Multi-mode intelligent sterilization and deodorization air purification method and system facing pet environment

Abstract

The invention discloses a multi-mode intelligent sterilization and deodorization air purification method and system for a pet environment, and belongs to the technical field of smart home, and the method comprises the following steps: collecting pet motion data and predicted pollution event probability and position; matching and outputting a pollutant category label and identifying the pollutant concentration; calculating a concentration estimation value of the pollutant in the indoor space grid in a future period of time; calculating the starting intensity and the energy consumption estimation value of the module in the corresponding area; and mapping the purification strategy matrix into an equipment control instruction through a power mapping function, and broadcasting the equipment control instruction to a corresponding equipment node. By constructing a closed-loop control structure from perception recognition to prediction to control feedback, the timeliness, the adaptability and the energy efficiency ratio of the pet environment air purification system are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the field of smart home technology, and in particular relates to a multi-mode intelligent sterilization and deodorization air purification method and system for pet environments. Background Art

[0002] With the accelerating pace of urban life and growing emotional needs, pet ownership has become a common practice in modern families. However, the presence of pets in domestic spaces can contribute to a range of air pollution issues, including ammonia released from feces, volatile organic compounds (such as mercaptans and amines) in pet body odor, organic particulate matter in saliva, and bacteria and fungi attached to shed hair and dander. These pollution sources are characterized by strong behavioral triggering, irregular onset times, wide variations in pollutant composition, and localized but rapidly spreading pollution. In contrast, traditional indoor air purification systems are often designed to target homogeneous pollutants such as formaldehyde, PM2.5, or kitchen fumes. These systems rely on simple concentration threshold triggering strategies, activating intensive purification only when pollutant concentrations exceed acceptable levels. This "post-event" response mechanism is often slow to respond to transient pollution caused by pet behavior, making it difficult to promptly contain the spread of peak pollutants, resulting in a significant decline in indoor air quality.

[0003] At the same time, there are significant differences in the odor characteristics between individual pets. Existing technologies often fail to identify and classify pollution types and lack personalized purification and adjustment strategies, resulting in either inefficient purification or excessive operation and energy waste. In addition, due to the sudden location of pet pollution, traditional equipment cannot evaluate the actual diffusion path of pollutants, and there are problems with purification blind spots and wind field mismatches. Some products have attempted to introduce pet recognition or odor recognition modules, but they are limited to static perception and lack predictive capabilities and linkage control mechanisms, making it difficult to achieve true intelligent scheduling. Therefore, the air purification systems currently on the market have not yet formed an effectively adapted technical system for pet home environments, and cannot meet the comprehensive requirements of "rapid identification of pollution sources + differentiation of pollution types + precise response and adjustment."

[0004] To this end, we propose a multi-mode intelligent sterilization and deodorization air purification method and system for pet environments to solve the above problems. Summary of the Invention

[0005] The purpose of the present invention is to solve the problem in the prior art that it is impossible to quickly identify pollution sources and distinguish pollution types and then accurately respond and adjust, and to propose a multi-mode intelligent sterilization and deodorization air purification method and system for pet environments.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] Multi-mode intelligent sterilization and deodorization air purification method for pet environments, including:

[0008] S1: Collect pet movement data and process the movement data to obtain behavior time series characteristics;

[0009] Inputting the behavior time series features into the Sigmoid activation function to obtain the predicted probability of the pollution event;

[0010] Performing weighted averaging on the position sequence within the sliding window of the behavior time series feature to obtain a predicted pollution event location;

[0011] S2: When the predicted probability of a pollution event is higher than a preset threshold, a detection point is selected based on the location of the pollution event and a local VOC sensor array is activated to obtain a sensor response value, the sensor response value is matched with a pollutant response template library, and a pollutant category label is output;

[0012] A piecewise cubic interpolation model is constructed for each type of pollutant. The sensor response value is input into the model based on the identified pollutant category label to obtain the pollutant concentration.

[0013] S3: constructing a discrete diffusion model, wherein the discrete diffusion model is a three-dimensional grid model;

[0014] The discrete diffusion model incorporates a pollutant type control coefficient and a furniture blocking coefficient; the pollutant type control coefficient is obtained by looking up a table based on the pollutant type;

[0015] Output the pollution concentration of the node at the corresponding time, and aggregate the results to obtain a concentration distribution tensor, which represents the concentration value of the pollutant in the future time window in the indoor space grid;

[0016] S4: Preset different types of purification modules and input the purification efficiency of all modules per unit power;

[0017] Generate a strategy matrix through the purification strategy function, and obtain the activation strength of the module in the corresponding area through the strategy matrix;

[0018] The purification strategy function includes a pollution hotspot residual term, an activation sparse term, and a spatial smoothing term;

[0019] The strategy matrix is ​​composed of the activation strength values ​​of all modules in the corresponding areas;

[0020] calculating an estimated energy consumption value based on the startup intensity value;

[0021] S5: Mapping the purification strategy matrix into device control instructions through a power mapping function, and broadcasting the device control instructions to corresponding device nodes.

