incremental backup (automatic backup at 2:00 am every day) is supported, and the data compression rate is ≥ 50%; The artificial intelligence model of the data analysis module is continuously optimized based on the federated learning framework, the number of participating nodes is ≥100, the model aggregation cycle is 24 hours, and the optimization process follows the parameter update rule: ; Wherein: is the model parameter of the th iteration; is the learning rate, the initial value is 0.001, and the exponential decay ; is the gradient of the loss function at .

[0054] With the increase of the number of iterations, the prediction error of the model for the health score of different species (dogs / cats / small mammals), age pets is gradually reduced (target: test set error ≤3 points), and the system adaptability is improved. The cloud server realizes efficient management and safe backup of data through hierarchical storage strategy, and optimizes the model based on federated learning under the premise of protecting privacy, so that the system can continuously improve the analysis accuracy with the accumulation of pet samples.

[0055] More specifically, the management control module has a real-time alarm function, which triggers the human-computer interaction module to send an alarm to the user when the abnormal state judgment condition is met: ; Wherein: is an abnormal state indicator, indicates an abnormality; is the health score threshold: 60 points for dogs, 55 points for cats, and 50 points for small mammals (such as rabbits) (which can be manually adjusted ±10 points); Physiological parameter safety range: Heart rate: 60-140 beats per minute for dogs, 120-140 beats per minute for cats, and 200-300 beats per minute for rabbits; Body temperature: 38.0-39.0℃ for dogs, 38.5-39.5℃ for cats, and 38.5-40.0℃ for rabbits; Alarm form: When the abnormal state indicator is greater than the health score threshold, trigger the "emergency" level push and phone notification (repeat dialing within 10 minutes if not confirmed), and at the same time, the APP interface flashes a warning light.

[0056] The system executes three levels of alarms through the human-computer interaction module: APP emergency push: send a pop-up window notification to the user's mobile APP with the highest priority, which contains the abnormal type (such as "heart rate exceeds the safety range"), the current value (such as "heart rate 150 beats per minute"), and the recommended measures (such as "please check the pet's condition"); Phone notification: automatically dial the user's preset emergency contact phone and play a voice prompt (e.g., "Your pet is currently in an abnormal state, please check the APP details in time"); if no user confirmation (clicking the "Viewed" button in the APP) is received within 10 minutes, the phone will be dialed again every 5 minutes, up to a maximum of 3 times; Interface warning: the warning light at the top of the APP interface flashes continuously (red) until the user confirms or the abnormal state is resolved.

[0057] Through the dual judgment mechanism of "health score and physiological parameters", it is ensured that there is no omission of abnormal state; combined with multi-level alarm forms (push, phone and interface warning), the risk of intervention delay is significantly reduced, especially for sudden health problems of pets when they are alone (such as acute heart rate abnormalities, sudden temperature rise), which provides immediate protection for pet health.

[0058] Although embodiments of the present application have been shown and described, it will be understood by those having ordinary skill in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent pet management system based on artificial intelligence, characterized by, Comprise: Data acquisition module, data storage module, data analysis module, management control module and man-machine interaction module; Data acquisition module, through the synchronous acquisition of multi-type sensor group multi-dimensional data, including: pet behavior data, physiological data and environmental data; After 3 times of processing by the data preprocessing unit, structured data frames are generated and transmitted to the data storage module; Data storage module, using MongoDB distributed database architecture, storing data according to time stamp, data type and pet ID three-dimensional index structure, supporting millisecond-level parallel retrieval, single-node data storage capacity ≥10TB, and having off-site disaster recovery backup function; Data analysis module, calling artificial intelligence algorithm to deeply analyze the stored data, first reducing dimension through principal component analysis, then using attention mechanism to weight and fuse behavior and physiological characteristics, inputting pre-trained CNN-LSTM hybrid model, and outputting health status score and future 24-hour behavior trend probability distribution; Management control module, based on health status score and behavior trend prediction, generating multi-dimensional management scheme, and executing through RS485 bus driving execution unit in real time; Man-machine interaction module, building a hardware and software bidirectional interaction channel, hardware end integrating pickup and speaker, software end deploying mobile APP, realizing management scheme visual display, user instruction real-time response and abnormal information hierarchical push. 2.The intelligent pet management system based on artificial intelligence according to claim 1, wherein The data acquisition module comprises: The behavior sensing unit adopts a UWB positioning sensor and a three-axis acceleration sensor to collect pet motion trajectory coordinates at a sampling rate of 10 Hz and motion acceleration values After Kalman filtering optimization, a 128x128 behavior pattern feature matrix is generated, with rows representing time steps and columns representing feature dimensions. Physiological sensing unit, integrating MAX30102 heart rate sensor and SHT30 temperature and humidity sensor, transmitting continuous physiological parameter sequence through Bluetooth BLE5.0; Environmental sensing unit, equipped with MQ-135 air quality sensor and DS18B20 temperature sensor, collecting environmental parameters of pet activity area at 1Hz sampling rate, generating environmental quality index. 3.The intelligent pet management system based on artificial intelligence according to claim 2, characterized in that, The deep learning algorithm of the data analysis module performs the following operations: Behavior data processing: sliding window is used to divide time series data, and 12-dimensional behavior characteristics are extracted, including motion frequency, stay area ratio, posture conversion times, motion trajectory complexity, average motion speed, active period ratio, high-frequency motion time, vertical direction motion ratio, specific area stay time, motion interval mean, posture holding time, and sudden turning times, wherein motion frequency = motion state frame number in window / total frame number × 60; Physiological data processing: time series alignment is performed on heart rate and body temperature data, and 8 physiological characteristic indexes are calculated, including heart rate variability, body temperature fluctuation amplitude, average heart rate, heart rate fluctuation range, body temperature mean, heart rate abnormality times, body temperature peak occurrence time, and respiratory rate variability, wherein HRV≤50ms is determined as heart rate abnormality; Feature fusion and model inference: 12-dimensional behavior characteristics and 8 physiological characteristics are spliced into 20-dimensional fusion feature vector, which is input into CNN-LSTM hybrid model, and health status score and future 24-hour behavior trend probability distribution are output through Softmax activation function, each time slice containing probability values of active / quiet / feeding / sleeping 4 behaviors.

