A lightweight, multi-particle cleaning method for construction waste disposal data

Abstract

This invention designs a lightweight, multi-granular cleaning method for construction waste disposal data, belonging to the field of construction waste disposal process and intelligent cleaning technology. It achieves precise cleaning of key product quality indicators such as belt scale flow rate and velocity during construction waste disposal. Construction waste disposal data is the foundation for subsequent waste recycling and treatment decisions. Low-quality data may lead to misjudgments of equipment status, production line load, and material throughput capacity, thereby affecting the setting of production line control parameters and operational decisions. Therefore, cleaning construction waste disposal data is of great significance for improving the accuracy of subsequent analysis. This invention constructs a lightweight, multi-granular cleaning model consisting of an input layer, a multi-granularity decomposition module, and a granularity sensing module to adapt to on-site online cleaning requirements.

Description

Technical Field

[0001] This invention presents a lightweight, multi-granular cleaning method for construction waste disposal data, achieving precise cleaning of key product quality indicators such as belt scale flow rate and velocity during construction waste disposal. Construction waste disposal data forms the basis for subsequent decisions regarding waste recycling and treatment. Low-quality data can lead to misjudgments of equipment status, production line load, and material throughput capacity, thereby affecting production line control parameter settings and operational decisions. Therefore, cleaning construction waste disposal data is crucial for improving the accuracy of subsequent analysis and belongs to the field of construction waste disposal process and intelligent cleaning technology. Background Technology

[0002] In the resource recovery and reduction of construction waste, the optimization of production line operation, energy consumption control, equipment health management, and output and quality assessment increasingly rely on the accuracy and continuity of process monitoring data. In actual engineering projects, the construction waste disposal process typically uses data collection devices such as belt scales to continuously collect data on material flow rate, flow rate, and other indicators of multiple key pieces of equipment, forming a multivariate, strongly time-correlated data stream. This provides a basis for subsequent operating condition identification, production line scheduling, and control decisions.

[0003] In complex construction waste disposal environments, sensors and data acquisition links are prone to drift, noise interference, transient anomalies, packet loss, and missing data, resulting in incomplete or distorted raw monitoring data. Improper handling of abnormal / missing data can lead to misjudgments of equipment status, production line load, and material throughput capacity, thereby affecting production line control parameter settings and operational decisions, reducing disposal efficiency, and increasing maintenance costs. Therefore, high-quality data cleaning for construction waste disposal processes is a crucial foundation for ensuring the reliability of subsequent analysis, prediction, and control.

[0004] Existing data cleaning methods often employ traditional techniques such as threshold rules, sliding smoothing, and interpolation completion, which typically struggle to simultaneously address the coupling relationships between variables from multiple devices and their changing patterns across different time scales. While some deep learning-based modeling methods can improve representation capabilities, their complex models and high computational costs make them unsuitable for deployment and real-time operation at the edge of the disposal site or in resource-constrained online systems. To address these issues, a method is needed that can fully characterize the dynamic features of multivariate time-series data from construction waste disposal at different granularities with low computational cost, and effectively clean data to remove anomalies, missing data, and noise. Therefore, this invention constructs a lightweight multi-granularity cleaning model consisting of an input layer, a multi-granularity decomposition module, and a granularity perception module to adapt to on-site online cleaning requirements. Summary of the Invention

[0005] The technical problem that this invention needs and can solve.

[0006] (1) The problem of unstable cleaning effect caused by multi-device and multi-scale fluctuations.

[0007] In construction waste disposal production lines, the flow rates and volumes of key equipment such as jaw crushers, hammer crushers, drum screens, vibrating screens, magnetic separators, and air separators are continuously collected in a time-series format, forming process data containing 12 input variables. Due to the long equipment chain and numerous links, there are obvious dynamic transmission characteristics between the variables, and the superposition of changes in the operating conditions of different equipment can generate multi-scale fluctuations. This makes it difficult for cleaning methods that rely solely on single-variable or single-scale processing to simultaneously ensure global consistency and local details, easily leading to unstable cleaning results as the operating conditions change. This invention constructs a multi-granularity decomposition module to achieve multi-granularity feature extraction and modeling of input data, fundamentally solving the problem of unstable cleaning effects caused by multi-scale fluctuations from multiple equipment, and achieving high-precision cleaning of construction waste disposal data.

