Intelligent dehumidifying and mildew-removing kitchen garbage can

Abstract

According to the intelligent dehumidification and mildew removal kitchen garbage can, environment data are collected through multiple sensors, environment state evaluation is conducted through a lightweight edge calculation model, optimal control parameters are calculated through a convex data driving shrinkage algorithm based on Riemannian measurement, and intelligent dehumidification and mildew removal control is achieved. According to the scheme, multi-device cooperative work is realized by adopting the smart contract based on the block chain, the technical problems of single control strategy, low energy utilization efficiency and lack of multi-device cooperative capability in the prior art are solved, and the technical effects of high control precision and high energy utilization efficiency are achieved.

Description

Technical Field

[0001] The present application relates to the field of environmental sanitation technology, and in particular to an intelligent dehumidifying and mildew-removing kitchen waste bin. Background Art

[0002] The treatment of kitchen waste has always been an important issue in urban environmental sanitation management. Reasonable collection and treatment of kitchen waste is not only related to the quality of life of residents, but also an important part of urban environmental sanitation.

[0003] At present, the kitchen waste bins on the market mainly include traditional mechanical bins, simple electric bins and the first generation of smart bins. Among them, the traditional mechanical bins only have basic garbage collection functions; although the simple electric bins are equipped with basic deodorization and dehumidification functions, they often waste energy due to the lack of intelligent control systems; the first generation of smart bins introduced some intelligent control functions, but adopted a single control strategy.

[0004] Existing smart trash cans usually use simple threshold control or fixed mode control, and cannot dynamically adjust working parameters according to actual environmental conditions. This type of control method has problems such as unsatisfactory dehumidification and mildew removal effects and low energy efficiency. At the same time, the existing technology lacks an effective multi-device coordination mechanism, and each trash can operates independently, which cannot achieve optimal resource allocation and coordinated task processing.

[0005] The above technology has the following problems: first, it lacks intelligent control strategies based on environmental perception, resulting in poor dehumidification and mildew removal effects and low energy efficiency; second, it cannot achieve the coordinated work of multiple devices, resulting in low overall system efficiency. These problems seriously restrict the actual application effect of smart trash cans. Summary of the invention

[0006] In view of this, the present application provides an intelligent dehumidifying and mildew-removing kitchen waste bin, which solves the problems in the prior art of lack of intelligent control strategies based on environmental perception and inability to achieve collaborative work of multiple devices.

[0007] The present application embodiment provides a smart dehumidifying and mildew-removing kitchen waste bin, comprising: A data acquisition module is used to collect environmental data through multiple sensors arranged in the trash can, perform outlier detection and sliding window average processing on the environmental data, and obtain standardized environmental data; wherein the environmental data includes temperature data, humidity data, harmful gas concentration data and light intensity data; A state assessment module, used to analyze the standardized environmental data using a lightweight edge computing model to generate an environmental state assessment result; wherein the environmental state assessment result includes a temperature anomaly score, a humidity anomaly score, a harmful gas concentration anomaly score, and a mold risk level; A parameter calculation module, used for mapping the environmental state assessment result to a Riemann manifold space, constructing a kernel regression model based on the mapped Riemann manifold space, learning the state control mapping relationship between the environmental state and the control parameters through the kernel regression model, and performing convex optimization calculation on the kernel regression model to obtain optimal control parameters; wherein the optimal control parameters include a PTC heating power parameter, a ventilation system speed parameter, and an ultraviolet irradiation intensity parameter; A control execution module, configured to generate a control instruction according to the optimal control parameter, send the control instruction to the corresponding actuator through a hardware driver interface set on the trash can, and execute the control instruction; and estimate the real-time state in the trash can based on Kalman filtering to obtain a real-time state value, compare the real-time state value with a preset state threshold, and when the real-time state value is less than the preset state threshold, adjust the control instruction according to the deviation degree of the real-time state value; The collaborative work module is used to build a blockchain-based smart contract, which conducts resource bidding and task allocation based on the geographical location, processing capacity and energy efficiency between garbage bins, thereby realizing collaborative work among multiple garbage bins.

[0008] Preferably, the data acquisition module includes: A temperature and humidity collection unit, used to set a temperature and humidity sensor to collect temperature and humidity data in the barrel in real time; The gas concentration collection unit is used to set the gas sensor to detect the concentration data of ammonia and hydrogen sulfide in the barrel; A light collection unit is used to set a photosensitive sensor to monitor the light intensity data in the barrel; The data preprocessing unit is used to perform outlier detection on the environmental data and perform smoothing using a sliding window averaging method to obtain the standardized environmental data.

[0009] Preferably, the status assessment module comprises: A model building unit is used to build a three-layer lightweight neural network. During the construction, it is ensured that the number of nodes in the input layer matches the number of sensors provided in the trash can, the hidden layer uses 32 neurons, and the output layer corresponds to different dimensional indicators of the environmental state; A knowledge migration unit, configured to adopt a soft label conversion strategy with specific parameters to migrate knowledge in an existing pre-trained environment evaluation model to the lightweight edge computing model; The model optimization unit is used to reduce the accuracy of model parameters from high floating points to low fixed points using quantization-aware training methods, and to compress the model size using model pruning technology that sets weight thresholds.

[0010] Preferably, the status assessment module further includes: A feature extraction unit, used to divide the standardized environmental data into time windows according to specific time lengths, and calculate the average value, standard deviation and peak value of the data in each time window; A dimension reduction processing unit is used to perform principal component analysis on the mean value, standard deviation and peak value, retain the principal component with an explained variance rate of 95%, and obtain a feature vector after dimension reduction; An evaluation and calculation unit is used to input the dimensionally reduced feature vector into the lightweight edge computing model to calculate the temperature anomaly score, the humidity anomaly score, the harmful gas concentration anomaly score and the mildew risk level.

[0011] Preferably, the parameter calculation module includes: A manifold construction unit, used for mapping the environmental state assessment result to a symmetric positive definite matrix manifold space, and designing a corresponding Riemann metric tensor based on the physical properties of the environmental state assessment result; The mapping conversion unit is used to establish a distance-preserving mapping from the Euclidean space to the symmetric positive definite matrix manifold space to ensure the continuity when performing the state conversion from the Euclidean space to the symmetric positive definite matrix manifold space.