[0022] Preferably, the motion data includes an acceleration sequence, a posture sequence and a position sequence; the acceleration sequence is used to describe the intensity and frequency of the pet's movement in different directions; the posture sequence represents the angle change around three axes, which is used to assist in judging the pet's posture; and the position sequence is used to track the pet's position trajectory.

[0023] Preferably, in the weighted average in step S1 to obtain the predicted pollution event location, the weight is calculated based on the acceleration amplitude.

[0024] Preferably, a pollution behavior a priori regularization term is introduced into the matching process in step S2, and the pollution behavior a priori regularization term is obtained through historical data based on a co-occurrence probability table of pollution types and behavior characteristics.

[0025] Preferably, the furniture blocking coefficient in step S3 is obtained by the structural permeability, and the structural permeability is obtained according to the following rules:

[0026] When there is no physical obstruction between the regions, the structural permeability is 1;

[0027] When there is furniture blocking the area, the structural permeability is an empirical coefficient less than 1 and greater than 0;

[0028] When the area is completely closed or is a wall, the structural permeability is 0.

[0029] Preferably, the pollution hotspot residual term is calculated by pollution concentration, grid position, purification efficiency of the module under unit power and startup intensity of the module in the corresponding area.

[0030] Preferably, a minimum executable threshold is set in the power mapping function, and the function is not activated when the threshold is lower than the minimum executable threshold.

[0031] A multi-mode intelligent sterilization and deodorization air purification system for pet environments, including:

[0032] A pet behavior prediction module is configured to collect pet motion data, process the motion data to obtain behavior time series features, input the behavior time series features into a Sigmoid activation function to obtain a predicted probability of a pollution event, and perform a weighted average of a position sequence within a sliding window of the behavior time series features to obtain a predicted pollution event location.

[0033] A pollutant component identification module is used to select a detection point based on the location of the pollution event and activate a local VOC sensor array to obtain a sensor response value when the predicted probability of the pollution event is higher than a preset threshold. The sensor response value is matched with a pollutant response template library to output a pollutant category label. A piecewise cubic interpolation model is constructed for each type of pollutant. The sensor response value is input into the model based on the identified pollutant category label to obtain the pollutant concentration.

[0034] A pollution diffusion modeling module is used to construct a discrete diffusion model, which is a three-dimensional grid model. The discrete diffusion model incorporates a pollutant type control coefficient and a furniture blocking coefficient. The pollutant type control coefficient is obtained by looking up the pollutant type in a table. The pollution concentration of the output node at the corresponding time is aggregated to obtain a concentration distribution tensor, which represents the concentration value of the pollutant in the indoor space grid within a future time window.

[0035] A purification strategy generation module is used to preset different types of purification modules and input the purification efficiency of all modules per unit power. A purification strategy function is used to generate a strategy matrix, which is used to obtain the activation intensity of the modules in the corresponding area. The purification strategy function includes a pollution hotspot residual term, an activation sparsity term, and a spatial smoothing term. The strategy matrix is ​​composed of the activation intensity values ​​of all modules in the corresponding area. Energy consumption estimates are calculated based on the activation intensity values.

[0036] The device execution module is used to map the purification strategy matrix into device control instructions through a power mapping function, and broadcast the device control instructions to corresponding device nodes.

[0037] To sum up, the technical effects and advantages of the present invention are as follows: the present invention constructs a pollution behavior monitoring and prediction mechanism, judges the potential pollution behavior of pets based on their activity characteristics in a specific environment, and realizes early response to air pollution events; combined with the pollutant characteristic identification mechanism, the system can determine the type of pollution source, estimate the pollution intensity, and provide personalized purification configuration support; the system establishes pollution diffusion trend modeling and spatial targeted purification strategy, which can dynamically evaluate the pollution diffusion range for complex structures within the home, and formulate parameter calling schemes for multi-mode purification modules, which significantly improves the timeliness, adaptability and energy efficiency of the pet environment air purification system, and overcomes the problems of delayed response, single strategy and high energy consumption in the existing technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a flow chart of the method of the present invention;

[0039] Figure 2Schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION

[0040] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0041] like Figure 1 As shown, the multi-mode intelligent sterilization and deodorization air purification method for pet environments includes:

[0042] S1: Collect pet movement data and process the movement data to obtain behavior time series characteristics;

[0043] Inputting the behavior time series features into the Sigmoid activation function to obtain the predicted probability of the pollution event;

[0044] Performing weighted averaging on the position sequence within the sliding window of the behavior time series feature to obtain a predicted pollution event location;

[0045] S2: When the predicted probability of a pollution event is higher than a preset threshold, a detection point is selected based on the location of the pollution event and a local VOC sensor array is activated to obtain a sensor response value, the sensor response value is matched with a pollutant response template library, and a pollutant category label is output;

[0046] A piecewise cubic interpolation model is constructed for each type of pollutant. The sensor response value is input into the model based on the identified pollutant category label to obtain the pollutant concentration.

