Multi-parameter decision deep networking muck resource utilization device

Abstract

The invention belongs to the technical field of tunnel engineering muck treatment, and particularly relates to a multi-parameter decision deep networked muck resource utilization device, which comprises a sensor module, a key parameter central processing unit, a data acquisition and transmission module, a data acquisition and transmission module and a resource utilization module, the multi-parameter decision-making module is used for operating a deep learning model and a multi-objective optimization algorithm, realizing multi-parameter decision-making and generating optimal process parameters; and the execution module is connected with the central processing unit and each processing device, and is used for receiving the control instruction and adjusting the operation parameters of the device. According to the multi-parameter decision deep networked muck resource utilization device, a sensing-decision-execution closed-loop control system is constructed by integrating the sensor module, the central processing unit and the execution module, automation and intelligentization of the whole process from muck feeding to finished product output are achieved, and manual dependence and subjective misjudgment risks are remarkably reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tunnel engineering spoil treatment, and particularly relates to a multi-parameter decision deep networked spoil resource utilization device. BACKGROUND

[0002] At present, shield spoil treatment and utilization face the following main problems:

[0003] 1. The composition of shield spoil is complex and unstable, and there is a lack of efficient intelligent monitoring means. The composition of shield spoil has high complexity and instability, especially under different geological conditions. Traditional detection means cannot monitor key indicators such as spoil discharge amount and water content in real time, resulting in unsatisfactory pretreatment effect.

[0004] 2. The screening efficiency of spoil of the same particle size and different components is low, which affects the resource utilization effect. Shield spoil often contains large particle size materials such as pebbles and mudstone. Although they have similar particle sizes, there are significant differences in recycling value. The efficiency of traditional screening technology is low, and materials with high utilization value cannot be effectively separated, affecting the resource effect of different components in the spoil.

[0005] 3. There is a lack of real-time detection means in the spoil disposal process, which affects the resource utilization efficiency. The content of fine-grained soil directly affects the improvement and resource utilization effect of spoil, but the existing traditional density meter method is low in efficiency and difficult to adapt to the rapidly changing construction environment. The flocculation process is sensitive to PH value, but the existing method relies on manual detection and cannot monitor and optimize in real time, which is difficult to adapt to the dynamic change of PH value. The filter pressing effect directly affects the resource utilization efficiency of spoil, but the existing method relies on manual detection and cannot monitor in real time, which is difficult to adapt to the rapidly changing construction environment.

[0006] Therefore, the present application proposes a multi-parameter decision deep networked spoil resource utilization device for improving the resource utilization efficiency of spoil. SUMMARY

[0007] Based on the above technical problems, the present application proposes a multi-parameter decision deep networked spoil resource utilization device.

[0008] The multi-parameter decision deep networked spoil resource utilization device proposed by the present application comprises:

[0009] A spoil inlet receives spoil generated by a shield machine.

[0010] A preliminary screening device is connected to the spoil inlet and is used to separate large stones and impurities.

[0011] A stirring conditioning tank is connected to the discharge port of the preliminary screening device and is used to fully mix the spoil by mechanical stirring to form a slurry with uniform composition.

[0012] A flocculation conditioning tank is connected to the discharge port of the stirring conditioning tank, for adding a flocculating agent to the slurry to cause fine particles to aggregate to form flocs.

[0013] A filter press is connected to the discharge port of the flocculation conditioning tank, for mechanically filter-pressing the flocculated slurry to dewater and form a filter cake.

[0014] A sensor module integrates multiple sensors for real-time monitoring of key parameters in the entire process of the spoil treatment.

[0015] A data acquisition and transmission module is electrically connected to the sensor module, for acquiring sensor data and transmitting the data through a wireless network, while having edge computing capabilities and cloud collaborative communication, and performing local data caching and preliminary processing in the event of network interruption.

[0016] A central processing unit is communicatively connected to the data acquisition and transmission module, for running a deep learning model and a multi-objective optimization algorithm to achieve multi-parameter decision-making and generate optimal process parameters, and running the following algorithms:

[0017] A1. Cleaning, normalizing, and feature extraction of multi-source data.