[0012] Preferably, the parameter calculation module further includes: A data acquisition unit, used to collect state transition data and control input instructions of the garbage bin under different working conditions; A model learning unit, used for constructing a kernel regression model in the symmetric positive definite matrix manifold space, and defining a geodesic distance by using an exponential mapping and a logarithmic mapping; The parameter optimization unit is used to construct an objective function based on the kernel regression model and solve the optimal control parameters using a gradient projection algorithm with an adaptive step size.

[0013] Preferably, the control execution module includes: A strategy generating unit, configured to generate a control instruction according to the optimal control parameter, wherein the control instruction includes a start timing of the PTC heating power, an adjustment timing of the ventilation system speed, and a duration of ultraviolet irradiation; An instruction execution unit, used to send control instructions to corresponding actuators through a hardware driver interface, wherein the actuators include a PTC heater, a ventilation system, and an ultraviolet lamp; The feedback adjustment unit estimates the real-time state in the trash can based on the Kalman filter to obtain a real-time state value, compares the real-time state value with a preset state threshold, and when the real-time state value is less than the preset state threshold, adjusts the optimal control parameter online through the PI controller.

[0014] Preferably, the control execution module further includes: A monitoring unit, used to obtain the working status of the actuator and changes in environmental parameters with a sampling period of 100 milliseconds; An evaluation unit, for calculating the dehumidification rate, energy efficiency and mold inhibition rate based on the working state and the change of the environmental parameters, and performing result analysis in a sliding manner with a 24-hour window and a 1-hour step length to obtain an execution result; The optimization unit is used to build a multi-objective evaluation model based on the execution results, and optimize the initial value selection and dynamic adjustment rules of the control parameters by using a reinforcement learning method.

[0015] Preferably, the collaborative working module includes: The bidding unit is used to build a blockchain-based smart contract to bid for resources based on the geographical location, processing capacity, and energy efficiency of the garbage bins; The scheduling unit is used to synchronize data between trash cans using a publish-subscribe model, and to implement distributed task scheduling and fault tolerance; The optimization unit is used to collect the operating data of the collaborative work between multiple trash cans and use deep reinforcement learning methods to optimize the resource allocation strategy and task scheduling algorithm.

[0016] The present application also provides a computer device, the computer device comprising: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the control method of the above-mentioned intelligent dehumidification and mildew-removing kitchen waste bin.

[0017] An embodiment of the present application also provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute the control method of the above-mentioned intelligent dehumidification and mildew-removing kitchen waste bin.

[0018] An embodiment of the present application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the control method of the above-mentioned intelligent dehumidification and mildew-removing kitchen waste bin.

[0019] This application has the following technical effects: By collecting environmental data through multiple sensors and evaluating the environmental status through lightweight edge computing models, accurate perception and real-time monitoring of the environment inside the trash can are achieved; the convex data-driven contraction algorithm based on Riemann metric is used to calculate the optimal control parameters, which improves the accuracy and energy efficiency of dehumidification and mildew removal control; and the introduction of a blockchain-based smart contract mechanism enables multi-device collaboration, thereby improving the overall operating efficiency of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the following is a brief introduction to the drawings required for use in the embodiment: Figure 1 A schematic diagram of the overall structure of the intelligent dehumidifying and mildew-removing kitchen waste bin provided in an embodiment of the present application; Figure 2 A schematic diagram of the structure of the data acquisition module provided in the embodiment of the present application; Figure 3 A schematic diagram of the structure of a lightweight edge computing model of a state assessment module provided in an embodiment of the present application; Figure 4 A schematic diagram of the Riemannian manifold mapping and optimization process of the parameter calculation module provided in an embodiment of the present application; Figure 5 A schematic diagram of the workflow of the control execution module provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] like Figure 1 As shown, the embodiment of the present application provides an intelligent dehumidifying and mildew-removing kitchen waste bin, comprising: The data acquisition module 10 is used to collect environmental data through multiple sensors arranged in the trash can, perform outlier detection and sliding window average processing on the environmental data, and obtain standardized environmental data; wherein the environmental data includes temperature data, humidity data, harmful gas concentration data and light intensity data; like Figure 2 As shown, the data acquisition module 10 includes: The temperature and humidity collection unit 11 is used to set a temperature and humidity sensor to collect temperature and humidity data in the barrel in real time; The gas concentration collection unit 12 is used to set a gas sensor to detect the concentration data of ammonia and hydrogen sulfide in the barrel; A light collection unit 13 is used to set a photosensitive sensor to monitor the light intensity data in the barrel; The data preprocessing unit 14 is used to perform abnormal value detection on the environmental data and perform smoothing processing using a sliding window averaging method to obtain the standardized environmental data.

[0022] The specific implementation process of the data acquisition module 10 is described in detail below in conjunction with the accompanying drawings. The data acquisition module 10 mainly completes the collection and preprocessing tasks of environmental data, providing a reliable data basis for subsequent environmental status assessment.

[0023] Specifically, multiple sensors are evenly arranged around the inner wall of the trash can, including temperature and humidity sensors, gas sensors and photosensors. Among them, the temperature and humidity sensor uses the DHT22 digital temperature and humidity sensor, with a measurement accuracy of ±0.5℃ and ±2%RH, and a sampling frequency of 0.5Hz; the gas sensor uses the MQ-137 ammonia sensor and the MQ-136 hydrogen sulfide sensor, with sensitivities of 10ppm and 1ppm respectively, and a response time of less than 10 seconds; the photosensor uses the BH1750FVI digital light intensity sensor, with a measurement range of 1-65535lx and a resolution of 1lx. By arranging these sensors reasonably, it is ensured that the environmental status inside the trash can can be fully perceived.

[0024] Secondly, for the raw data collected by the sensor, this application designs a complete outlier detection mechanism. In the specific implementation, an improved Z-score algorithm is used for outlier detection, and outliers are identified by calculating the standard deviation multiple between the data point and the mean. The calculation formula is:

[0025] in is the data point to be detected, is the mean value in the time window, is the standard deviation. When the value is greater than 3, the data point is determined to be an outlier. For the detected outliers, linear interpolation method is used to correct them to ensure the continuity and reliability of the data.

[0026] Then, in order to suppress data noise and random fluctuations, this application uses an adaptive sliding window averaging method to smooth the data. The window size is dynamically adjusted according to the data change rate. The specific algorithm is: when the data change rate is detected to exceed the preset threshold (such as the temperature change rate>0.5℃ / s), the window size is reduced to increase the response speed; when the data is relatively stable, the window size is increased to enhance the smoothing effect. The window size can be adjusted in the range of 5-30 sampling points, with a step size of 5 sampling points.