[0047] S3: constructing a discrete diffusion model, wherein the discrete diffusion model is a three-dimensional grid model;

[0048] The discrete diffusion model incorporates a pollutant type control coefficient and a furniture blocking coefficient; the pollutant type control coefficient is obtained by looking up a table based on the pollutant type;

[0049] Output the pollution concentration of the node at the corresponding time, and aggregate the results to obtain a concentration distribution tensor, which represents the concentration value of the pollutant in the future time window in the indoor space grid;

[0050] S4: Preset different types of purification modules and input the purification efficiency of all modules per unit power;

[0051] Generate a strategy matrix through the purification strategy function, and obtain the activation strength of the module in the corresponding area through the strategy matrix;

[0052] The purification strategy function includes a pollution hotspot residual term, an activation sparse term, and a spatial smoothing term;

[0053] The strategy matrix is ​​composed of the activation strength values ​​of all modules in the corresponding areas;

[0054] calculating an estimated energy consumption value based on the startup intensity value;

[0055] S5: Mapping the purification strategy matrix into device control instructions through a power mapping function, and broadcasting the device control instructions to corresponding device nodes.

[0056] The specific steps are as follows:

[0057] Step 1:

[0058] The goal of this step is to predict the imminent occurrence of certain air pollution-causing behaviors, such as defecation, violent rolling, or fur licking. Given the short-term, sudden, and spatially localized nature of pet pollution behaviors, predicting these behaviors before they occur allows for the preemptive activation of subsequent pollution identification and purification response modules, enabling proactive and efficient air treatment. To achieve this, this step establishes a lightweight, time-series behavior prediction model. Based on acceleration and posture data collected by the collar and position trajectory data captured by the camera, it outputs the probability of occurrence and spatial location estimates of pollution events.

[0059] This step inputs actual sensor data from the device side:

[0060] The pet collar integrates a three-axis acceleration sensor and a posture fusion module with a sampling frequency of 50Hz. The acceleration sensor outputs an acceleration vector sequence a(t) = [a x (t),a y (t),a z (t)], which describes the intensity and frequency of the pet's movements in different directions. The posture fusion module outputs a posture angle sequence θ(t) = [φ(t), ψ(t), γ(t)], representing the angular changes around the three axes, to assist in determining behaviors such as rolling and lying down.

[0061] The pet's spatial trajectory, p(t) = [x(t), y(t)], is captured by an indoor positioning system. For example, a camera deployed in a home, combined with a lightweight SLAM algorithm (such as ORB-SLAM2), tracks the pet in real time on a 2D plane. Tracking accuracy can reach centimeters, with a frame rate of 10 fps.

[0062] All data are uniformly sampled and modeled within a 10-second sliding time window, and the window size is adjustable.

[0063] For example, in a real-world deployment, a pet might move around in a bedroom. The camera can capture its spatial trajectory, and the collar's acceleration can detect jumping and running. Sudden stillness after these intense movements could be a precursor to defecation. The system combines this behavioral sequence with its location trajectory and inputs it into the model to predict whether a contamination event is likely to occur within the next 10 seconds.

[0064] First, all input sequences are synchronously sampled to the same temporal resolution, forming three types of data on a unified time axis: acceleration sequence a(t), posture sequence θ(t), and position sequence p(t). The acceleration and posture signals are each fed into a two-layer bidirectional gated recurrent neural network (Bi-GRU), with 128 units per layer. The Bi-GRU is a temporal modeling structure that can simultaneously capture temporal dependencies between the preceding and following time periods and is suitable for modeling behavioral signals. At the end of the model, the output of the fused hidden state representation h is T , representing the dynamic characteristics of the current behavior sequence.

[0065] In obtaining the behavior time series feature h T Finally, in order to estimate the location of the potential pollution event, we perform a weighted average of the position sequence p(t) within the sliding window, and the weight is calculated based on the acceleration amplitude:

[0066] where w i =exp(α·||a(t i )||)

[0067] L event : The predicted pollution event location (two-dimensional vector), which is the weighted location center with higher motion intensity in this window;

[0068] p(t i ): spatial coordinates at the i-th moment, obtained by camera + SLAM estimation;

[0069] ||a(t i )||: acceleration amplitude at the i-th moment, collected by the collar;

[0070] w i : The weighting factor at each moment. The more intense the exercise, the greater the weight.

[0071] α: A hyperparameter that controls the sensitivity of acceleration weighting, with a typical value of 0.3.

[0072] In terms of behavior prediction, h T Input a fully connected neural network (with one hidden layer and Sigmoid output) and output the probability of a pollution event occurring in the future time window:

[0073] P event =σ(Wp ·h T +b p )

[0074] P event : The predicted probability of a pollution event, with a value between 0 and 1;

[0075] σ: Sigmoid activation function;

[0076] W p and b p : Fully connected weights and bias parameters obtained from model training;

[0077] h T : The temporal behavior representation vector generated by Bi-GRU.