[0018] A2. Establishing a mapping relationship from multiple parameters to product performance using a deep learning model, which can learn the complex relationship between multiple parameters and the performance of ecological materials and achieve prediction of the performance of ecological materials, for predicting the performance of recycled products and environmental risks.

[0019] A3. Finding optimal process parameters using a multi-objective optimization algorithm.

[0020] A4. Converting the optimized process parameters into control instructions.

[0021] An execution module is connected to the central processing unit and each processing device, and builds a distributed control network, for receiving control instructions and adjusting device operating parameters.

[0022] A human-machine interface is connected to the central processing unit, for displaying real-time data, prediction results, device states, and alarm information, and supporting a manual override mode of process parameters, allowing an operator to manually input process parameters when the automatic decision-making system is being maintained or has a significant deviation.

[0023] Preferably, the execution module includes an adjustable vibration mechanism disposed on the surface of the primary screening equipment, a flocculant addition mechanism disposed on the surface of the flocculation conditioning tank, and a stirring mechanism disposed on the surface of the stirring conditioning tank. The adjustable vibration mechanism is used to control the screening efficiency and particle size, the flocculant addition mechanism is used to precisely control the amount of flocculant added, and the stirring mechanism is used to adjust the stirring speed and stirring time.

[0024] Preferably, the sensor module includes:

[0025] A 3D laser scanner is installed on the feeding conveyor belt of the preliminary screening equipment to monitor the discharge volume of the slag in real time.

[0026] A dielectric constant sensor, installed on the side wall of the mixing and conditioning tank, is used to measure the dielectric constant of the slurry and calculate the overall moisture content.

[0027] A machine vision camera, mounted on the discharge conveyor belt of the primary screening equipment and equipped with a stable light source system, is used to identify the coarse aggregate gradation of the slag.

[0028] A laser particle size analyzer is installed on the pipeline of the filter press feed pump for online analysis of the particle size distribution of fine particles.

[0029] A heavy metal ion concentration sensor is installed inside the flocculation conditioning tank to detect the concentration of dissolved heavy metal ions in the slurry.

[0030] A pH sensor is installed inside the flocculation conditioning tank to monitor the pH value during the flocculation process.

[0031] A moisture content sensor is installed at the outlet of the filter press to detect the final moisture content of the filter cake.

[0032] Preferably, the deep learning model run by the central processing unit is a multi-task learning network, and its model structure includes:

[0033] Long Short-Term Memory (LSTM) network layers are used to extract temporal features from multi-source sensor data.

[0034] Convolutional neural network layers are used to extract spatial features between multiple parameters.

[0035] The attention mechanism layer is used to focus on key parameters and time steps.

[0036] The multi-task output layer simultaneously outputs the classification of slag and soil types, the prediction of the compressive strength of recycled products, the prediction of the risk of heavy metal leaching, and the prediction of the chloride ion diffusion coefficient.

[0037] Preferably, the loss function of the multi-task learning network is a weighted multi-task loss function, expressed as:

[0038]

[0039] The cross-entropy loss function for the task of classifying waste soil types is calculated as follows:

[0040]

[0041] in, For the sample size, In order to test the soil samples, This represents the total number of waste soil types. In order to measure the type of waste soil, It is a sample In category The real labels on The sample predicted by the model Category The probability of.

[0042] The mean squared error loss function for predicting the compressive strength of recycled products is expressed by the following formula:

[0043]

[0044] in, It is a sample The true compressive strength value, It is the compressive strength value predicted by the model.

[0045] The mean squared error loss function represents the task of predicting heavy metal leaching rates, and its calculation formula is as follows:

[0046]

[0047] in, It is a sample The true heavy metal leaching rate It is the leaching rate predicted by the model.

[0048] The log-mean-square error loss function for the chloride ion diffusion coefficient prediction task is expressed by the following formula:

[0049]

[0050] in, It is a sample The true chloride ion diffusion coefficient, It is the chloride ion diffusion coefficient predicted by the model.