[0008] A lightweight, multi-granular cleaning method for construction waste disposal data, characterized by: collecting construction waste disposal data, establishing a lightweight, multi-granular cleaning model for construction waste disposal data, training the lightweight, multi-granular cleaning model for construction waste disposal data, and cleaning the construction waste disposal data, including the following steps:

[0009] (1) Collect data on construction waste disposal

[0010] Taking the construction waste disposal process as the research object, the flow velocity and flow rate data of the jaw crusher, hammer crusher, drum screen, vibrating screen, magnetic separator, and air separator were collected by belt scale. The flow velocity x of the output material of the jaw crusher at time t was also collected. 1 (t), in meters per second, represents the flow rate x of the material output from the jaw crusher at time t. 2 (t), in kilograms per second, represents the velocity (x) of the material output from the hammer crusher at time t. 3 (t), in meters per second, represents the flow rate x of the material output from the hammer crusher at time t. 4 (t), in kilograms per second, represents the velocity x of the material output from the drum screen at time t. 5 (t), in meters per second, represents the flow rate x of the material output from the drum screen at time t. 6 (t), in kilograms per second, represents the velocity x of the material output from the vibrating screen at time t. 7 (t), in meters per second, represents the flow rate x of the material output from the vibrating screen at time t. 8 (t), in kilograms per second, represents the velocity x of the material output from the magnetic separator at time t. 9 (t), in meters per second, represents the flow rate x of the material output from the magnetic separator at time t. 10 (t), in kilograms per second, represents the velocity x of the material output from the air classifier at time t.11 (t), in meters per second, represents the flow rate x of the material output from the air classifier at time t. 12 (t), in kilograms per second; select x 1 (t), x 2 (t), x 3 (t), x 4 (t), x 5 (t), x 6 (t), x 7 (t), x 8 (t), x 9 (t), x 10 (t), x 11 (t) and x 12 (t) serves as the input variable for the lightweight multi-granularity cleaning model of construction waste disposal data at time t, and its value at the next time step serves as the output variable of the model; t represents the current time step;

[0011] (2) Establish a lightweight, multi-granular cleaning model for construction waste disposal data.

[0012] A lightweight multi-granularity cleaning model is constructed, which consists of an input layer, a multi-granularity decomposition module, and a granularity perception module.

[0013] Input layer: The input variable vector is constructed by arranging the collected data at time t and the previous B-1 historical time steps in chronological order, X. m (t) represents the m-th input variable vector, X m (t)=[x m (t-B+1),…,x m (t)],x m (t-B+1) represents the value of the m-th input variable at time step t-B+1, where B is the length of the input variable's time window, and m represents the index of the input variable, m=1,2,…,12; the output matrix of the input layer is X(t)=[X 1 (t) T ,…,X n (t) T ,…,X 12 (t) T ] T T denotes the transpose of the matrix;

[0014] Multi-granularity decomposition module: includes adaptive convolution methods and lightweight residual gating methods; calculates the output of the adaptive convolution method:

[0015]

[0016] in This represents the adaptive convolution feature vector at time t. Indicates the first Data at any given time This represents the value of the l-th element of the convolution kernel; in the lightweight residual gating module, the output of the lightweight residual gating method is calculated:

[0017]

[0018]

[0019] Where W1 and W2 represent learnable weights, It is the ReLU activation function. The gate vector representing time. This represents the output vector of the lightweight residual gating module. and Indicates learnable weights, Represents the Sigmoid function;

[0020] Output of the computational granularity awareness module:

[0021]

[0022]

[0023] in This represents the feature vector after granularity sensing at time t. This represents the result of data cleaning at time t. Represents a learnable vector. Indicates learnable weights, This indicates a splicing operation. Indicates a fully connected state;

[0024] (3) Training a lightweight, multi-granular cleaning model for construction waste disposal data

[0025] ① Define the loss function of the model as:

[0026]

[0027] in, Indicates the first Loss in lightweight, multi-granular cleaning models for construction waste disposal data. This represents the data vector of the construction waste disposal process at the next time step after the cleaning model at time t-1. It is obtained from the lightweight multi-granularity cleaning model of construction waste disposal data and includes an input layer, a multi-granularity decomposition module, a lightweight residual gating method, and a granularity-aware module, for a total of 2 layers. This represents the data vector of the actual construction waste disposal process at time t-1, which is used as a monitoring signal at time t to participate in the training of the cleaning model.

