Landscaping waste recycling system and method

Abstract

The invention discloses a garden greening waste recycling system and method, and relates to the field of garden waste, and the method comprises the steps: S1, three-dimensional grid monitoring and data acquisition, S2, multi-dimensional feature data set construction, S3, spatial distribution anomaly recognition, S4, control decision parameter generation, S5, layered and partitioned compost regulation and control execution, S6, layered and partitioned compost regulation and control execution, and S5, layered and partitioned compost regulation and control execution. And updating the compost state prediction model. According to the invention, through the three-dimensional grid monitoring network, three-dimensional and omnibearing perception of the internal state of the pile body is realized, heterogeneity of spatial distribution can be accurately captured, and the defects of few monitoring points and incomplete information of a traditional method are overcome.

Description

Technical Field

[0001] This application relates to the field of garden waste, and more specifically, to a system and method for recycling garden greening waste. Background Technology

[0002] Garden and landscaping waste mainly includes organic materials such as branches and leaves from pruning, lawn mowers, and fallen leaves. If not properly disposed of, this waste will occupy a large amount of land resources and may cause environmental pollution. Composting is an effective way to convert garden and landscaping waste into organic fertilizer, which can both realize the resource utilization of waste and produce high-quality soil conditioners.

[0003] Traditional composting methods rely heavily on manual experience for management, which presents several problems: monitoring methods are limited, typically only monitoring temperature on the surface or at a few points, failing to comprehensively understand the spatial distribution within the compost pile; control measures are rudimentary, often relying on timed turning or uniform ventilation, failing to precisely control localized abnormal areas within the pile; predictive capabilities are lacking, unable to anticipate potential problems during composting, leading to long processing cycles and unstable product quality; energy consumption is high, as insufficient control strategies result in overly frequent or inadequate turning and ventilation, causing energy waste or poor composting results; and gas emission control is ineffective, failing to effectively monitor and control odorous gases such as ammonia and hydrogen sulfide generated during composting.

[0004] In recent years, although some scholars have proposed applying sensor technology and automated control to the composting process, existing technologies still have significant shortcomings: First, the monitoring network is not refined enough and it is difficult to capture the spatial heterogeneity inside the compost pile; second, there is a lack of intelligent prediction models, which cannot predict future development trends based on the current state; and third, the control strategy is too simplistic and fails to seek the optimal balance among multiple objectives such as composting cycle, product quality, energy consumption, and environmental impact.

[0005] Therefore, there is an urgent need to develop a method for composting garden waste that integrates refined monitoring, intelligent prediction, and multi-objective optimization. Summary of the Invention

[0006] In order to overcome a series of defects in the existing technology, the purpose of this application is to provide, in response to the above problems, a method for recycling garden and landscaping waste, characterized in that,

[0007] Step S1, Three-dimensional grid monitoring and data acquisition: The compost pile is divided into layers and zones to form multiple monitoring units. Temperature and humidity sensors are installed in the monitoring units, and oxygen concentration and pH sensors are installed in each layer. Multi-point monitoring data inside the pile are acquired at a preset acquisition frequency.

[0008] Step S2, Construction of multidimensional feature dataset: Based on the multi-point monitoring data, spatial distribution features, statistical features and gradient features are extracted, and combined with the temporal change features of the initial composition parameters and monitoring parameters of the compost material within a preset time period, to construct a multidimensional feature dataset;

[0009] Step S3, Spatial Distribution Anomaly Identification: Compare the multidimensional feature dataset with the normal threshold range of each monitoring parameter to identify abnormal regions of temperature, humidity, oxygen concentration and interlayer gradient;

[0010] Step S4, generating control decision parameters: input the multidimensional feature dataset into the composting state prediction model to obtain the temperature, maturity and risk prediction results within the target prediction time period, and input them together with the abnormal area information into a multi-objective optimization algorithm to solve for the control decision parameters of turning frequency, ventilation volume and water addition volume.

[0011] Step S5, Layered and Zoned Composting Control Execution: Based on the control decision parameters and the type of abnormal area, perform local turning, local enhanced ventilation, or local watering on local abnormal areas, and perform full pile turning or full pile forced ventilation on the overall abnormal trend. The execution interval between each control operation shall not be less than the preset minimum interval time.