[0027] Finally, the smoothed data is standardized to make different types of environmental data comparable. The standardization uses the Min-Max method to map various types of data to the [0,1] interval. The calculation formula is:

[0028] Where x is the original data, and are the historical minimum and maximum values ​​of this type of data respectively. At the same time, the exponential sliding average method is introduced to dynamically update and The value of enables the standardization process to adapt to the long-term changing trend of environmental data.

[0029] In summary, the data acquisition module 10 achieves accurate acquisition and effective preprocessing of environmental data in the trash can through multi-sensor collaborative acquisition and multi-level data processing, providing high-quality data support for subsequent environmental status assessment.

[0030] A state assessment module 20, for analyzing the standardized environmental data using a lightweight edge computing model to generate an environmental state assessment result; wherein the environmental state assessment result includes a temperature anomaly score, a humidity anomaly score, a harmful gas concentration anomaly score, and a mold risk level; The state evaluation module 20 is one of the core modules of the present application, which realizes intelligent evaluation of the environmental state through a lightweight neural network. The specific implementation process of the state evaluation module 20 is described in detail below.

[0031] like Figure 3 As shown, the state assessment module 20 includes: The model building unit 21 is used to build a three-layer lightweight neural network. During the construction, it is ensured that the number of nodes in the input layer is compatible with the number of sensors provided in the trash can, the hidden layer uses 32 neurons, and the output layer corresponds to different dimensional indicators of the environmental state; The model building unit 21 adopts a three-layer structure to design a lightweight neural network. The number of nodes in the input layer is determined according to the number of sensors installed in the trash can, specifically, including 4 temperature sensor nodes, 4 humidity sensor nodes, 2 ammonia sensor nodes, 2 hydrogen sulfide sensor nodes and 4 light sensor nodes, totaling 16 input nodes. The hidden layer adopts a compact structure of 32 neurons and uses ReLU (Rectified Linear Unit) as the activation function, and its mathematical expression is:

[0032] The output layer is designed with 4 nodes, corresponding to the temperature anomaly score, humidity anomaly score, harmful gas concentration anomaly score and mold risk level. The first three scores are normalized using the Sigmoid activation function to ensure that the output value is within the range of [0,1]; the mold risk level uses the Softmax function to output the probability distribution of three levels (low, medium and high).

[0033] A knowledge migration unit 22, configured to adopt a soft label conversion strategy with specific parameters to migrate knowledge in an existing pre-trained environment evaluation model to the lightweight edge computing model; The knowledge transfer unit 22 realizes the knowledge transfer from the pre-trained large model to the lightweight model. In the specific implementation, a large environmental assessment model is first pre-trained based on the ResNet-50 architecture, and an environmental data set containing 1 million annotated data is used for training. Then, a soft label conversion strategy with a temperature coefficient T=3 is used for knowledge distillation, where the calculation formula of the soft label is:

[0034] in is the logits value output by the large model. By adjusting the temperature coefficient T, a smoother probability distribution can be generated, which is conducive to the effective transfer of knowledge. During the training process, a mixed loss function is used:

[0035] Where LCE is the cross entropy loss, LKD is the knowledge distillation loss, and α is set to 0.7 to balance the contribution of the two losses.

[0036] The model optimization unit 23 is used to reduce the accuracy of model parameters from high floating points to low fixed points by using quantization-aware training methods, and compress the model size by using model pruning technology with set weight thresholds.

[0037] The model optimization unit 23 implements the lightweight model through two steps: quantization-aware training and model pruning. In quantization-aware training, the model parameters are first reduced from 32-bit floating point numbers to 8-bit fixed point numbers. The quantization process can be expressed as:

[0038] Where scale is the quantization scale factor. In order to reduce the quantization error, the floating-point weights and quantized weights are maintained simultaneously during the training process, the floating-point weights are updated by back propagation, and then the quantized weights are requantized. In terms of model pruning, a structured pruning method based on weight importance is adopted. Specifically, the L1 norm of each weight parameter is calculated as an importance indicator. When the L1 norm of the weight is less than the preset threshold (set to 0.01), the connection is pruned. Through the iterative pruning and retraining process, the model parameters are finally reduced by about 60%, while keeping the model performance loss within 5%.

[0039] After the above optimization, the model size of the state assessment module 20 is reduced to less than 40% of the original model, and the single inference time is less than 10ms, which fully meets the deployment requirements of edge computing devices. At the same time, through knowledge transfer and sophisticated optimization strategies, the evaluation accuracy of the model is ensured, providing reliable environmental state assessment results for subsequent control decisions.

[0040] The state assessment module 20 also includes three functional units: feature extraction, dimension reduction processing and assessment calculation, which together constitute the data processing pipeline for environmental state assessment. The specific implementation process of these three units is described in detail below.

[0041] A feature extraction unit 24 is used to divide the standardized environmental data into time windows according to specific time lengths, and calculate the average value, standard deviation and peak value of the data in each time window; The feature extraction unit 24 first divides the standardized environmental data into time windows. Considering the changing characteristics of environmental parameters, this application adopts a basic time window of 180 seconds and introduces a 50% overlap rate to improve the time resolution of feature extraction. In each time window, statistical features including mean, standard deviation and peak value are calculated. Specifically, for temperature data, the calculation formula is: Mean:

[0042] Standard Deviation:

[0043] Peak:

[0044] in is the sampling point in the window, and n is the number of sampling points in the window. In order to capture the changing characteristics of different time scales, a multi-scale window analysis method is also used to expand the basic window length to 360 seconds and 720 seconds, respectively, to form a feature set of three time scales. In this way, for each type of environmental data (temperature, humidity, gas concentration, light), 9 statistical features (3 time scales × 3 statistics) can be obtained.

[0045] A dimension reduction processing unit 25 is used to perform principal component analysis on the mean value, standard deviation and peak value, retain the principal component with an explained variance rate of 95%, and obtain a feature vector after dimension reduction; The dimension reduction processing unit 25 uses the principal component analysis (PCA) method to reduce the dimension of the extracted high-dimensional features. First, the feature matrix X is centralized, that is:

[0046] Where μ is the mean vector of each feature. Then calculate the covariance matrix:

[0047] And solve the eigenvalue equation:

[0048] Get the eigenvalue and the corresponding eigenvector . The eigenvalues ​​are arranged in descending order , calculate the cumulative explained variance rate:

[0049] This application sets ≥95% is used as the threshold for principal component selection, that is, retaining principal components that can explain 95% or more of the variance of the original data. Through experimental verification, it is usually only necessary to retain the first 4-6 principal components to meet the requirements, which significantly reduces the feature dimension. At the same time, in order to improve the robustness of PCA in practical applications, the incremental PCA algorithm is used, which can dynamically update the feature space according to the newly added data, and the computational complexity is reduced from O(n²) to O(n).