[0028] ② Set the current training time to t and initialize the number of training rounds. =1, the number of training iterations is fixed at 100; initialize the model's weight parameters and bias parameters, the weight parameters are randomly selected in the interval [−0.2,0.2], and the bias parameters are set to 0;

[0029] ③ Update the weight matrix using gradient descent. The calculation formula is as follows:

[0030]

[0031]

[0032]

[0033]

[0034] in and Indicates the first In the next iteration, the weight matrix in the lightweight residual gating module and Indicates the first In the next iteration, the weight matrix in the lightweight residual gating module Indicates the first In the next iteration, the learnable vectors in the granularity-aware module Indicates the first In the next iteration, the learnable vectors in the granularity-aware module Indicates the first In the next iteration, the weight matrix in the granularity-aware module Indicates the first In the next iteration, the weight matrix in the granularity-aware module Index representing the training round number, =1,2,…,100;

[0035] ④ If the number of training rounds <100, Increase by 1, proceed to step ③ and continue training; if the number of training rounds... If the value is ≥100, then terminate the training and update of the model parameters and proceed to step (4) to perform cleaning.

[0036] (4) Cleaning construction waste disposal data

[0037] Using a trained lightweight multi-particle cleaning model based on construction waste disposal data, the flow rates of the jaw crusher, hammer crusher, drum screen, vibrating screen, magnetic separator, air separator, jaw crusher flow rate, hammer crusher flow rate, drum screen flow rate, vibrating screen flow rate, magnetic separator flow rate, and air separator flow rate collected at time t are used as the model inputs. The model output vector at time t is obtained according to formulas (1)-(5). , to output vectors from the model This serves as the data for construction waste disposal at the next moment after cleaning at time t.

[0038] The effects that this invention can achieve:

[0039] (1) Social effects

[0040] This invention constructs a lightweight, multi-granular data cleaning model for construction waste disposal, enabling precise correction and intelligent optimization of operational data in key processes such as crushing, screening, magnetic separation, and air separation. This improves the accuracy and stability of core data like flow rate and volume, reduces manual intervention and reliance on experience, and enhances the operational reliability and management efficiency of construction waste disposal systems. This method helps strengthen digital supervision throughout the entire process, reduces resource waste and operational risks, promotes the reduction, recycling, and intelligent development of construction waste, and provides strong support for green and low-carbon transformation and the construction of "zero-waste cities."

[0041] (2) Economic effects

[0042] This invention improves the accuracy and stability of construction waste disposal data cleaning, enhances the reliability of equipment operation data and the scientific nature of decision analysis, reduces misjudgments and resource waste caused by data anomalies, and minimizes unreasonable energy consumption expenditures such as equipment idling and overload operation, thus effectively controlling the operating costs of the disposal process. Simultaneously, this method reduces manual data screening and repetitive verification work, lowering the costs of information management and maintenance. With the widespread application of this technology, it will help improve the overall operational profitability of construction waste resource utilization projects, accelerating the industry's cost reduction, efficiency improvement, and large-scale, intelligent development.