[0051] This is the L2 regularization term applied to all parameters of the deep learning model.

[0052] , , , , These are configurable hyperparameters used to balance the weights of classification task loss, compressive strength loss, leaching rate loss, diffusion coefficient loss, and regularization term in the total loss.

[0053] Preferably, the multi-objective optimization algorithm run by the central processing unit is an improved multi-objective evolutionary algorithm based on NSGA-II, and its objective function includes:

[0054] C1. Maximize the compressive strength of recycled products Its expression is:

[0055]

[0056] in, The compressive strength of the recycled product is predicted by the deep learning model according to claim 4.

[0057] C2. Minimize heavy metal leaching rate Its expression is:

[0058]

[0059] in, The heavy metal leaching rate is predicted by the deep learning model according to claim 4.

[0060] C3. Minimize overall processing cost Its expression is:

[0061]

[0062] in, To comprehensively consider processing costs, Indicates energy consumption cost, Indicates the cost of the medicine. This indicates the cost of disposal.

[0063] It is a vector of decision variables, including process parameters such as stirring speed, flocculant dosage, and filter press pressure.

[0064] Preferably, the adjustable vibration mechanism includes a vibrating screen, which is inclinedly arranged inside the support of the primary screening equipment. A vibration motor is installed at the bottom of the vibrating screen, and the discharge conveyor belt is arranged below the vibrating screen. A receiving box is arranged below one side of the primary screening equipment, and the lower surface of the vibrating screen is connected to the inner surface of the support of the primary screening equipment through a plurality of shock-absorbing springs arranged in a rectangular array.

[0065] Through the above technical solution, the high end of the vibrating screen is the feeding end and the low end is the waste discharge end, so that the inclined vibrating screen uses gravity to assist screening, thereby improving the processing efficiency; the central control unit changes the intensity of screening by adjusting the frequency or amplitude of the vibrating motor, thereby controlling the particle size distribution of the undersize material and meeting the aggregate particle size requirements of different resource utilization pathways; the rectangular array of damping springs effectively absorbs the high-frequency vibration generated by the vibrating motor, prevents the vibration from being transmitted to the support of the primary screening equipment, ensures the stable operation of the entire system, and reduces noise.

[0066] Preferably, the stirring mechanism includes a stirring motor fixedly installed on the upper surface of the stirring and conditioning tank. One end of the output shaft of the stirring motor extends into the stirring and conditioning tank and is fixedly fitted with a stirring shaft paddle. The upper end of the stirring shaft paddle is mounted on the inner top wall of the stirring and conditioning tank through a bearing. The feed inlet of the stirring and conditioning tank is connected to one end of the discharge conveyor belt. The discharge outlet of the stirring and conditioning tank is connected to the feed inlet of the flocculation conditioning tank through a pipe. A suction pump is provided on the surface of the pipe.

[0067] The above technical solution facilitates the reception of screened slag into the mixing and conditioning tank. The rotation of the output shaft of the mixing motor drives the connected mixing shaft paddle to rotate. After uniform mixing, the homogenized slag slurry is pumped from the mixing and conditioning tank to the flocculation conditioning tank by a suction pump. The central control unit adjusts the power of the suction pump to match the processing rhythm of the entire system, thus achieving precise process control.

[0068] Preferably, the flocculant addition mechanism includes a connecting pipe fixedly installed on the upper surface of the flocculation conditioning tank, the upper end of the connecting pipe being fixedly connected to a U-shaped pipe, two branch pipes of the U-shaped pipe being respectively equipped with solenoid valves, and the free end faces of the two branch pipes of the U-shaped pipe being respectively provided with metering pumps, and the central processing unit being electrically connected to the metering pumps and the solenoid valves.