[0050] The evaluation calculation unit 26 is used to input the dimension-reduced feature vector into the lightweight edge computing model to calculate the temperature anomaly score, the humidity anomaly score, the harmful gas concentration anomaly score and the mildew risk level.

[0051] The evaluation calculation unit 26 is responsible for inputting the reduced feature vector into the lightweight edge computing model to generate the final evaluation result. Taking into account the correlation between different evaluation indicators, this application adopts a multi-task learning framework, that is, the model outputs four evaluation indicators at the same time. In the specific implementation, four parallel output branches are designed in the last layer of the model, and each branch corresponds to an evaluation indicator. Among them, the temperature anomaly score, humidity anomaly score and harmful gas concentration anomaly score are represented by continuous values ​​in the range of [0,1] and implemented by the Sigmoid function; the mildew risk level adopts a three-category output (low, medium and high), and the Softmax function is used to calculate the probability distribution of each level. In order to balance the training of different evaluation tasks, a dynamically weighted loss function is introduced:

[0052] in is the loss function of each task, is the corresponding weight coefficient. The weight coefficient is adaptively adjusted through the task uncertainty: is the prediction uncertainty of the task, is the smoothing factor (set to 0.5).

[0053] Through the collaborative work of the above three functional units, the state assessment module 20 can extract key features from the original environmental data and generate accurate environmental state assessment results. Experiments show that the scheme has achieved a good balance between assessment accuracy and computational efficiency. The average accuracy of the assessment results is over 95%, and the computational time of a single assessment is controlled within 20ms.

[0054] A parameter calculation module 30 is used to map the environmental state assessment result to a Riemann manifold space, construct a kernel regression model based on the mapped Riemann manifold space, learn the state control mapping relationship between the environmental state and the control parameters through the kernel regression model, and perform convex optimization calculation on the kernel regression model to obtain optimal control parameters; wherein the optimal control parameters include a PTC heating power parameter, a ventilation system speed parameter, and an ultraviolet irradiation intensity parameter; like Figure 4 As shown, the parameter calculation module 30 includes: A manifold construction unit 31 is used to map the environmental state assessment result to a symmetric positive definite matrix manifold space, and design a corresponding Riemann metric tensor based on the physical characteristics of the environmental state assessment result; The core task of the parameter calculation module 30 is to achieve the mapping optimization from the environmental state to the control parameters. Among them, the manifold construction unit 31 and the mapping conversion unit 32 jointly construct a state representation framework based on Riemannian geometry. The specific implementation process of these two units is described in detail below.

[0055] The manifold construction unit 31 first maps the environmental status assessment results to the symmetric positive definite matrix (SPD) manifold space. In specific implementation, the four-dimensional assessment result vector v=[v1,v2,v3,v4] (corresponding to the temperature anomaly score, humidity anomaly score, harmful gas concentration anomaly score and mildew risk level, respectively) is converted into a 4×4 symmetric positive definite matrix P. The conversion process adopts the exponential mapping method:

[0056] Where S is an antisymmetric matrix whose elements are constructed from the evaluation result vector in the following way:

[0057] i,j=1,2,3,4. This construction method ensures the positive definiteness of the matrix P while retaining the correlation information between the dimensions of the evaluation results.

[0058] In order to accurately describe the geometric characteristics of the SPD manifold space, this application designs a Riemann metric tensor based on the physical characteristics of the environmental state. At the point P of the matrix, for two tangent vectors U and V in the tangent space, the Riemann metric is defined as:

[0059] This metric takes into account the physical meaning of each dimension of the environmental state. For example, the coupling effect of temperature and humidity changes will be reflected in the non-diagonal elements of the metric tensor. At the same time, the weight coefficient is introduced Adjust the importance of different dimensions:

[0060] in is the basic matrix. The weight coefficient is determined by analyzing the influence of each dimension on the control effect in the historical data. For example, the weight coefficient of temperature and humidity is set to 0.3, the concentration of harmful gases is 0.2, and the mold risk level is 0.2.

[0061] The mapping conversion unit 32 is used to establish a distance-preserving mapping from the Euclidean space to the symmetric positive definite matrix manifold space to ensure the continuity when performing the state conversion from the Euclidean space to the symmetric positive definite matrix manifold space.

[0062] The mapping conversion unit 32 is responsible for establishing a distance-preserving mapping relationship from the Euclidean space to the SPD manifold space. The distance-preserving mapping needs to satisfy two key properties: isometricity and continuity. To achieve isometricity, the present application designs a mapping function based on geodesic distance. Given two points in the Euclidean space and , their corresponding points in the SPD manifold space are and , the geodesic distance is defined as:

[0063] in represents the Frobenius norm, and log represents the matrix logarithm operation.

[0064] To ensure the continuity of the mapping, this application adopts a piecewise smooth mapping strategy. First, the Euclidean space is divided into multiple local regions, and a quadratically differentiable mapping function is constructed in each local region:

[0065] Where A(x) is the affine transformation matrix and e is the matrix exponential operation. Continuous transition is achieved between adjacent regions by introducing a smooth transition function:

[0066] Where h(t) is the cubic spline function and r is the radius of the transition area. In this way, the global mapping function can be expressed as:

[0067] in is the weight coefficient, satisfying .

[0068] Through the above design, the manifold construction unit 31 and the mapping conversion unit 32 realize the effective representation of the environmental state in the Riemann manifold space, providing a good mathematical basis for subsequent parameter optimization. Experiments show that the scheme can accurately capture the intrinsic geometric structure of the environmental state, the mapping error is controlled within 3%, and has good numerical stability.