[0012] Step S6, update the composting state prediction model: record the multidimensional feature dataset, control decision parameters, regulation operations and product quality parameters during multiple batches of composting. When the prediction error between the prediction model output and the actual monitoring exceeds a preset threshold, update the composting state prediction model using the recorded data as training samples.

[0013] Furthermore, the three-dimensional gridded monitoring includes: dividing the compost pile into several vertical layers in the height direction, and dividing each layer into rectangular monitoring units in the horizontal direction; temperature sensors and humidity sensors are installed inside each monitoring unit, and oxygen concentration sensors and pH sensors are installed at multiple sampling points along each layer; all types of sensors are connected to the central control unit for data transmission, and the time interval for collecting monitoring parameters is from several minutes to tens of minutes.

[0014] Furthermore, the multidimensional feature dataset includes at least:

[0015] Spatial distribution characteristics: temperature, humidity, oxygen concentration, and pH value of each monitoring unit at the same time;

[0016] Statistical characteristics: mean and dispersion indices of monitoring parameters at each level;

[0017] Gradient characteristics: Parameter differences between adjacent layers and between adjacent monitoring units within the same layer;

[0018] Compositional characteristics: carbon-nitrogen ratio, moisture content, organic matter content, and particle size distribution of compost materials;

[0019] Time-series characteristics: The trend and rate of change of the monitored parameters within a time window of at least 24 hours.

[0020] Furthermore, the composting status prediction model is a deep learning-based multi-output prediction model that takes the multidimensional feature dataset as input and outputs the temperature distribution, maturity index, and abnormal risk index within the target prediction time period.

[0021] Furthermore, the optimization objectives of the multi-objective optimization algorithm include at least: shortening the composting cycle, improving the quality of compost products, reducing energy consumption, and reducing gas emissions; the constraints include at least: turning frequency, ventilation volume, water addition volume, and the maximum temperature of the compost pile not exceeding the safety limit.

[0022] Furthermore, when the height of the compost pile is between approximately 1.5m and 3.0m, the pile is divided into 3 to 6 vertical layers, each with a thickness of approximately 30cm to 50cm. Each layer is horizontally divided into rectangular monitoring units with sides of approximately 40cm to 60cm. Temperature and humidity sensors are placed at the geometric center of each monitoring unit, and oxygen concentration and pH sensors are set at no less than 3 sampling points in each layer. The temperature and humidity sampling interval is approximately 10min to 30min, and the oxygen concentration and pH sampling interval is approximately 20min to 60min.

[0023] Furthermore, the statistical features include the mean, standard deviation, and range of the monitoring parameters for each layer, and the gradient features include the vertical temperature gradient, humidity gradient, oxygen concentration gradient, and pH gradient. In anomaly identification, when the temperature of the monitoring unit remains outside the normal range of the corresponding composting stage for no less than 8 hours, or when the absolute value of the temperature gradient of the adjacent layer exceeds a preset threshold, the corresponding area is marked as a temperature anomaly area or a gradient anomaly area.

[0024] Furthermore, the composting state prediction model includes a one-dimensional convolutional network layer for extracting temporal features, a long short-term memory network layer for modeling long-term dependencies, and an attention mechanism layer for assigning weights to different features. The output layer has three branches, which are used to predict the temperature distribution, maturity index, and probability of abnormal risks in the next 24 hours, 48 ​​hours, and 72 hours, respectively.

[0025] Furthermore, the multi-objective optimization algorithm employs a non-dominated sorting genetic algorithm with a population size of tens to hundreds of individuals and an evolutionary generation of tens to hundreds of generations. By obtaining a Pareto optimal solution set, control decision parameters such as turning frequency, ventilation volume, and water addition volume are selected from the Pareto optimal solution set according to the current composting conditions.

[0026] A system for recycling garden and landscaping waste includes: a three-dimensional gridded monitoring module responsible for layering and partitioning the compost pile, and deploying temperature, humidity, oxygen concentration, and pH sensors in each monitoring unit.