[0078] Model training uses manually labeled start times of pollution behaviors as supervisory signals, which can be derived from synchronized video recordings combined with manual labels, or achieved through semi-automatic labeling using behavior recognition auxiliary tools.

[0079] The innovation of this step lies in locating the likely location of pollution events through a spatial weighting mechanism driven by acceleration amplitude, and employing a lightweight bidirectional RNN structure to model temporal behavior, resulting in excellent performance when deployed locally. Furthermore, the model inputs come from multiple heterogeneous sensors (collars and cameras), enabling the system to integrate behavioral intensity and spatial distribution characteristics, improving prediction accuracy and real-time performance. This is particularly applicable to scenarios with high pet density in small spaces.

[0080] This step outputs the following two variables, which are used for the next step of VOC sampling triggering and pollution diffusion simulation:

[0081] P event : The predicted probability of a pollution incident occurring in the short term in the future;

[0082] L event : The predicted location of the pollution event, used to guide the VOC sampling and diffusion model source item settings.

[0083] This step implements a locally deployable pet pollution behavior prediction module that integrates multimodal data. Using data collected by collar sensors and a visual positioning system, it extracts features of potential pollution behaviors within a lightweight time series model. This model, combined with an acceleration-guided weighting mechanism, predicts the location and probability of pollution events. This design addresses the sudden and complex spatial distribution of pet pollution behavior by establishing a pollution identification triggering mechanism based on behavioral priors, providing a predictive-driven logical foundation for precise responses in subsequent modules.

[0084] Step 2:

[0085] In a pet home environment, air pollution behaviors are often caused by individual behaviors. For example, the defecation, licking, and rubbing of the ground by a specific pet will release different types of odor molecules. The components of these pollutants are individual-specific, behavior-dependent, and spatially instantaneous, and often appear in the form of small doses, high concentrations, and local releases. Therefore, traditional VOC detection methods (fixed-point sampling, periodic detection) often have problems such as delayed response, inaccurate regions, and ambiguous judgments on pollution types. To this end, this step is based on the output of step 1 and combined with the probability of occurrence of the pollution event P event Its predicted position L event Dynamically selecting detection points in physical space and activating a local VOC sensor array enables rapid identification of pollution types and estimation of local pollution intensity. In the recognition model design, this step uses a local response matching approach guided by behavioral priors and introduces a pollution behavior regularization term to improve the recognition model's robustness to pet behavioral variability.

[0086] All inputs for this step come from the output of step 1, including:

[0087] P event : Output from the previous time series prediction model, indicating the probability of a pollution event occurring within a short time window in the future;

[0088] L event : Output by the acceleration weighting mechanism, indicating the location coordinates (in the two-dimensional plane) where the event occurred;

[0089] System preset threshold τ p (such as 0.7), when P event ≥τ p When the local VOC array gas sampling module is started, the gas sampling area is l event Within a 1-meter radius of the designated space, the system uses a directional blower or solenoid valve to direct the air in that area to a VOC sensor array (consisting of multiple metal oxide or MEMS sensors with different gas response characteristics). The sampling duration is 5 seconds and the sampling frequency is 10 Hz.

[0090] For example, the system predicts that defecation is about to occur in an area near a pet toilet and activates a VOC detection array in that area to collect characteristic gas signals to identify whether it is ammonia, thiols, or some individual odor mixture.

[0091] The raw responses of all sensors form a vector v = [v1, v2, ..., v m ], where v i is the average voltage or conductivity of the i-th sensor during this sampling period. The system stores a response template library of k typical pet pollutants. Each row represents a standard response vector for a pollutant type, obtained by factory calibration. Since different pets may have different odor characteristics for the same pollutant behavior, existing matching methods cannot adapt to individual differences. Therefore, this step introduces a behavioral regularization term Ω based on the traditional Euclidean residual. beh , expressed as a penalty term for the similarity measure between the predicted behavior intensity and the sensor response.

[0092] Pollutant identification is accomplished by minimizing the following objective function with a regularization term:

[0093]

[0094] C VOC : The final identified pollutant category label;

[0095] v i : The response value of the i-th sensor in this sampling;

[0096] T j,i : The response value of the j-th type of pollutant in the template library on the i-th sensor;

[0097] λ: regularization coefficient, controlling the influence of behavioral priors on decision making;

[0098] Ω beh (j,a): pollution behavior prior regularization term, whose value comes from the statistical co-occurrence relationship between pollution type j and behavior signal a(t);

[0099] The regularization term Ω beh This is a key innovative design in this step, used to reduce the risk of unstable pollutant identification due to individual pet behavior changes. The system has built a co-occurrence probability table of pollutant types and behavioral characteristics over long-term operation. This term is estimated using the behavioral feature vector a(t) (obtained from the acceleration sequence statistics in step 1) and the historical correlation P(j|a) of the pollutant. The specific form is as follows:

[0100] Ω beh (j,a)=―logP(j|a)

[0101] This feature makes it easier to match certain pollutant types with specific behavioral patterns. For example, defecation is often accompanied by high-amplitude, low-frequency acceleration, while licking is characterized by low-amplitude, high-frequency acceleration. The system will then favor pollutant types that better match these behaviors, thus avoiding misjudgments caused by cross-sensor interference.