[0069] Wherein, the parameter calculation module 30 also includes: A data collection unit 33, used to collect state transition data and control input instructions of the garbage bin under different working conditions; The data acquisition unit 33 collects state transition data and control instructions by running experiments in the system. In the specific implementation, a set of working condition coverage experimental schemes are designed, including normal temperature conditions (15-25℃), high temperature conditions (25-35℃), high humidity conditions (relative humidity 70-90%) and composite conditions. Under each working condition, the control parameters (PTC heating power, ventilation system speed, ultraviolet irradiation intensity) are combined according to the orthogonal experimental design to form a complete experimental matrix. The sampling period is set to 1 second, and each group of experiments lasts for 1 hour. The state data is collected through an industrial-grade data acquisition card, including 16 state quantities such as temperature, humidity, and gas concentration. At the same time, the control instructions output by the controller are recorded, including the control parameters of the three actuators and their timing information. In order to improve the data quality, the data screening method based on Shannon entropy is used to remove data points with low information content, and finally a training data set containing 10,000 sets of valid state transition samples is established.

[0070] A model learning unit 34, configured to construct a kernel regression model in the symmetric positive definite matrix manifold space, and define a geodesic distance using an exponential mapping and a logarithmic mapping; The model learning unit 34 constructs a kernel regression model in the symmetric positive definite matrix manifold space. First, the kernel function is defined:

[0071] in is the geodesic distance, and σ is the kernel width parameter.

[0072] Geodesic distances are defined using exponential and logarithmic mappings:

[0073] in represents matrix logarithm operation, is the Frobenius norm. In order to improve the computational efficiency, Taylor expansion is used to approximate the matrix logarithm:

[0074] Taking the number of expansion terms n=3 can obtain sufficient accuracy. Based on the above kernel function, a kernel regression model is constructed:

[0075] in is the weight coefficient to be learned, is the training sample. Considering the large number of training samples, the random feature approximation method is used to reduce the computational complexity and map the kernel function to a finite-dimensional feature space:

[0076] in is a random feature map.

[0077] The parameter optimization unit 35 is used to construct an objective function based on the kernel regression model and solve the optimal control parameters using a gradient projection algorithm with an adaptive step size.

[0078] The parameter optimization unit 35 is responsible for solving the optimal control parameters. First, an optimization function considering multiple objectives is constructed:

[0079] in is the control parameter vector, J1 represents the control effect term (such as temperature and humidity deviation), J2 represents the energy consumption term, J3 represents the system response time term, and w i is the weight coefficient. The constraints include parameter range constraints:

[0080] And the state constraints:

[0081] The gradient projection algorithm with adaptive step size is used to solve the optimization problem. The iterative format is:

[0082] Where P[] represents the projection operator, which projects the solution into the feasible domain. is the adaptive step size. The step size update adopts the Armijo criterion:

[0083] in , m is the minimum non-negative integer that makes the objective function fully decrease. In order to improve the optimization efficiency, the momentum term is introduced to accelerate the convergence:

[0084] in is the momentum coefficient, and its value is 0.9.

[0085] Through the collaborative work of the above three units, the parameter calculation module 30 can construct an accurate state-control mapping relationship in the Riemann manifold space based on historical operation data, and quickly solve the optimal control parameters. Experimental verification shows that compared with the traditional Euclidean space optimization method, this solution improves control accuracy by about 20%, optimizes convergence speed by about 30%, and has strong anti-disturbance ability.

[0086] The control execution module 40 is used to generate a control instruction according to the optimal control parameter, send the control instruction to the corresponding actuator through the hardware driver interface set on the trash can, and execute the control instruction; and estimate the real-time state in the trash can based on Kalman filtering to obtain a real-time state value, compare the real-time state value with a preset state threshold, and when the real-time state value is less than the preset state threshold, adjust the control instruction according to the deviation degree of the real-time state value; The control execution module 40 includes a strategy generation unit 41, an instruction execution unit 42 and a feedback adjustment unit 43, which together complete the closed-loop control from control parameters to actual execution. The specific implementation process of these three units is described in detail below. Figure 5 As shown, the control execution module 40 includes: A strategy generating unit 41, for generating a control instruction according to the optimal control parameters, wherein the control instruction includes a start timing of the PTC heating power, an adjustment timing of the ventilation system speed, and a duration of ultraviolet irradiation; The strategy generation unit 41 is responsible for converting the optimal control parameters into a specific control instruction sequence. For PTC heating power control, a soft start strategy is adopted to avoid high current shock. The startup process is divided into three stages: preheating stage (10% rated power, lasting 5 seconds), slow rise stage (power increases linearly at a slope of 20% / second) and stabilization stage (reaching the target power). The ventilation system speed adopts a fuzzy PID control strategy, and the speed change is determined according to the temperature and humidity deviation. The control rule is: if the temperature deviation is large and the upward trend is obvious, the speed is greatly increased; if the temperature is close to the target value, a small step size is used for fine-tuning. The control cycle is set to 2 seconds, and the maximum speed change allowed in each cycle is 200rpm. Ultraviolet irradiation adopts an intermittent control mode, and the working cycle is "on for 30 minutes, off for 15 minutes". At the same time, the opening time is adaptively adjusted according to the ambient humidity. For every 10% increase in humidity, the opening time is correspondingly extended by 5 minutes.

[0087] An instruction execution unit 42, used to send control instructions to corresponding actuators through a hardware driver interface, wherein the actuators include a PTC heater, a ventilation system and an ultraviolet lamp; The instruction execution unit 42 communicates with the actuator through a standardized hardware driver interface. The PTC heater uses PWM modulation to control power output. The PWM frequency is set to 1kHz and the duty cycle resolution is 0.1%. In order to improve the control accuracy, a feedforward compensation link is added:

[0088] Where uff(t) is the feedforward control quantity calculated based on the system model, and ufb(t) is the feedback correction quantity. The ventilation system uses the 485 bus communication protocol, and the control frame format is: [address code (1byte)][function code (1byte)][data length (1byte)][speed value (2bytes)][check code (2bytes)], and the communication baud rate is 115200bps. The ultraviolet lamp control adopts the relay switch mode. In order to extend the life of the lamp tube, a soft start circuit is designed. The voltage rises slowly during startup, and the time is about 0.5 seconds. At the same time, the hardware-level fault protection function is realized. When abnormal conditions such as overcurrent and overtemperature are detected, the power supply of the corresponding actuator is automatically cut off.

[0089] The feedback adjustment unit 43 estimates the real-time state in the trash can based on the Kalman filter to obtain a real-time state value, compares the real-time state value with a preset state threshold, and when the real-time state value is less than the preset state threshold, adjusts the optimal control parameter online through the PI controller.