[0027] Multidimensional feature dataset construction module: This module extracts spatial distribution features, statistical features, gradient features, composition features and time series features during the composting process based on multi-point monitoring data to form a multidimensional feature dataset;

[0028] Spatial distribution anomaly identification module: Based on a multidimensional feature dataset, this module identifies abnormal areas in the composting process by comparing them with normal threshold ranges. Specifically, it identifies abnormal areas in terms of temperature, humidity, oxygen concentration, etc., and marks them for subsequent processing.

[0029] Composting Status Prediction Module: Based on deep learning technology, it predicts the composting process and outputs prediction results such as temperature distribution, maturity, and abnormal risks in the future time period.

[0030] Multi-objective optimization module: By using a multi-objective optimization algorithm, the control decision parameters in the composting process are solved while considering multiple optimization objectives;

[0031] Composting regulation execution module: Based on the control decision parameters, this module executes specific regulation operations and adjusts the composting status and abnormal area types accordingly;

[0032] The composting state prediction model update module records information such as multidimensional feature datasets, control decision parameters, regulation operations, and product quality during each batch of composting. When the prediction error between the output of the composting state prediction model and the actual monitoring data exceeds a preset threshold, the composting state prediction model is updated using these recorded data.

[0033] Compared with the prior art, the beneficial effects of this application are as follows: This application realizes three-dimensional and all-round perception of the internal state of the stack through a three-dimensional gridded monitoring network, which can accurately capture the heterogeneity of spatial distribution and overcome the shortcomings of traditional methods such as few monitoring points and incomplete information.

[0034] This application utilizes a deep learning prediction model to predict future development trends based on the current state, identify potential problems in advance, and achieve a shift from passive response to proactive prevention. It employs a multi-objective optimization algorithm to generate control decisions, seeking the optimal balance among multiple objectives such as composting cycle, product quality, energy consumption, and environmental impact.

[0035] This application implements precise control in a layered and zoned manner based on the location and type of abnormal areas, which can effectively solve local problems while avoiding unnecessary interference with normal areas, thus improving control efficiency.

[0036] Instruction manual illustrations

[0037] Figure 1 This is a schematic diagram of the overall process of a method for recycling landscaping waste, as disclosed in a preferred embodiment of this application. Detailed Implementation

[0038] The technical solution of this application will be clearly and completely described below with reference to the embodiments of this application, but these embodiments should not be construed as limiting the scope of protection of this application.

[0039] Example 1: This example discloses a method for recycling garden and greening waste. Step S1: Three-dimensional grid monitoring and data acquisition: The compost pile is divided into layers and zones to form multiple monitoring units. Temperature sensors and humidity sensors are installed in the monitoring units, and oxygen concentration sensors and pH sensors are installed in each layer. Multi-point monitoring data inside the pile is acquired at a preset acquisition frequency.

[0040] Step S2, Construction of multidimensional feature dataset: Based on the multi-point monitoring data, spatial distribution features, statistical features and gradient features are extracted, and combined with the temporal change features of the initial composition parameters and monitoring parameters of the compost material within a preset time period, to construct a multidimensional feature dataset;

[0041] Step S3, Spatial Distribution Anomaly Identification: Compare the multidimensional feature dataset with the normal threshold range of each monitoring parameter to identify abnormal regions of temperature, humidity, oxygen concentration and interlayer gradient;

[0042] Step S4, generating control decision parameters: input the multidimensional feature dataset into the composting state prediction model to obtain the temperature, maturity and risk prediction results within the target prediction time period, and input them together with the abnormal area information into a multi-objective optimization algorithm to solve for the control decision parameters of turning frequency, ventilation volume and water addition volume.

[0043] Step S5, Layered and Zoned Composting Control Execution: Based on the control decision parameters and the type of abnormal area, perform local turning, local enhanced ventilation, or local watering on local abnormal areas, and perform full pile turning or full pile forced ventilation on the overall abnormal trend. The execution interval between each control operation shall not be less than the preset minimum interval time.