[0069] Through the above technical solution, the solenoid valve is used to control the opening and closing of the U-tube, forming a switchable dual-path dosing system. When one path becomes blocked or malfunctions and requires cleaning or maintenance, the central processing unit automatically controls the solenoid valve to switch to the other path to continue working, ensuring uninterrupted operation of the dosing process and the entire production line. It is particularly suitable for use in engineering environments that require continuous production. At the same time, conventional flocculants can be added in one path, and coagulant aids for adjusting pH value can be added in the other path. The central processing unit can intelligently select the dosing path and reagent ratio according to the real-time pH value of the slag slurry to achieve more complex conditioning processes. Moreover, the metering pump ensures the high accuracy of flocculant dosing and avoids reagent waste.

[0070] Preferably, the central processing unit is pre-loaded with a model trained on soil samples from nine major geological zones across the country, which can be quickly adapted to the geological type of a new project.

[0071] The beneficial effects of this invention are as follows:

[0072] 1. By integrating sensor modules, central processing units and execution modules, a closed-loop control system of "perception-decision-execution" was constructed, realizing full automation and intelligence from slag feeding to finished product output, significantly reducing reliance on manual labor and the risk of subjective misjudgment.

[0073] 2. By adopting a multi-task deep learning model, it can simultaneously process real-time sensor data from multiple sources and heterogeneous structures, accurately predict the type of slag and the performance of recycled products, solve the problem of process adaptation caused by differences in geological conditions, and simultaneously optimize the three major objectives of "compressive strength, leaching rate and processing cost" using a multi-objective optimization algorithm. While ensuring the performance of recycled products, it controls the risk of heavy metal pollution within environmental protection standards, promoting the high-value transformation of slag from "waste" to "resource".

[0074] 3. By setting an adjustable vibration mechanism, efficient and stable particle size control and impurity separation are achieved. At the same time, the use of a U-shaped tube dual-path flocculant addition mechanism ensures the continuity and accuracy of dosing, effectively improving the overall treatment efficiency and reliability. By dynamically adjusting the process parameters of flocculant dosage, stirring speed and filter pressure, the consumption of reagents, energy consumption and landfill costs are effectively reduced. Attached Figure Description

[0075] Figure 1 This is a schematic diagram of a multi-parameter decision-making deep networked waste soil resource utilization device proposed in this invention;

[0076] Figure 2 This is a three-dimensional view of the slag inlet structure of a slag resource utilization device with multi-parameter decision-making deep networking proposed in this invention.

[0077] Figure 3 This is a three-dimensional view of the discharge conveyor belt structure of a multi-parameter decision-making deep networked waste soil resource utilization device proposed in this invention.

[0078] Figure 4 This is a three-dimensional view of the vibrating screen structure of a multi-parameter decision-making deep networked waste soil resource utilization device proposed in this invention.

[0079] Figure 5 This is a three-dimensional view of the mixing shaft and impeller structure of a multi-parameter decision-making deep networked waste soil resource utilization device proposed in this invention.

[0080] Figure 6 This is a three-dimensional view of the flocculation conditioning tank structure of a multi-parameter decision-making deep networked waste soil resource utilization device proposed in this invention.

[0081] Figure 7 This is a system block diagram of a multi-parameter decision-making deep networked waste soil resource utilization device proposed in this invention;

[0082] Figure 8 This is a system overall workflow diagram of a multi-parameter decision-making deep networked waste soil resource utilization device proposed in this invention;

[0083] Figure 9 This is a control flowchart of the central processing unit of a multi-parameter decision-making deep networked waste soil resource utilization device proposed in this invention.

[0084] Figure 10 This is a flowchart illustrating the deep learning model training process for a multi-parameter decision-making deep network-based waste soil resource utilization device proposed in this invention.

[0085] In the diagram: 1. Slag inlet; 2. Preliminary screening equipment; 21. Feeding conveyor belt; 22. Discharge conveyor belt; 3. Mixing and conditioning tank; 4. Flocculation and conditioning tank; 5. Filter press; 51. Feed pump; 6. Vibrating screen; 61. Vibrating motor; 62. Receiving box; 63. Shock-absorbing spring; 7. Mixing motor; 71. Mixing shaft propeller; 72. Suction pump; 8. Connecting pipe; 81. U-shaped pipe; 82. Solenoid valve; 83. Metering pump. Detailed Implementation

[0086] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0087] Reference Figures 1-10 A multi-parameter decision-making deep network-based waste soil resource utilization device, comprising:

[0088] Slag import 1: Receives slag produced by the tunnel boring machine.