[0090] The feedback adjustment unit 43 implements state estimation and online parameter adjustment based on Kalman filtering. The state estimation adopts the extended Kalman filter (EKF) algorithm, and the state equation is:

[0091] The observation equation is:

[0092] Where w(k) and v(k) are process noise and observation noise, respectively, and their covariance matrices are Q and R, respectively.

[0093] Prediction steps:

[0094] Update steps:

[0095] Among them, F(k) and H(k) are the Jacobian matrices of the state equation and observation equation, respectively.

[0096] After comparing the state estimation value with the preset threshold, the improved PI controller is used to adjust the parameters. The proportional gain Kp and integral time Ti of the controller are adjusted online using the fuzzy adaptive method:

[0097] Where e is the state deviation, is the deviation change rate, f1 and f2 are fuzzy rules designed based on expert experience. The output of the PI controller is used to correct the optimal control parameters: ,in The correction amount is used to ensure that the corrected parameters are still within the feasible domain through limiting processing. In order to improve the control stability, a dead zone mechanism is introduced, and the parameter adjustment is only started when the state deviation exceeds the set threshold (temperature ±0.5℃, humidity ±2%).

[0098] Through the cooperation of the above three units, the control execution module 40 realizes the accurate mapping from the optimal control parameters to the actual control effect, and can adjust the control strategy in time according to the real-time status. The experimental results show that the control accuracy of this scheme is about 25% better than that of the traditional PID control, the system response time is shortened by about 30%, and it has good anti-interference ability and robustness.

[0099] The control execution module 40 further includes a monitoring unit 44, an evaluation unit 45 and an optimization unit 46, which constitute the performance evaluation and optimization feedback link of the system. The specific implementation process of these three units is described in detail below.

[0100] A monitoring unit 44 is used to obtain the working state of the actuator and the change of environmental parameters with a sampling period of 100 milliseconds; The monitoring unit 44 uses a high-precision data acquisition system to achieve real-time monitoring of the actuator working state and environmental parameters. The sampling period is set to 100 milliseconds to ensure the capture of the fast dynamic characteristics of the system. The monitoring of the actuator working state includes: the actual power output (measurement accuracy ±1W) and surface temperature (measurement accuracy ±0.1℃) of the PTC heater, the actual speed (measurement accuracy ±10rpm) and air flow (measurement accuracy ±0.1m³ / h) of the ventilation system, the real-time power (measurement accuracy ±0.1W) and cumulative working time (accuracy 1s) of the ultraviolet lamp. The environmental parameter monitoring adopts a distributed sensor network, including the temperature field distribution (measurement accuracy ±0.2℃), humidity field distribution (measurement accuracy ±1%RH) and gas concentration distribution (measurement accuracy ±1ppm) of 16 measuring points. All collected data are processed by digital filtering, using a second-order Butterworth low-pass filter with a cut-off frequency of 5Hz to eliminate high-frequency noise interference.

[0101] An evaluation unit 45 is used to calculate the dehumidification rate, energy efficiency and mold inhibition rate based on the working state and the change of the environmental parameters, and perform result analysis in a sliding manner with a 24-hour window and a 1-hour step length to obtain an execution result; The evaluation unit 45 calculates the key performance indicators of the system based on the monitoring data. Dehumidification rate The calculation uses the mass balance method:

[0102] in is the mass of water removed per unit time, is the air density, and V is the effective volume of the trash can. Energy efficiency Defined as the environmental improvement effect produced by unit energy consumption:

[0103] in is the temperature change, and P is the total input power. Evaluation by image recognition methods:

[0104] Where At is the current mold area, and A0 is the initial mold area. The evaluation adopts a sliding analysis method with a 24-hour window and a 1-hour step size, that is, the performance indicators are updated every hour and the data of the last 24 hours are retained. The periodic characteristics of the performance indicators are analyzed by Fourier transform to extract the daily variation pattern and fluctuation characteristics. At the same time, the weighted moving average method is introduced to smooth short-term fluctuations:

[0105] in is the weight coefficient, satisfying , giving greater weight to recent data.

[0106] The optimization unit 46 is used to build a multi-objective evaluation model based on the execution result, and optimize the initial value selection and dynamic adjustment rules of the control parameters by using a reinforcement learning method.

[0107] The optimization unit 46 builds a parameter optimization framework based on deep reinforcement learning. First, a multi-objective evaluation model is established:

[0108] in is the weight coefficient of each objective, C is the system constraint violation measure, is the penalty factor. The Double DQN structure is used to achieve parameter optimization. The state space S includes the environment state and the system operation state. The action space A is the adjustment amount of the control parameter. The reward function is the above evaluation model R. The network structure adopts a four-layer fully connected network. The number of hidden layer neurons is 256, 128, and 64 respectively. The activation function is LeakyReLU. In order to improve the training efficiency, the priority experience replay mechanism is introduced:

[0109] in is the TD error, is the priority index (value is 0.6). The parameter update adopts the soft update strategy:

[0110] in is the soft update coefficient (value is 0.01).

[0111] In order to adapt to the dynamic changes of the environment, the optimization unit 46 also implements a fast adaptation mechanism based on Meta-Learning. Specifically, a task distribution p(T) is constructed, and each task corresponds to a different combination of environmental conditions. A good parameter initialization is learned through the Model-Agnostic Meta-Learning (MAML) algorithm. , so that only a small number of samples are needed to quickly adapt to new tasks. The outer loop optimization goal of the algorithm is:

[0112] in This method significantly improves the system's ability to adapt to new environmental conditions, and the average adaptation time is shortened from 4 hours in traditional methods to less than 1 hour.

[0113] Through the collaborative work of the above three units, the system achieves continuous monitoring, evaluation and optimization of performance. Experimental results show that this solution can maintain the system performance indicators at a high level, improve dehumidification efficiency by 20%, reduce energy consumption by 15%, and increase mold inhibition rate by 25%. It also has strong environmental adaptability and robustness.

[0114] The collaborative work module 50 is used to build a blockchain-based smart contract, which performs resource bidding and task allocation based on the geographical location, processing capacity and energy efficiency between the garbage bins, so as to realize collaborative work among multiple garbage bins.

[0115] The collaborative working module 50 includes: A bidding unit 51 is used to construct a smart contract based on blockchain to bid for resources according to the geographical location, processing capacity and energy efficiency between garbage bins; Bidding unit 51 builds a decentralized resource bidding mechanism based on Ethereum smart contracts. Smart contracts contain three core functions: resource registration, bidding execution, and transaction confirmation. The resource registration part defines the basic attribute structure of the trash can: struct Resource {uint256 id; bytes32 location; uint256 capacity; uint256 efficiency; uint256 availability;}.