[0044] Step S6, update the composting state prediction model: record the multidimensional feature dataset, control decision parameters, regulation operations and product quality parameters during multiple batches of composting. When the prediction error between the prediction model output and the actual monitoring exceeds a preset threshold, update the composting state prediction model using the recorded data as training samples.

[0045] As an embodiment of this application, the three-dimensional gridded monitoring includes: dividing the compost pile into several vertical layers in the height direction, and dividing each layer into rectangular monitoring units in the horizontal direction; temperature sensors and humidity sensors are installed inside each monitoring unit, and oxygen concentration sensors and pH sensors are installed at multiple sampling points along each layer; all types of sensors are connected to the central control unit for data transmission, and the time interval for collecting monitoring parameters is from several minutes to tens of minutes.

[0046] As one embodiment of this application, the multidimensional feature dataset includes at least:

[0047] Spatial distribution characteristics: temperature, humidity, oxygen concentration, and pH value of each monitoring unit at the same time;

[0048] Statistical characteristics: mean and dispersion indices of monitoring parameters at each level;

[0049] Gradient characteristics: Parameter differences between adjacent layers and between adjacent monitoring units within the same layer;

[0050] Compositional characteristics: carbon-nitrogen ratio, moisture content, organic matter content, and particle size distribution of compost materials;

[0051] Time-series characteristics: The trend and rate of change of the monitored parameters within a time window of at least 24 hours.

[0052] As an embodiment of this application, the composting state prediction model is a deep learning-based multi-output prediction model that takes the multidimensional feature dataset as input and outputs the temperature distribution, maturity index, and abnormal risk index within the target prediction time period.

[0053] As an embodiment of this application, the optimization objectives of the multi-objective optimization algorithm include at least: shortening the composting cycle, improving the quality of compost products, reducing energy consumption, and reducing gas emissions; the constraints include at least: turning frequency, ventilation volume, water addition volume, and the maximum temperature of the compost pile not exceeding the safety limit.

[0054] As an embodiment of this application, when the height of the compost pile is between approximately 1.5m and 3.0m, the pile is divided into 3 to 6 vertical layers, each layer being approximately 30cm to 50cm thick; each layer is horizontally divided into rectangular monitoring units with sides of approximately 40cm to 60cm; temperature and humidity sensors are installed at the geometric center of each monitoring unit, and oxygen concentration and pH sensors are installed at no less than 3 sampling points in each layer. The temperature and humidity sampling interval is approximately 10min to 30min, and the oxygen concentration and pH sampling interval is approximately 20min to 60min.

[0055] As an embodiment of this application, the statistical features include the mean, standard deviation, and range of the monitoring parameters of each layer, and the gradient features include the vertical temperature gradient, humidity gradient, oxygen concentration gradient, and pH gradient; in anomaly identification, when the temperature of the monitoring unit is outside the normal range of the corresponding composting stage for no less than 8 hours, or when the absolute value of the temperature gradient of the adjacent layer exceeds a preset threshold, the corresponding area is marked as a temperature anomaly area or a gradient anomaly area.

[0056] As an embodiment of this application, the composting state prediction model includes a one-dimensional convolutional network layer for extracting temporal features, a long short-term memory network layer for modeling long-term dependencies, and an attention mechanism layer for assigning weights to different features. The output layer has three branches, which are used to predict the temperature distribution, maturity index, and probability of abnormal risks in the next 24 hours, 48 ​​hours, and 72 hours, respectively.

[0057] As an embodiment of this application, the multi-objective optimization algorithm adopts a non-dominated sorting genetic algorithm with a population size of tens to hundreds of individuals and an evolutionary generation of tens to hundreds of generations. By obtaining a Pareto optimal solution set, control decision parameters such as turning frequency, ventilation volume, and water addition volume are selected from the Pareto optimal solution set according to the current composting conditions.

[0058] As an embodiment of this application, the method further includes a gas emission monitoring and control step: setting up a gas sampling port in the compost pile and connecting it to a gas analysis device to monitor the concentrations of ammonia, hydrogen sulfide and methane. When the concentration of any gas exceeds the corresponding preset threshold, measures such as increasing ventilation, adjusting the covering layer or spraying deodorant are triggered. The gas emission monitoring data is used as one of the inputs to the environmental impact target in the multi-objective optimization algorithm.