4. The intelligent pet management system based on artificial intelligence according to claim 3, characterized in that, The model training of the deep learning algorithm uses the loss function: ； wherein, is the total loss function; is the number of training samples; is the true health score of the sample; is the model predicted score; is the matrix of model weight parameters; is the L2 regularization coefficient. 5.The intelligent pet management system based on artificial intelligence according to claim 4, wherein The management control module calculates a pet health score The formula is: ； wherein, Behavior score, 0-100: calculated based on movement frequency, activity duration compliance rate, and posture diversity, movement frequency > 5 times / minute is counted as full score; For the physiological score, 0-100 points: based on heart rate, body temperature calculation, heart rate in the normal range of species and body temperature fluctuation amplitude ≤0.5℃ is full marks; Environmental adaptation score, 0-100 points: based on environmental temperature, humidity, air quality calculation, temperature 20-26℃, humidity 40%-60%, air quality index ≤100 as full marks; are weight coefficients , initial values are 0.3, 0.5, 0.2 respectively, and user can customize adjustment. 6.The intelligent pet management system based on artificial intelligence according to claim 5, wherein, The weight coefficient is dynamically adjusted according to the pet species and age, satisfying: ； wherein, Bit behavior score weight function: Puppies < 1 year, Adult dogs 1-8 years, Senior dogs > 8 years, ; Kittens < 1 year, Adult cats 1-7 years, Senior cats > 7 years, ; Bit physiological score weight function: puppies , adult dogs , senior dogs ; kittens , adult cats , senior cats ; For environment score weight function: all species juvenile stage , adult stage , old stage . 7.The intelligent pet management system based on artificial intelligence according to claim 6, wherein, The management control module generates an optimization objective function of the management scheme as: ； wherein: is an utility function, , is a management period, by default 24 hours, , ; The health benefit function is: , Each 1-point increase in health, increases utility by 2 units per hour; The feeding cost, per gram of food where The feeding cost, per gram of food The utility value, The device energy cost, per kWh The utility value. 8.The intelligent pet management system based on artificial intelligence of claim 7, wherein, The human-computer interaction module comprises: The voice interaction unit adopts the ASR speech recognition engine of Kedou Xunfei, receives user voice instructions, and the instruction response delay is ≤1s; The text interaction unit displays structured data through an APP interface: health score curve, diet suggestion list, and activity plan, supports user text input, and the input response delay is ≤500ms; The feedback unit feeds back the system execution results in two channels: voice feedback and APP push notification, with priority levels of normal, important, and urgent. 9.The intelligent pet management system based on artificial intelligence of claim 8, wherein, The intelligent pet management system based on artificial intelligence further comprises a cloud server, which is used for: Storing historical pet data: structured data is stored in a MySQL database, unstructured data is stored in a MongoDB database, nearly 3 years of data are retained, incremental backup is supported, and the data compression rate is ≥50%; Continuously optimizing the artificial intelligence model of the data analysis module based on a federal learning framework, the number of participating nodes is ≥100, the model aggregation period is 24 hours, and the optimization process follows a parameter update rule: ； where: is the model parameter for the th iteration; is the learning rate, initial value 0.001, exponential decay ; is the gradient of the loss function at . 10.The intelligent pet management system based on artificial intelligence according to claim 9, wherein, The management control module has a real-time alarm function, which triggers the human-computer interaction module to send an alarm to the user when the abnormal state judgment condition is met: ； wherein: is an indicator of an abnormal state, represents an abnormality; is a health score threshold: 60 for dogs, 55 for cats, 50 for small mammals, adjustable manually by ±10 points; Physiological parameter safety range: Heart rate: dog 60-140 times / minute, cat 120-140 times / minute, and rabbit 200-300 times / minute; Body temperature: dog 38.0-39.0℃, cat 38.5-39.5℃, and rabbit 38.5-40.0℃; Alert form: When, trigger "emergency" level push and phone notification, while the APP interface flashing warning light.

Images