[0089] The preliminary screening equipment 2 is connected to the slag inlet 1 and is used to separate large stones and debris.

[0090] The mixing and conditioning tank 3 is connected to the discharge port of the preliminary screening equipment 2 and is used to fully mix the slag and soil through mechanical mixing to form a slurry with uniform composition.

[0091] The flocculation conditioning tank 4 is connected to the outlet of the mixing conditioning tank 3 and is used to add flocculant to the slurry so that fine particles aggregate to form flocs.

[0092] The filter press 5 is connected to the outlet of the flocculation conditioning tank 4 and is used to mechanically filter and dewater the flocculated slurry to form mud cake.

[0093] The sensor module integrates multiple sensors to monitor key parameters throughout the entire process of waste disposal in real time.

[0094] The data acquisition and transmission module is electrically connected to the sensor module. It is used to acquire sensor data and transmit it through a wireless network. It also has edge computing capabilities and can communicate collaboratively with the cloud. When the network is interrupted, it can perform local data caching and preliminary processing.

[0095] The central processing unit (CPU), communicating with the data acquisition and transmission module, is used to run deep learning models and multi-objective optimization algorithms to achieve multi-parameter decision-making and generate optimal process parameters. Simultaneously, the CPU has a pre-installed model trained on slag samples from nine major geological zones across China, enabling rapid adaptation to the geological type of new projects and the execution of the following algorithms:

[0096] A1. Clean, normalize, and extract features from multi-source data.

[0097] A2. A deep learning model is used to establish a mapping relationship from multiple parameters to product performance. This model can learn the complex relationship between multiple parameters and the performance of eco-materials, and realize the prediction of the performance of eco-materials, which can be used to predict the performance of recycled products and environmental risks.

[0098] A3. Use a multi-objective optimization algorithm to find the optimal process parameters.

[0099] A4. Convert the optimized process parameters into control commands.

[0100] The execution module connects to the central processing unit and various processing devices, and constructs a distributed control network to receive control commands and adjust device operating parameters.

[0101] The human-machine interface, connected to the central processing unit, is used to display real-time data, prediction results, equipment status and alarm information, and supports manual overwrite mode of process parameters. When the automatic decision system is under maintenance or a major deviation occurs, operators are allowed to manually input process parameters.

[0102] To precisely control the processing of construction waste, the execution module includes an adjustable vibration mechanism installed on the surface of the primary screening equipment 2, a flocculant addition mechanism installed on the surface of the flocculation conditioning tank 4, and a stirring mechanism installed on the surface of the mixing conditioning tank 3. The adjustable vibration mechanism is used to control the screening efficiency and particle size, the flocculant addition mechanism is used to precisely control the amount of flocculant added, and the stirring mechanism is used to adjust the stirring speed and stirring time.

[0103] To monitor key parameters throughout the entire waste disposal process, the sensor module includes:

[0104] A 3D laser scanner is installed on the feeding conveyor belt 21 of the preliminary screening equipment 2 to monitor the discharge volume of the slag in real time.

[0105] A dielectric constant sensor is installed on the side wall of the mixing and conditioning tank 3 to measure the dielectric constant of the slurry and calculate the overall moisture content.

[0106] A machine vision camera, installed on the discharge conveyor belt 22 of the primary screening equipment 2, is equipped with a stable light source system and is used to identify the coarse aggregate gradation of the slag.

[0107] A laser particle size analyzer is installed on the pipeline of the feed pump 51 of the filter press 5 for online analysis of the particle size distribution of fine particles.

[0108] A heavy metal ion concentration sensor is installed inside the flocculation conditioning tank 4 to detect the concentration of dissolved heavy metal ions in the slurry.

[0109] A pH sensor is installed inside the flocculation conditioning tank 4 to monitor the pH value during the flocculation process.