[0059] As one embodiment of this application, a landscaping waste recycling system includes: a three-dimensional gridded monitoring module: responsible for layering and partitioning the compost pile, and deploying temperature, humidity, oxygen concentration, and pH sensors in each monitoring unit. By collecting multi-point monitoring data within the compost pile, it provides a basis for subsequent data analysis and control decisions. Main functions: compost pile layering and partitioning; deployment and data acquisition of sensors for temperature, humidity, oxygen concentration, and pH; periodic collection of environmental data within the compost pile.

[0060] Multidimensional Feature Dataset Construction Module: Based on multi-point monitoring data, this module extracts spatial distribution features, statistical features, gradient features, composition features, and temporal features during the composting process, constructing a multidimensional feature dataset to provide input data for subsequent anomaly identification, compost status prediction, and optimized control. Main functions: Data extraction: spatial distribution features, statistical features, gradient features, composition features, and temporal features.

[0061] Spatial distribution anomaly identification module: Based on a multidimensional feature dataset, this module identifies abnormal areas in the composting process by comparing them with normal threshold ranges. These abnormal areas include those related to temperature, humidity, and oxygen concentration, and are marked for subsequent processing.

[0062] Composting Status Prediction Module: This module uses deep learning techniques (such as multi-output prediction models, convolutional networks, long short-term memory networks, attention mechanisms, etc.) to predict the composting process and output prediction results such as temperature distribution, maturity, and abnormal risks in the future time period.

[0063] Multi-objective optimization module: This module uses multi-objective optimization algorithms (such as non-dominated sorting genetic algorithm) to solve for control decision parameters (such as turning frequency, ventilation volume and water volume) in the composting process while considering multiple optimization objectives (such as shortening the composting cycle, improving product quality, reducing energy consumption, etc.).

[0064] Composting regulation execution module: Based on the control decision parameters, this module executes specific regulation operations (such as local turning, local ventilation, watering, etc.) and makes adjustments according to the composting status and abnormal area types.

[0065] Compost Status Prediction Model Update Module

[0066] During each batch of composting, this module records information such as multidimensional feature datasets, control decision parameters, regulation operations, and product quality. When the prediction error between the output of the composting state prediction model and the actual monitoring data exceeds a preset threshold, the composting state prediction model is updated using these recorded data.

[0067] Example 2, Material preparation: The main waste materials for landscaping include: 50% pruned tree branches, crushed to 2-4cm, 30% lawn mowing materials, and 20% fallen leaves.

[0068] The initial carbon-to-nitrogen ratio was 28:1, the moisture content was 62%, and the organic matter content was 85%. After the materials were mixed evenly, they were piled into a heap 5m long, 2.5m wide, and 2m high, with a total compost volume of approximately 20 cubic meters.

[0069] Monitoring System Deployment: The stack height is divided into four vertical layers, each 50cm high. Each layer is further divided horizontally into rectangular monitoring units with sides of 50cm, totaling 50 monitoring units per layer (10 along the length and 5 along the width). Temperature Sensors: One sensor is placed at the center of each monitoring unit, totaling 200, with a measurement range of 0-100℃ and an accuracy of ±0.5℃. Humidity Sensors: One sensor is placed at the center of each monitoring unit, totaling 200, with a measurement range of 0-100% and an accuracy of ±2%. Oxygen Concentration Sensors: Five sensors are placed in each layer, totaling 20, with a measurement range of 0-25% and an accuracy of ±0.5%. pH Sensors: Three sensors are placed in each layer, totaling 12, with a measurement range of 4-10 and an accuracy of ±0.1%. All sensors are connected to the central control unit via a wireless data transmission module. Temperature and humidity data are collected every 15 minutes, while oxygen concentration and pH data are collected every 30 minutes.