[0043] (3) Technical effects

[0044] This invention technically overcomes the limitations of traditional data cleaning methods in adapting to multi-source heterogeneous time-series data and inaccurate anomaly identification. By constructing a lightweight multi-granularity decomposition module and a granularity sensing module, combined with adaptive convolution and a lightweight residual gating mechanism, it accurately extracts the temporal correlation features and dynamic change patterns of variables such as flow rate and volume during construction waste disposal. This effectively suppresses the impact of noise interference and abnormal fluctuations on the model output, significantly improving the accuracy and stability of the data cleaning results. The method is lightweight, computationally efficient, and can seamlessly integrate with existing construction waste disposal monitoring and control systems, providing reliable technical support for intelligent sensing and refined management of the disposal process. Attached Figure Description

[0045] Figure 1 This is a diagram showing the results of data cleaning according to the present invention; Detailed Implementation

[0046] The present invention adopts the following technical solution and implementation steps:

[0047] A lightweight, multi-granular cleaning method for construction waste disposal data, characterized by: collecting construction waste disposal data, establishing a lightweight, multi-granular cleaning model for construction waste disposal data, training the lightweight, multi-granular cleaning model for construction waste disposal data, and cleaning the construction waste disposal data, including the following steps:

[0048] (1) Collect data on construction waste disposal

[0049] Taking the construction waste disposal process as the research object, the flow velocity and flow rate data of the jaw crusher, hammer crusher, drum screen, vibrating screen, magnetic separator, and air separator were collected by belt scale. The flow velocity x of the output material of the jaw crusher at time t was also collected. 1 (t), in meters per second, represents the flow rate x of the material output from the jaw crusher at time t. 2 (t), in kilograms per second, represents the velocity (x) of the material output from the hammer crusher at time t. 3 (t), in meters per second, represents the flow rate x of the material output from the hammer crusher at time t. 4 (t), in kilograms per second, represents the velocity x of the material output from the drum screen at time t. 5 (t), in meters per second, represents the flow rate x of the material output from the drum screen at time t. 6 (t), in kilograms per second, represents the velocity x of the material output from the vibrating screen at time t. 7 (t), in meters per second, represents the flow rate x of the material output from the vibrating screen at time t. 8 (t), in kilograms per second, represents the velocity x of the material output from the magnetic separator at time t. 9(t), in meters per second, represents the flow rate x of the material output from the magnetic separator at time t. 10 (t), in kilograms per second, represents the velocity x of the material output from the air classifier at time t. 11 (t), in meters per second, represents the flow rate x of the material output from the air classifier at time t. 12 (t), in kilograms per second; select x 1 (t), x 2 (t), x 3 (t), x 4 (t), x 5 (t), x 6 (t), x 7 (t), x 8 (t), x 9 (t), x 10 (t), x 11 (t) and x 12 (t) serves as the input variable for the lightweight multi-granularity cleaning model of construction waste disposal data at time t, and its value at the next time step serves as the output variable of the model; t represents the current time step;

[0050] (2) Establish a lightweight, multi-granular cleaning model for construction waste disposal data.

[0051] A lightweight multi-granularity cleaning model is constructed, which consists of an input layer, a multi-granularity decomposition module, and a granularity perception module.

[0052] Input layer: The input variable vector is constructed by arranging the collected data at time t and the previous B-1 historical time steps in chronological order, X. m (t) represents the m-th input variable vector, X m (t)=[x m (t-B+1),…,x m (t)],x m (t-B+1) represents the value of the m-th input variable at time step t-B+1, where B is the length of the input variable's time window, and m represents the index of the input variable, m=1,2,…,12; the output matrix of the input layer is X(t)=[X 1 (t) T ,…,X n (t) T ,…,X 12 (t) T ] T T denotes the transpose of the matrix;

[0053] Multi-granularity decomposition module: includes adaptive convolution methods and lightweight residual gating methods; calculates the output of the adaptive convolution method:

[0054]

[0055] in This represents the adaptive convolution feature vector at time t. Indicates the first Data at any given time This represents the value of the l-th element of the convolution kernel; in the lightweight residual gating module, the output of the lightweight residual gating method is calculated:

[0056]

[0057]

[0058] Where W1 and W2 represent learnable weights, It is the ReLU activation function. The gate vector representing time. This represents the output vector of the lightweight residual gating module. and Indicates learnable weights, Represents the Sigmoid function;

[0059] Output of the computational granularity awareness module:

[0060]

[0061]

[0062] in This represents the feature vector after granularity sensing at time t. This represents the result of data cleaning at time t. Represents a learnable vector. Indicates learnable weights, This indicates a splicing operation. Indicates a fully connected state;