[0102] In terms of pollutant concentration estimation, since the response value of each type of sensor does not correspond linearly to the concentration, this system constructs a piecewise cubic interpolation model f for each type of pollutant j (v), by identifying C VOC The current output v of its main response sensor* Enter this function to estimate the pollutant concentration:

[0103]

[0104] Function f j The calibration curve provided by the manufacturer is fitted, and each pollutant corresponds to a set of curve parameters. The output value is the relative concentration estimate of the pollutant at the current detection point.

[0105] This step outputs two variables used in subsequent pollution diffusion simulation:

[0106] C VOC : Identified pollutant category label (integer index);

[0107] Q VOC : The current local concentration estimate of the corresponding pollutant (normalized real number).

[0108] This step is based on the output of the pollution behavior prediction in the previous stage. It not only focuses on the sampling area in space, but also introduces a behavioral prior response matching mechanism to build a local recognition module that integrates dynamic response, pollutant identification and concentration estimation. In particular, by introducing the pollution behavior regularization term in the pollutant matching objective function, the system can output recognition results more stably in the scenario of sensor cross-interference, thereby improving the individual adaptability and robustness of recognition. This design fully reflects the targeted innovation in the patent scenario: the system not only identifies pollution, but also determines "who", "when" and "where" the pollution is caused, providing highly reliable pollution source data for subsequent spatial diffusion modeling and multi-mode purification response, with practical engineering deployment value and significant patent creativity.

[0109] Step 3:

[0110] The goal of this step is to model the spatial diffusion of pollutants from the pollution sources identified in the previous step. This involves predicting the concentration distribution of pollutants over time in the three-dimensional indoor space, thereby providing a basis for developing a purification strategy. Pet home environments exhibit significant non-uniform diffusion characteristics, such as pollution sources located close to the ground, complex furniture obstructions, and uneven local ventilation. Therefore, a rapid modeling approach with spatial structural adaptability is required. This step utilizes a simplified mesh modeling strategy combined with lightweight, approximate, decoupled diffusion modeling to simulate pollution diffusion trends, while maintaining a balance between prediction accuracy and response time.

[0111] The input received by this step all comes from the output of step 2, including:

[0112] C VOC : The pollutant type number identified in the previous step, used to match the pollutant diffusion rate parameters;

[0113] Q VOC : The estimated concentration of the pollutant at the current location, used as the initial value of the source term;

[0114] L event : The location of the pollution incident from step 1 is the starting coordinate of the pollution spread;

[0115] These three variables together determine the boundary conditions and initial value settings of the pollution diffusion model.

[0116] This step first constructs a regular 3D grid model in the indoor space to simulate the diffusion of pollutants in the home space. The space is divided into uniform voxel units, and the center coordinate of each unit represents a simulation node. For example, a room of 6m×4m×2.5m is divided into cubic units with a side length of 0.5m, totaling 1920 nodes. Pollution source point L event The cell where it is located is called node V s , whose initial concentration is Q VOC , the initial concentration of all other nodes is 0.

[0117] Considering that high-precision CFD simulations cannot be directly deployed in real-time systems, this step constructs a fast simulation method based on the discrete form of the diffusion equation. Based on the traditional discrete diffusion model, a simplified update formula that integrates the pollutant type control coefficient and the furniture barrier coefficient is designed as follows:

[0118]

[0119] The pollution concentration of the i-th node at time t;

[0120] The set of node indexes adjacent to node i;

[0121] κ j→i : Diffusion coefficient from node j to node i, defined as κ·η j→i ;

[0122] κ: by pollutant type C VOC Basic diffusion coefficient obtained by looking up the table;

[0123] η j→i : The structural permeability (blocking factor) from node j to i, generated by the indoor structure scanning module;

[0124] t: discrete time step. The simulation usually completes the prediction after iterating to t=10 (approximately equal to the concentration distribution after 10 seconds).

[0125] Structural permeability η j→iThe occlusion matrix is ​​generated by scanning the space through the indoor layout mapping system (for example, using Realsense depth camera or LiDAR). If there is no physical obstruction between j and i, η j→i =1; if there is some furniture blocking the way, η j→i is an empirical coefficient less than 1 (e.g. 0.4); if it is completely enclosed or a wall, η j→i = 0. Through this design, even without building a complete physical model, the inhibitory effect of structural heterogeneity on pollutant propagation can be reflected.

[0126] The basic coefficient k for controlling the diffusion rate of pollutant type is a lookup table parameter, for example:

[0127] Ammonia

[0128] Thiols

[0129] Aldehydes

[0130] This part comes from experimental calibration or existing gas database.

[0131] The entire simulation process runs on a local microcontroller or edge server, and the future concentration distribution can be output under 3 layers of convolution (10 iterations). The total simulation time is controlled within 1 second, meeting quasi-real-time requirements.