[0070] Composting process control: Days 1-3 (heating stage): Monitoring showed that the temperature in the central area of ​​the compost pile rose rapidly, from 25℃ to 55℃ within 24 hours, while the temperature in the edge area rose more slowly, remaining at 35-40℃. The multidimensional feature dataset showed that the temperature gradient between the center and the edge reached 20℃, exceeding the preset threshold of 15℃, and was identified as an anomaly. The prediction model output that the temperature in the central area would reach 70℃ in the next 24 hours, indicating a risk of overheating. The multi-objective optimization algorithm yielded the following control decision: the entire compost pile was turned over once, and the ventilation rate was set to 0.15 m³ / (min·m³) after turning. After turning, the temperature distribution was more uniform, and the overall temperature stabilized in the range of 50-60℃.

[0071] Days 4-15 (High-Temperature Fermentation Stage): The entire pile enters the high-temperature stage, with the average temperature maintained at 58-63℃. Monitoring revealed that the oxygen concentration at the bottom layer dropped to 3%, below the normal threshold of 5%, and it was identified as an oxygen-deficient area. Localized enhanced ventilation was implemented at the bottom layer, increasing the ventilation rate from 0.15 m³ / (min·m³) to 0.25 m³ / (min·m³). After 3 hours, the oxygen concentration at the bottom layer recovered to 8%, returning to normal. On day 7, the humidity in some edge monitoring units dropped to 42%, below the normal threshold of 45%. Localized water was added to these units, with 1-2 liters of water added to each unit, restoring the humidity to 50-55%. On day 10, a second full pile turning was carried out to promote the degradation of organic matter and the redistribution of heat.

[0072] Days 16-30 (Cooling and Maturation Stage): The pile temperature gradually decreases to 40-50℃, and the degradation rate of organic matter slows down; monitoring shows that the pH value of each layer gradually increases from 6.5 to 7.5-8.0, indicating that the compost is stabilizing; the prediction model assesses that the maturity index continues to rise, and it is expected that the maturity standard (maturity index > 0.7) will be reached on day 28; the entire pile is turned over once on day 20 and day 25, and the ventilation rate is gradually reduced to 0.08 m³ / (min·m³); composting ends on day 30, the pile temperature drops to ambient temperature +5℃, and the maturity index reaches 0.75.

[0073] Product quality testing: Compost product test results: Organic matter content: 68%, which is about 15% higher than the traditional method (55-60%); Carbon-nitrogen ratio is reduced to 15:1, which is suitable as a soil conditioner; Maturity index: 0.75, which meets the standard of complete decomposition; Seed germination index: 92%, indicating no plant toxicity; Moisture content: 35%, which is convenient for subsequent packaging and storage.

[0074] Energy consumption and gas emissions: Number of turnings: 5 times (traditional methods usually require 8-10 times); Total energy consumption: reduced by about 28% (mainly savings in turning and ventilation energy consumption); Total ammonia emissions: reduced by about 35% compared to traditional methods; Composting cycle: 30 days (traditional methods usually require 40-45 days), shortened by about 33%.

[0075] Model Update: After composting is completed, all monitoring data, control decisions, actual operations, and product quality parameters for this batch will be saved. Evaluation of the prediction model's performance in this batch: The average error in temperature prediction was 3.2℃, and the error in maturity prediction was 5.8%, both within acceptable ranges; therefore, a model update will not be triggered at this time.

[0076] Example 3, Material Preparation: The composition of this batch of materials differs from that of Example 1: 40% pruned branches, crushed to 2-4cm; 40% lawn mow; 15% fallen leaves; and 5% chicken manure added as a nitrogen source regulator. The initial carbon-to-nitrogen ratio is 22:1, the moisture content is 68%, and the organic matter content is 78%. The pile size is the same.

[0077] Differences in the composting process: Due to the low initial carbon-nitrogen ratio and high moisture content, this batch of compost exhibited different characteristics: = faster heating, with the central area temperature reaching 55℃ within 12 hours; monitoring on the second day revealed that the humidity of multiple upper monitoring units exceeded 72%, and the ammonia concentration reached 30ppm, exceeding the preset threshold of 25ppm; the predictive model judged that there was a risk of excessive humidity and ammonia emission.