[0063] (3) Training a lightweight, multi-granular cleaning model for construction waste disposal data

[0064] ① Define the loss function of the model as:

[0065]

[0066] in, Indicates the first Loss in lightweight, multi-granular cleaning models for construction waste disposal data. This represents the data vector of the construction waste disposal process at the next time step after the cleaning model at time t-1. It is obtained from the lightweight multi-granularity cleaning model of construction waste disposal data and includes an input layer, a multi-granularity decomposition module, a lightweight residual gating method, and a granularity-aware module, for a total of 2 layers. This represents the data vector of the actual construction waste disposal process at time t-1, which is used as a monitoring signal at time t to participate in the training of the cleaning model.

[0067] ② Set the current training time to t and initialize the number of training rounds. =1, the number of training iterations is fixed at 100; initialize the model's weight parameters and bias parameters, the weight parameters are randomly selected in the interval [−0.2,0.2], and the bias parameters are set to 0;

[0068] ③ Update the weight matrix using gradient descent. The calculation formula is as follows:

[0069]

[0070]

[0071]

[0072]

[0073] in and Indicates the first In the next iteration, the weight matrix in the lightweight residual gating module and Indicates the first In the next iteration, the weight matrix in the lightweight residual gating module Indicates the first In the next iteration, the learnable vectors in the granularity-aware module Indicates the first In the next iteration, the learnable vectors in the granularity-aware module Indicates the first In the next iteration, the weight matrix in the granularity-aware module Indicates the first In the next iteration, the weight matrix in the granularity-aware module Index representing the training round number, =1,2,…,100;

[0074] ④ If the number of training rounds <100, Increase by 1, proceed to step ③ and continue training; if the number of training rounds... If the value is ≥100, then terminate the training and update of the model parameters and proceed to step (4) to perform cleaning.

[0075] (4) Cleaning construction waste disposal data

[0076] Using a trained lightweight multi-particle cleaning model based on construction waste disposal data, the flow rates of the jaw crusher, hammer crusher, drum screen, vibrating screen, magnetic separator, air separator, jaw crusher flow rate, hammer crusher flow rate, drum screen flow rate, vibrating screen flow rate, magnetic separator flow rate, and air separator flow rate collected at time t are used as the model inputs. The model output vector at time t is obtained according to formulas (1)-(5). , to output vectors from the model This serves as the data for construction waste disposal at the next moment after cleaning at time t.