[0132] For example, if a pet is detected releasing high concentrations of ammonia on a bedroom carpet, the system will simulate the pollutant's distribution in the surrounding area 10 seconds after the pet's arrival. The system will find that the pollution is primarily concentrated near the pet's bed and upstream of the air conditioner's airflow, while the area behind the cabinets is severely obstructed, making it difficult for pollutants to diffuse in. These results will provide a basis for subsequent targeted purification strategies.

[0133] This step outputs a three-dimensional concentration distribution tensor D(x, y, z), which represents the estimated concentration of pollutants in the indoor space grid over a period of time in the future, and can be used by the subsequent purification strategy optimization module.

[0134] Step 4:

[0135] The goal of this step is to calculate the spatial distribution of pollutants D(x,y,z), the type of pollutants C VOC and its source intensity Q VOC , generate a targeted, energy-saving, spatially differentiated multi-module purification strategy matrix A modeIn pet home environment, pollution behavior has the characteristics of suddenness, locality and diversification, so it is necessary to avoid the waste of resources and response lag caused by traditional air purification system "whole house high intensity purification". Therefore, this step designs a pollution hot spot guidance, type adaptive control, module energy consumption and space smoothness regulation purification strategy generation mechanism, so that the air purification behavior has intelligent differential response ability.

[0136] The inputs of this step are all from the previous module (pollution diffusion modeling) and the cooperation of the previous two steps, including:

[0137] D(x, y, z): pollution concentration distribution tensor, the future concentration estimate value of pollutants at each point in three-dimensional space;

[0138] C VOC : pollutant type number;

[0139] Q VOC : pollution source concentration estimate value;

[0140] L event : pollution source point position, used to ensure that the pollution response does not miss the source point.

[0141] These inputs jointly determine the response range, regulation type and priority area of the purification strategy.

[0142] This system has n different types of purification modules built-in (such as activated carbon filter, UV lamp sterilization, negative ion diffusion, etc.), each module is arranged in m air outlet areas, and the corresponding space area is pollution hotspot area composed of grid points with more than 95% quantile value in tensor D(x, y, z), and L event The area where it is located is forced to contain.

[0143] The purification ability of each module k to pollutants C VOC is given by the lookup table matrix , which represents the purification effect under unit power. Let the strategy matrix represent the starting intensity of module k in area j (the value range is [0, 1], for example, the air speed ratio).

[0144] We construct the following purification strategy generation objective function:

[0145]

[0146] D i : the pollution concentration of the i-th grid;

[0147] δ ij : whether grid i belongs to area j;

[0148] The activation strength of module k in region j;

[0149] The purification efficiency of module k for the pollutant;

[0150] λ1: Sparse regularization coefficient that controls the overall activation level;

[0151] λ2: Smoothing coefficient, used to control the activation fluctuation of the same module in adjacent areas.

[0152] The objective function consists of three components: a pollution hotspot residual term, an activation sparsity term, and a spatial smoothing term. These three components are jointly regulated to ensure precise and differentiated control of the system's pollution response and minimize unnecessary module activations.

[0153] After the optimization is completed, the following two items are output:

[0154] Strategy Matrix A mode That is, all the value;

[0155] Estimated energy consumption E cost Calculated by the following formula:

[0156]

[0157] Energy consumption constant of module k under unit activation intensity in region j (normalized value per unit power);

[0158] The actual startup strength of the optimized module in the corresponding area;

[0159] E cost : Estimated total energy consumption under the entire strategy.

[0160] For example: If strong ammonia pollution is detected near the toilet and D(x,y,z) shows that the pollution is spreading to the living room, the system will activate the activated carbon module and local UV auxiliary module near the boundary between the toilet and the living room, and output the corresponding intensity value (such as Indicates that the first type of module works at 70% power in the third area), and outputs the energy consumption required for the entire strategy as E cost =1.45 (unit is normalized value).

[0161] Output:

[0162] A mode : purification strategy matrix, size is n×, each element represents the power output coefficient of module k in region j;

[0163] E cost: The estimated total energy consumption required for strategy execution provides a basis for system optimization and feedback mechanism;

[0164] Based on pollution hotspot identification, this step constructs a multi-module purification strategy optimization mechanism that adapts to pollution types, adjusts to specific regions, and controls energy consumption. By explicitly introducing module-pollution type response matching, L1 sparse activation, and spatial difference smoothing, the generated strategy not only covers key areas of pollutant diffusion but also intelligently distributes module loads to minimize noise and energy consumption.

[0165] Step 5:

[0166] The goal of this step is to transform the purification strategy matrix A generated in the previous step mode These instructions are mapped into executable device control commands, driving home air purification devices to perform precise and differentiated purification actions, achieving a complete closed-loop implementation from pollution prediction to pollution control. Unlike traditional purifiers that operate at high intensity across the entire system, this step activates specific air outlet units in specific modules based on regional pollution conditions and module compatibility, achieving low-energy, high-efficiency localized pollution response.