[0078] The multi-objective optimization algorithm suggests: immediately turn the entire pile over, increase the ventilation rate to 0.20 m³ / (min·m³) after turning, and add a dry straw covering layer to the surface of the pile to a thickness of 15 cm to promote moisture evaporation and ammonia adsorption; after taking these measures, the humidity should drop to 60-65% within 6 hours and the ammonia concentration should drop to below 15 ppm.

[0079] Model prediction and validation: On day 5, the prediction model predicted that an anaerobic zone would appear at the bottom layer (oxygen concentration would drop to 2%) in the next 48 hours.

[0080] Preemptive measures were taken: ventilation at the bottom was increased and localized turning of the compost was carried out; actual monitoring showed that the prediction was accurate and timely control measures prevented the formation of anaerobic conditions.

[0081] Composting Results: Composting period: 28 days (2 days shorter than Example 1, due to a more suitable initial carbon-to-nitrogen ratio for microbial activity). Excellent product quality, with an organic matter content of 70% and a maturity index of 0.78. Energy consumption was reduced by 25%, and ammonia emissions were reduced by 40%.

[0082] Model Update: The average temperature prediction error for this batch was 4.8℃, with errors reaching 7℃ in some periods, exceeding the preset threshold of 5℃, triggering a model update. The data from the first two batches were merged into a training set, and the prediction model was fine-tuned. The updated model performed better in the third batch of compost, with the temperature prediction error reduced to 2.5℃.

Claims

1. A method for recycling garden greening waste, characterized in that, Step S1, three-dimensional gridding monitoring and data acquisition: the compost pile is divided into layers and zones to form multiple monitoring units, temperature sensors and humidity sensors are arranged in the monitoring units, oxygen concentration sensors and pH sensors are arranged at multiple sampling points in each layer, and multi-point monitoring data inside the pile are acquired at a preset acquisition frequency; Step S2, multi-dimensional feature data set construction: spatial distribution features, statistical features and gradient features are extracted based on the multi-point monitoring data, and initial ingredient parameters of the compost material and time series variation features of the monitoring parameters within a preset time period are combined to form a multi-dimensional feature data set; Step S3, spatial distribution anomaly identification: the multi-dimensional feature data set is compared with the normal threshold interval of each monitoring parameter to identify abnormal areas of temperature, humidity, oxygen concentration and interlayer gradient; Step S4, control decision parameter generation: the multi-dimensional feature data set is input into a compost state prediction model to obtain temperature, maturity and risk prediction results within a target prediction time period, and the abnormal area information is input into a multi-objective optimization algorithm to obtain control decision parameters of turning frequency, ventilation volume and water addition amount; Step S5, hierarchical and zonal compost regulation and control: based on the control decision parameters and abnormal area types, local turning, local enhanced ventilation or local water addition are performed on local abnormal areas, and whole-pile turning or whole-pile forced ventilation are performed on overall abnormal trends, and the execution interval between each regulation and control operation is not less than a preset minimum interval time; Step S6, compost state prediction model updating: the multi-dimensional feature data set, control decision parameters, regulation and control operations and product quality parameters in multiple batches of composting processes are recorded, and when the prediction error between the prediction model output and the actual monitoring exceeds a preset threshold, the compost state prediction model is updated using the recorded data as training samples.

2. The method according to claim 1, characterized in that, The three-dimensional gridding monitoring includes: dividing the compost pile height direction into several vertical layers, and dividing each layer into a rectangular monitoring unit in the horizontal direction; temperature sensors and humidity sensors are arranged inside each monitoring unit, oxygen concentration sensors and pH sensors are arranged at multiple sampling points along each layer, and each type of sensor is data-connected with a central control unit, and the collection time interval of the monitoring parameters is several minutes to several tens of minutes.

3. The method according to claim 1, characterized in that, The multi-dimensional feature data set at least includes: Spatial distribution features: temperature, humidity, oxygen concentration and pH value of each monitoring unit at the same time; Statistical features: mean value and dispersion index of monitoring parameters of each layer; Gradient features: parameter difference between adjacent layers and adjacent monitoring units in the same layer; Ingredient features: carbon-nitrogen ratio, moisture content, organic matter content and particle size distribution of the compost material; Time series features: change trend and change rate of the monitoring parameters within at least a 24-hour time window.