[0116] The bidding rules adopt a multi-round sealed bidding method, and each round of bidding lasts 300 seconds. The bidding information is encrypted using zero-knowledge proof technology: bid = keccak256(abi.encodePacked(value, nonce)), Where value is the bid price and nonce is a random number. The scoring function takes into account geographical proximity (D), processing power matching (C) and energy efficiency (E): Score = w1·(1 / D) + w2·C + w3·E The weight coefficients {w1, w2, w3} are obtained through historical data training. In order to improve the efficiency of contract execution, a hierarchical storage strategy is adopted to store frequently accessed state data in memory and other data on the blockchain.

[0117] The scheduling unit 52 is used to synchronize data between trash cans using a publish-subscribe mode to achieve distributed task scheduling and fault tolerance; The scheduling unit 52 implements a distributed collaboration mechanism based on the publish-subscribe model. Apache Kafka is used as the message middleware, and four topics are designed: state synchronization, task distribution, fault reporting, and performance monitoring. The message format uses Protocol Buffers encoding to reduce transmission overhead. The state synchronization adopts an incremental synchronization strategy, which only transmits the changed state data, and the synchronization cycle is 5 seconds. Task distribution adopts priority queue management: priority_queue <task>taskQueue, the priority is calculated based on task urgency and resource utilization. Fault tolerance mechanisms include: master-slave switching (fault detection time threshold 2 seconds), task redistribution (based on consistent hashing algorithm) and state rollback (using write-ahead log WAL mechanism). To ensure message reliability, a message replication mechanism based on the Raft algorithm is implemented, and the replication factor is set to 3.

[0118] The optimization unit 53 is used to collect the operation data of the collaborative work between multiple trash cans and optimize the resource allocation strategy and task scheduling algorithm using the deep reinforcement learning method.

[0119] The optimization unit 53 is responsible for the continuous optimization of the collaborative strategy. First, a data collection pipeline is constructed with a sampling period of 10 seconds. The collected content includes: task execution status, resource utilization, energy consumption data and environmental parameters. The data preprocessing uses a sliding window method (window size 1 hour, step length 10 minutes) for feature extraction. Based on the processed data, a deep reinforcement learning model is constructed to optimize the collaborative strategy. The state space includes: S = {s_resource, s_task, s_environment} Where s_resource represents the resource state vector, s_task represents the task state vector, and s_environment represents the environment state vector. The action space is defined as: A = {a_allocation, a_scheduling} Represents resource allocation and task scheduling decisions.

[0120] The reward function is designed to take into account multiple objectives:

[0121] in represents the processing efficiency reward, represents the load balancing reward, Indicates energy saving reward, Represents the constraint violation penalty. The deep neural network adopts the Actor-Critic architecture. Both the Actor network and the Critic network adopt a four-layer fully connected structure. The number of hidden layer units is [512, 256, 128, 64], and the activation function is PReLU. To improve the training efficiency, a hierarchical reinforcement learning method is adopted: complex collaborative decision-making is decomposed into two subtasks: resource allocation and task scheduling. Each subtask is trained independently and then integrated through the policy distillation method.

[0122] At the same time, the meta-learning mechanism is introduced to improve the generalization ability of the model. The task pool T is constructed to include collaborative decision-making tasks in different scenarios, and the MAML algorithm is used to learn a good parameter initialization The optimization goal is:

[0123] To handle the uncertainty of the environment, the Monte Carlo Tree Search (MCTS) method is integrated for decision optimization: at each decision point, the search tree is expanded by MCTS, and the action selection is combined with the value evaluation of the neural network. The search depth is set to 100 and the UCB exploration coefficient is set to 1.4.

[0124] Through the collaborative work of the above three units, the system realizes intelligent collaborative management among multiple trash cans. Experimental results show that the scheme significantly improves the overall performance of the system: resource utilization is increased by 25%, task processing efficiency is increased by 30%, system energy consumption is reduced by 20%, and it has good scalability and robustness. The network communication overhead is controlled at a low level, the average message delay is less than 100ms, and the packet loss rate is less than 0.1%.

[0125] The basic principles of the present disclosure are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present disclosure are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present disclosure. In addition, the specific details disclosed above are only for the purpose of illustration and ease of understanding, and are not limitations. The above details do not limit the present disclosure to the necessity of adopting the above specific details to be implemented.

[0126] In the present disclosure, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. The block diagrams of the devices, devices, equipment, and systems involved in the present disclosure are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagram. As will be recognized by those skilled in the art, these devices, devices, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including", "comprising", "having", etc. are open words, meaning "including but not limited to", and can be used interchangeably with them. The words "or" and "and" used here refer to the words "and / or" and can be used interchangeably with them, unless the context clearly indicates otherwise. The word "such as" used here refers to the phrase "such as but not limited to", and can be used interchangeably with it.

[0127] Additionally, as used herein, "or" used in a list of items beginning with "at least one" indicates a disjunctive list, so that, for example, a list of "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not mean that the example described is preferred or better than other examples.

[0128] It should also be noted that in the system and method of the present disclosure, each component or each step can be decomposed and / or recombined. Such decomposition and / or recombination should be regarded as equivalent solutions of the present disclosure.

[0129] Various changes, substitutions, and modifications of the techniques described herein may be made without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of the present disclosure is not limited to the specific aspects of the processes, machines, manufactures, compositions of events, means, methods, and actions described above. Currently existing or later to be developed processes, machines, manufactures, compositions of events, means, methods, or actions that perform substantially the same functions or achieve substantially the same results as the corresponding aspects described herein may be utilized. Thus, the appended claims include such processes, machines, manufactures, compositions of events, means, methods, or actions within their scope.