[0167] The core of this step is to mode The strategy value in the system is converted into a control instruction that can be executed by the hardware. Each purification module k in the system has a standardized control interface and its working state can be determined by the power mapping function P k (·) control, corresponding to hardware parameters such as fan speed, UV lamp brightness, negative ion release rate, etc. The settings are as follows:

[0168]

[0169] U dev : The final generated control instruction set;

[0170] The power mapping function of purification module k is provided by the equipment manufacturer and is generally implemented as a piecewise linear function or a lookup table, for example Indicates three wind speed levels;

[0171] ∈: minimum executable threshold of the device (e.g., 0.05), used to avoid false activation of the device with low intensity;

[0172] Indicates the specific control parameters of the kth module in the jth area.

[0173] This command set will be broadcast to the device nodes in each area of ​​the home through a local wireless protocol (such as BLE Mesh, ZigBee, etc.). Each node has a built-in micro MCU to receive the corresponding commands for the area and drive the module to execute according to the control parameters. For example, if the living room area is identified as a high-value area of ​​ammonia pollution diffusion, the system will send a command to the activated carbon module in this area. (i.e. wind speed 2) control command.

[0174] The technical solutions in the above-mentioned embodiments of the present application have at least the following technical effects or advantages: the present invention constructs a pollution behavior monitoring and prediction mechanism, judges the potential pollution behavior of pets based on their activity characteristics in a specific environment, and realizes early response to air pollution events; combined with the pollutant characteristic identification mechanism, the system can determine the type of pollution source, estimate the pollution intensity, and provide personalized purification configuration support; the system establishes pollution diffusion trend modeling and spatial targeted purification strategies, which can dynamically evaluate the pollution diffusion range for complex structures within the home, and formulate parameter calling schemes for multi-mode purification modules, significantly improving the timeliness, adaptability and energy efficiency of the pet environment air purification system, and overcoming the problems of delayed response, single strategy and high energy consumption in the existing technology.

[0175] The present application also provides a multi-mode intelligent sterilization and deodorization air purification system for pet environments, such as Figure 2 Shown, including:

[0176] A pet behavior prediction module is configured to collect pet motion data, process the motion data to obtain behavior time series features, input the behavior time series features into a Sigmoid activation function to obtain a predicted probability of a pollution event, and perform a weighted average of a position sequence within a sliding window of the behavior time series features to obtain a predicted pollution event location.

[0177] A pollutant component identification module is used to select a detection point based on the location of the pollution event and activate a local VOC sensor array to obtain a sensor response value when the predicted probability of the pollution event is higher than a preset threshold. The sensor response value is matched with a pollutant response template library to output a pollutant category label. A piecewise cubic interpolation model is constructed for each type of pollutant. The sensor response value is input into the model based on the identified pollutant category label to obtain the pollutant concentration.

[0178] A pollution diffusion modeling module is used to construct a discrete diffusion model, which is a three-dimensional grid model. The discrete diffusion model incorporates a pollutant type control coefficient and a furniture blocking coefficient. The pollutant type control coefficient is obtained by looking up the pollutant type in a table. The pollution concentration of the output node at the corresponding time is aggregated to obtain a concentration distribution tensor, which represents the concentration value of the pollutant in the indoor space grid within a future time window.

[0179] A purification strategy generation module is used to preset different types of purification modules and input the purification efficiency of all modules per unit power. A purification strategy function is used to generate a strategy matrix, which is used to obtain the activation intensity of the modules in the corresponding area. The purification strategy function includes a pollution hotspot residual term, an activation sparsity term, and a spatial smoothing term. The strategy matrix is ​​composed of the activation intensity values ​​of all modules in the corresponding area. Energy consumption estimates are calculated based on the activation intensity values.

[0180] The device execution module is used to map the purification strategy matrix into device control instructions through a power mapping function, and broadcast the device control instructions to corresponding device nodes.

[0181] The working principle is as follows: The present invention constructs a pollution behavior monitoring and prediction mechanism, judges the potential pollution behavior of pets based on their activity characteristics in a specific environment, and realizes early response to air pollution events; combined with the pollutant characteristic identification mechanism, the system can determine the type of pollution source, estimate the pollution intensity, and provide personalized purification configuration support; the system establishes pollution diffusion trend modeling and spatial targeted purification strategy, which can dynamically evaluate the pollution diffusion range for complex structures within the home and formulate parameter calling plans for multi-mode purification modules.