4. The method according to claim 1, characterized in that, The composting state prediction model is a multi-output prediction model based on deep learning, taking the multi-dimensional feature data set as input, and simultaneously outputting temperature distribution, maturity index and abnormal risk index in the target prediction period.

5. The garden greening waste recycling method according to claim 1, characterized in that, The optimization objectives of the multi-objective optimization algorithm include at least shortening the composting period, improving the quality of composting products, reducing energy consumption, and reducing gas emissions; and the constraint conditions include at least turning frequency, ventilation volume, water addition amount, and maximum temperature of the composting pile not exceeding the safety upper limit.

6. The garden greening waste recycling method according to claim 2, characterized in that, When the height of the composting pile is between about 1.5 m and 3.0 m, the pile is divided into 3 to 6 vertical layers, each layer being about 30 cm to 50 cm thick; each layer is divided into rectangular monitoring units with a side length of about 40 cm to 60 cm in the horizontal direction; temperature sensors and humidity sensors are arranged at the geometric center of each monitoring unit; oxygen concentration sensors and pH sensors are arranged at not less than 3 sampling points in each layer; the temperature and humidity collection interval is about 10 min to 30 min; and the oxygen concentration and pH collection interval is about 20 min to 60 min.

7. The garden greening waste recycling method according to claim 3, characterized in that, The statistical features include the mean, standard deviation and range of the monitoring parameters of each layer, and the gradient features include the vertical temperature gradient, humidity gradient, oxygen concentration gradient and pH gradient; in the abnormality identification, when the temperature of a monitoring unit is outside the normal interval for the corresponding composting stage for not less than 8 hours, or the absolute value of the temperature gradient of adjacent layers exceeds the preset threshold, the corresponding area is marked as a temperature abnormal area or a gradient abnormal area.

8. The garden greening waste recycling method according to claim 4, characterized in that, The composting state prediction model includes a one-dimensional convolutional network layer for extracting time series features, a long short-term memory network layer for modeling long-term dependencies, and an attention mechanism layer for assigning weights to different features; the output layer has three branches for predicting the temperature distribution, maturity index and abnormal risk probability in the next 24 hours, 48 hours and 72 hours, respectively.

9. The garden greening waste recycling method according to claim 5, characterized in that, The multi-objective optimization algorithm uses a non-dominated sorting genetic algorithm, with a population size of several tens to several hundreds of individuals and an evolution number of several tens to several hundreds of generations; the Pareto optimal solution set is obtained, and the control decision parameters of turning frequency, ventilation volume and water addition amount are selected from the Pareto optimal solution set according to the current composting conditions.

10. A landscaping waste recycling system applied to the landscaping waste recycling method of claim 1, characterized in that, It includes: A three-dimensional gridding monitoring module: responsible for layering and zoning the composting pile, and arranging temperature, humidity, oxygen concentration and pH sensors in each monitoring unit; A multi-dimensional feature data set construction module: based on multi-point monitoring data, the module extracts spatial distribution features, statistical features, gradient features, composition features and time series features during composting to form a multi-dimensional feature data set; Space distribution anomaly identification module: Based on the multi-dimensional feature dataset, this module identifies abnormal areas in the composting process by comparing with the normal threshold interval, including temperature, humidity, oxygen concentration and other aspects of abnormal areas, and marks them out for subsequent processing; Composting state prediction module: Based on deep learning technology, the composting process is predicted, and the prediction results of temperature distribution, maturity and abnormal risk in the future time period are output; Multi-objective optimization module: Through multi-objective optimization algorithm, the control decision parameters in the composting process are solved while considering multiple optimization objectives; Composting control execution module: According to the control decision parameters, this module executes specific control operations, and adjusts the control according to the composting state and abnormal area type; Composting state prediction model update module: In each batch of composting process, record the multi-dimensional feature dataset, control decision parameters, control operation and product quality information, when the prediction error between the output of the composting state prediction model and the actual monitoring data exceeds the preset threshold, update the composting state prediction model by using these recorded data.

Images