[0130] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present disclosure. Therefore, the present disclosure is not intended to be limited to the aspects shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

[0131] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present disclosure to the forms disclosed herein. Although multiple example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.< / task>

Claims

1. An intelligent dehumidifying and mildew-removing kitchen waste bin, comprising: A data acquisition module is used to collect environmental data through multiple sensors arranged in the trash can, perform outlier detection and sliding window average processing on the environmental data, and obtain standardized environmental data; wherein the environmental data includes temperature data, humidity data, harmful gas concentration data and light intensity data; A state assessment module, used to analyze the standardized environmental data using a lightweight edge computing model to generate an environmental state assessment result; wherein the environmental state assessment result includes a temperature anomaly score, a humidity anomaly score, a harmful gas concentration anomaly score, and a mold risk level; A parameter calculation module, used for mapping the environmental state assessment result to a Riemann manifold space, constructing a kernel regression model based on the mapped Riemann manifold space, learning the state control mapping relationship between the environmental state and the control parameters through the kernel regression model, and performing convex optimization calculation on the kernel regression model to obtain optimal control parameters; wherein the optimal control parameters include a PTC heating power parameter, a ventilation system speed parameter, and an ultraviolet irradiation intensity parameter; A control execution module, configured to generate a control instruction according to the optimal control parameter, send the control instruction to the corresponding actuator through a hardware driver interface set on the trash can, and execute the control instruction; and estimate the real-time state in the trash can based on Kalman filtering to obtain a real-time state value, compare the real-time state value with a preset state threshold, and when the real-time state value is less than the preset state threshold, adjust the control instruction according to the deviation degree of the real-time state value; The collaborative work module is used to build a blockchain-based smart contract, which conducts resource bidding and task allocation based on the geographical location, processing capacity and energy efficiency between garbage bins, thereby realizing collaborative work among multiple garbage bins.

2. The intelligent dehumidifying and mildew-removing kitchen waste bin according to claim 1 is characterized in that: The data acquisition module comprises: A temperature and humidity collection unit, used to set a temperature and humidity sensor to collect temperature and humidity data in the barrel in real time; The gas concentration collection unit is used to set the gas sensor to detect the concentration data of ammonia and hydrogen sulfide in the barrel; A light collection unit is used to set a photosensitive sensor to monitor the light intensity data in the barrel; The data preprocessing unit is used to perform outlier detection on the environmental data and perform smoothing using a sliding window averaging method to obtain the standardized environmental data.

3. The intelligent dehumidifying and mildew-removing kitchen waste bin according to claim 1 is characterized in that: The status assessment module includes: A model building unit is used to build a three-layer lightweight neural network. During the construction, it is ensured that the number of nodes in the input layer matches the number of sensors provided in the trash can, the hidden layer uses 32 neurons, and the output layer corresponds to different dimensional indicators of the environmental state; A knowledge migration unit, configured to adopt a soft label conversion strategy with specific parameters to migrate knowledge in an existing pre-trained environment evaluation model to the lightweight edge computing model; The model optimization unit is used to reduce the accuracy of model parameters from high floating points to low fixed points using quantization-aware training methods, and to compress the model size using model pruning technology that sets weight thresholds.

4. The intelligent dehumidifying and mildew-removing kitchen waste bin according to claim 3 is characterized in that: The status assessment module also includes: A feature extraction unit, used to divide the standardized environmental data into time windows according to specific time lengths, and calculate the average value, standard deviation and peak value of the data in each time window; A dimension reduction processing unit is used to perform principal component analysis on the mean value, standard deviation and peak value, retain the principal component with an explained variance rate of 95%, and obtain a feature vector after dimension reduction; An evaluation and calculation unit is used to input the dimensionally reduced feature vector into the lightweight edge computing model to calculate the temperature anomaly score, the humidity anomaly score, the harmful gas concentration anomaly score and the mildew risk level.

5. The intelligent dehumidifying and mildew-removing kitchen waste bin according to claim 1 is characterized in that: The parameter calculation module comprises: A manifold construction unit, used for mapping the environmental state assessment result to a symmetric positive definite matrix manifold space, and designing a corresponding Riemann metric tensor based on the physical properties of the environmental state assessment result; The mapping conversion unit is used to establish a distance-preserving mapping from the Euclidean space to the symmetric positive definite matrix manifold space to ensure the continuity when performing the state conversion from the Euclidean space to the symmetric positive definite matrix manifold space.

6. The intelligent dehumidifying and mildew-removing kitchen waste bin according to claim 5 is characterized in that: The parameter calculation module also includes: A data acquisition unit, used to collect state transition data and control input instructions of the garbage bin under different working conditions; A model learning unit, used for constructing a kernel regression model in the symmetric positive definite matrix manifold space, and defining a geodesic distance by using an exponential mapping and a logarithmic mapping; The parameter optimization unit is used to construct an objective function based on the kernel regression model and solve the optimal control parameters using a gradient projection algorithm with an adaptive step size.

7. The intelligent dehumidifying and mildew-removing kitchen waste bin according to claim 1 is characterized in that: The control execution module includes: A strategy generating unit, configured to generate a control instruction according to the optimal control parameter, wherein the control instruction includes a start timing of the PTC heating power, an adjustment timing of the ventilation system speed, and a duration of ultraviolet irradiation; An instruction execution unit, used to send control instructions to corresponding actuators through a hardware driver interface, wherein the actuators include a PTC heater, a ventilation system, and an ultraviolet lamp; The feedback adjustment unit estimates the real-time state in the trash can based on the Kalman filter to obtain a real-time state value, compares the real-time state value with a preset state threshold, and when the real-time state value is less than the preset state threshold, adjusts the optimal control parameter online through the PI controller.

8. The intelligent dehumidifying and mildew-removing kitchen waste bin according to claim 7 is characterized in that: The control execution module also includes: A monitoring unit, used to obtain the working status of the actuator and changes in environmental parameters with a sampling period of 100 milliseconds; An evaluation unit, for calculating the dehumidification rate, energy efficiency and mold inhibition rate based on the working state and the change of the environmental parameters, and performing result analysis in a sliding manner with a 24-hour window and a 1-hour step length to obtain an execution result; The optimization unit is used to build a multi-objective evaluation model based on the execution results, and optimize the initial value selection and dynamic adjustment rules of the control parameters by using a reinforcement learning method.

9. The intelligent dehumidifying and mildew-removing kitchen waste bin according to claim 1, characterized in that: The collaborative working module includes: The bidding unit is used to build a blockchain-based smart contract to bid for resources based on the geographical location, processing capacity, and energy efficiency of the garbage bins; The scheduling unit is used to synchronize data between trash cans using a publish-subscribe model, and to implement distributed task scheduling and fault tolerance; The optimization unit is used to collect the operating data of the collaborative work between multiple trash cans and use deep reinforcement learning methods to optimize the resource allocation strategy and task scheduling algorithm.

Images