Claims

1. A lightweight, multi-granular cleaning method for construction waste disposal data, characterized in that, Collect construction waste disposal data, establish a lightweight, multi-granular data cleaning model for construction waste disposal, train the lightweight, multi-granular data cleaning model for construction waste disposal, and clean the construction waste disposal data, including the following steps: (1) Collect data on construction waste disposal Taking the construction waste disposal process as the research object, the flow velocity and flow rate data of the jaw crusher, hammer crusher, drum screen, vibrating screen, magnetic separator, and air separator were collected by belt scale. The flow velocity x of the output material of the jaw crusher at time t was also collected. 1 (t), in meters per second, represents the flow rate x of the material output from the jaw crusher at time t. 2 (t), in kilograms per second, represents the velocity (x) of the material output from the hammer crusher at time t. 3 (t), in meters per second, represents the flow rate x of the material output from the hammer crusher at time t. 4 (t), in kilograms per second, represents the velocity x of the material output from the drum screen at time t. 5 (t), in meters per second, represents the flow rate x of the material output from the drum screen at time t. 6 (t), in kilograms per second, represents the velocity x of the material output from the vibrating screen at time t. 7 (t), in meters per second, represents the flow rate x of the material output from the vibrating screen at time t. 8 (t), in kilograms per second, represents the velocity x of the material output from the magnetic separator at time t. 9 (t), in meters per second, represents the flow rate x of the material output from the magnetic separator at time t. 10 (t), in kilograms per second, represents the velocity x of the material output from the air classifier at time t. 11 (t), in meters per second, represents the flow rate x of the material output from the air classifier at time t. 12 (t), in kilograms per second; select x 1 (t), x 2 (t), x 3 (t), x 4 (t), x 5 (t), x 6 (t), x 7 (t), x 8 (t), x 9 (t), x 10 (t), x 11 (t) and x 12 (t) serves as the input variable for the lightweight multi-granularity cleaning model of construction waste disposal data at time t, and its value at the next time step serves as the output variable of the model; t represents the current time step; (2) Establish a lightweight, multi-granular cleaning model for construction waste disposal data. A lightweight multi-granularity cleaning model is constructed, which consists of an input layer, a multi-granularity decomposition module, and a granularity perception module. Input layer: The input variable vector is constructed by arranging the collected data at time t and the previous B-1 historical time steps in chronological order, X. m (t) represents the m-th input variable vector, X m (t)=[x m (t-B+1),…,x m (t)],x m (t-B+1) represents the value of the m-th input variable at time step t-B+1, where B is the length of the input variable's time window, and m represents the index of the input variable, m=1,2,…,12; the output matrix of the input layer is X(t)=[X 1 (t) T ,…,X n (t) T ,…,X 12 (t) T ] T T denotes the transpose of the matrix; Multi-granularity decomposition module: includes adaptive convolution methods and lightweight residual gating methods; calculates the output of the adaptive convolution method: ； in This represents the adaptive convolution feature vector at time t. Indicates the first Data at any given time This represents the value of the l-th element of the convolution kernel; in the lightweight residual gating module, the output of the lightweight residual gating method is calculated: ； ； Where W1 and W2 represent learnable weights, It is the ReLU activation function. The gate vector representing time. This represents the output vector of the lightweight residual gating module. and Indicates learnable weights, Represents the Sigmoid function; Output of the computational granularity awareness module: ； ； in This represents the feature vector after granularity sensing at time t. This represents the result of data cleaning at time t. Represents a learnable vector. Indicates learnable weights, This indicates a splicing operation. Indicates a fully connected state; (3) Training a lightweight, multi-granular cleaning model for construction waste disposal data ① Define the loss function of the model as: ； in, Indicates the first Loss in lightweight, multi-granular cleaning models for construction waste disposal data. This represents the data vector of the construction waste disposal process at the next time step after the cleaning model at time t-1. It is obtained from the lightweight multi-granularity cleaning model of construction waste disposal data and includes an input layer, a multi-granularity decomposition module, a lightweight residual gating method, and a granularity-aware module, for a total of 2 layers. This represents the data vector of the actual construction waste disposal process at time t-1, which is used as a monitoring signal at time t to participate in the training of the cleaning model. ② Set the current training time to t and initialize the number of training rounds. =1, the number of training iterations is fixed at 100; initialize the model's weight parameters and bias parameters, the weight parameters are randomly selected in the interval [−0.2,0.2], and the bias parameters are set to 0; ③ Update the weight matrix using gradient descent. The calculation formula is as follows: ； ； ； ； in and Indicates the first In the next iteration, the weight matrix in the lightweight residual gating module and Indicates the first In the next iteration, the weight matrix in the lightweight residual gating module Indicates the first In the next iteration, the learnable vectors in the granularity-aware module Indicates the first In the next iteration, the learnable vectors in the granularity-aware module Indicates the first In the next iteration, the weight matrix in the granularity-aware module Indicates the first In the next iteration, the weight matrix in the granularity-aware module Index representing the training round number, =1,2,…,100; ④ If the number of training rounds <100, Increase by 1, proceed to step ③ and continue training; if the number of training rounds... If the value is ≥100, then terminate the training and update of the model parameters and proceed to step (4) to perform cleaning. (4) Cleaning construction waste disposal data Using a trained lightweight multi-particle cleaning model based on construction waste disposal data, the flow rates of the jaw crusher, hammer crusher, drum screen, vibrating screen, magnetic separator, air separator, jaw crusher flow rate, hammer crusher flow rate, drum screen flow rate, vibrating screen flow rate, magnetic separator flow rate, and air separator flow rate collected at time t are used as the model inputs. The model output vector at time t is obtained according to formulas (1)-(5). , to output vectors from the model This serves as the data for construction waste disposal at the next moment after cleaning at time t.

Images