[0182] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A multi-mode intelligent sterilization and deodorization air purification method for pet environments, characterized by: include: S1: Collect pet movement data and process the movement data to obtain behavior time series characteristics; Inputting the behavior time series features into the Sigmoid activation function to obtain the predicted probability of the pollution event; Performing weighted averaging on the position sequence within the sliding window of the behavior time series feature to obtain a predicted pollution event location; S2: When the predicted probability of a pollution event is higher than a preset threshold, a detection point is selected based on the location of the pollution event and a local VOC sensor array is activated to obtain a sensor response value, the sensor response value is matched with a pollutant response template library, and a pollutant category label is output; A piecewise cubic interpolation model is constructed for each type of pollutant. The sensor response value is input into the model based on the identified pollutant category label to obtain the pollutant concentration. S3: constructing a discrete diffusion model, wherein the discrete diffusion model is a three-dimensional grid model; The discrete diffusion model incorporates a pollutant type control coefficient and a furniture blocking coefficient; the pollutant type control coefficient is obtained by looking up a table based on the pollutant type; Output the pollution concentration of the node at the corresponding time, and aggregate the results to obtain a concentration distribution tensor, which represents the concentration value of the pollutant in the future time window in the indoor space grid; S4: Preset different types of purification modules and input the purification efficiency of all modules per unit power; Generate a strategy matrix through the purification strategy function, and obtain the activation strength of the module in the corresponding area through the strategy matrix; The purification strategy function includes a pollution hotspot residual term, an activation sparse term, and a spatial smoothing term; The strategy matrix is ​​composed of the activation strength values ​​of all modules in the corresponding areas; calculating an estimated energy consumption value based on the startup intensity value; S5: Mapping the purification strategy matrix into device control instructions through a power mapping function, and broadcasting the device control instructions to corresponding device nodes.

2. The multi-mode intelligent sterilization and deodorization air purification method for pet environments according to claim 1 is characterized in that: The motion data includes acceleration sequence, posture sequence and position sequence; the acceleration sequence is used to describe the pet's movement intensity and frequency in different directions; the posture sequence represents the angle change around three axes, which is used to assist in judging the pet's posture; the position sequence is used to track the pet's position trajectory.

3. The multi-mode intelligent sterilization and deodorization air purification method for pet environments according to claim 1 is characterized in that: In the weighted average method described in step S1 to obtain the predicted pollution event location, the weight is calculated based on the acceleration amplitude.

4. The multi-mode intelligent sterilization and deodorization air purification method for pet environments according to claim 1 is characterized in that: A pollution behavior priori regularization term is introduced into the matching process described in step S2. The pollution behavior priori regularization term is obtained through historical data based on a co-occurrence probability table of pollution types and behavior characteristics.

5. The multi-mode intelligent sterilization and deodorization air purification method for pet environments according to claim 1 is characterized in that: The furniture blocking coefficient in step S3 is obtained by the structural permeability, which is obtained according to the following rules: When there is no physical obstruction between the regions, the structural permeability is 1; When there is furniture blocking the area, the structural permeability is an empirical coefficient less than 1 and greater than 0; When the area is completely closed or is a wall, the structural permeability is 0.

6. The multi-mode intelligent sterilization and deodorization air purification method for pet environments according to claim 1 is characterized in that: The pollution hotspot residual term is calculated based on the pollution concentration, grid position, purification efficiency of the module per unit power, and the startup intensity of the module in the corresponding area.

7. The multi-mode intelligent sterilization and deodorization air purification method for pet environments according to claim 1 is characterized in that: A minimum executable threshold is set in the power mapping function, and the function is not activated when the threshold is lower than the minimum executable threshold.

8. A multi-mode intelligent sterilization and deodorization air purification system for pet environments, characterized by: include: A pet behavior prediction module is used to collect pet motion data, process the motion data to obtain behavior time series features, and input the behavior time series features into a Sigmoid activation function to obtain a predicted probability of a contamination event; Performing weighted averaging on the position sequence within the sliding window of the behavior time series feature to obtain a predicted pollution event location; A pollutant component identification module is used to select a detection point based on the location of the pollution event and activate a local VOC sensor array to obtain a sensor response value when the predicted probability of the pollution event is higher than a preset threshold. The sensor response value is matched with a pollutant response template library to output a pollutant category label. A piecewise cubic interpolation model is constructed for each type of pollutant. The sensor response value is input into the model based on the identified pollutant category label to obtain the pollutant concentration. A pollution diffusion modeling module is used to construct a discrete diffusion model, which is a three-dimensional grid model. The discrete diffusion model incorporates a pollutant type control coefficient and a furniture blocking coefficient. The pollutant type control coefficient is obtained by looking up the pollutant type in a table. The pollution concentration of the output node at the corresponding time is aggregated to obtain a concentration distribution tensor, which represents the concentration value of the pollutant in the indoor space grid within a future time window. A purification strategy generation module is used to preset different types of purification modules and input the purification efficiency of all modules per unit power. A purification strategy function is used to generate a strategy matrix, which is used to obtain the activation intensity of the modules in the corresponding area. The purification strategy function includes a pollution hotspot residual term, an activation sparsity term, and a spatial smoothing term. The strategy matrix is ​​composed of the activation intensity values ​​of all modules in the corresponding area. Energy consumption estimates are calculated based on the activation intensity values. The device execution module is used to map the purification strategy matrix into device control instructions through a power mapping function, and broadcast the device control instructions to corresponding device nodes.

Images