[0110] A moisture content sensor is installed at the sludge outlet of filter press 5 to detect the final moisture content of the sludge cake.

[0111] To generate optimal process parameters, the deep learning model run by the central processing unit is a multi-task learning network, and its model structure includes:

[0112] Long Short-Term Memory (LSTM) network layers are used to extract temporal features from multi-source sensor data.

[0113] Convolutional neural network layers are used to extract spatial features between multiple parameters.

[0114] The attention mechanism layer is used to focus on key parameters and time steps.

[0115] The multi-task output layer simultaneously outputs the classification of slag and soil types, the prediction of the compressive strength of recycled products, the prediction of the risk of heavy metal leaching, and the prediction of the chloride ion diffusion coefficient.

[0116] To achieve optimal parameter prediction, the loss function of the multi-task learning network is a weighted multi-task loss function, expressed as:

[0117]

[0118] The cross-entropy loss function for the task of classifying waste soil types is calculated as follows:

[0119]

[0120] in, For the sample size, In order to test the soil samples, This represents the total number of waste soil types. In order to measure the type of waste soil, It is a sample In category The real labels on The sample predicted by the model Category The probability of;

[0121] The mean squared error loss function for predicting the compressive strength of recycled products is expressed by the following formula:

[0122]

[0123] in, It is a sample The true compressive strength value, It is the compressive strength value predicted by the model;

[0124] The mean squared error loss function represents the task of predicting heavy metal leaching rates, and its calculation formula is as follows:

[0125]

[0126] in, It is a sample The true heavy metal leaching rate It is the leaching rate predicted by the model;

[0127] The log-mean-square error loss function for the chloride ion diffusion coefficient prediction task is expressed by the following formula:

[0128]

[0129] in, It is a sample The true chloride ion diffusion coefficient, It is the chloride ion diffusion coefficient predicted by the model;

[0130] This is the L2 regularization term applied to all parameters of the deep learning model;

[0131] , , , , These are configurable hyperparameters used to balance the weights of classification task loss, compressive strength loss, leaching rate loss, diffusion coefficient loss, and regularization term in the total loss.

[0132] To achieve optimal parameters for data processing, the central processing unit runs a multi-objective optimization algorithm based on an improved multi-objective evolutionary algorithm using NSGA-II. Its objective function includes:

[0133] C1. Maximize the compressive strength of recycled products Its expression is:

[0134]

[0135] in, The compressive strength of the recycled product predicted by the deep learning model according to claim 4;

[0136] C2. Minimize heavy metal leaching rate Its expression is:

[0137]

[0138] in, The heavy metal leaching rate predicted by the deep learning model according to claim 4;

[0139] C3. Minimize overall processing cost Its expression is:

[0140]

[0141] in, To comprehensively consider processing costs, Indicates energy consumption cost, Indicates the cost of the medicine. Indicates disposal costs;

[0142] It is a vector of decision variables, including process parameters such as stirring speed, flocculant dosage, and filter press pressure.

[0143] To achieve the screening action, the adjustable vibration mechanism includes a vibrating screen 6, which is inclined and installed inside the support of the primary screening equipment 2. The high end of the vibrating screen 6 is the feed end, and the low end is the discharge end. This allows the inclined vibrating screen 6 to utilize gravity to assist screening, improving processing efficiency. A vibrating motor 61 is installed at the bottom of the vibrating screen 6, and a discharge conveyor belt 22 is located below the vibrating screen 6. A receiving box 62 is located on one side of the primary screening equipment 2. The central control unit adjusts the frequency or amplitude of the vibrating motor 61 to change the intensity of screening by the vibrating screen 6, thereby controlling the particle size distribution of the undersize material and meeting the aggregate particle size requirements of different resource utilization pathways. The lower surface of the vibrating screen 6 is connected to the inner surface of the support of the primary screening equipment 2 through multiple shock-absorbing springs 63 arranged in a rectangular array. The rectangular array of shock-absorbing springs 63 effectively absorbs the high-frequency vibration generated by the vibrating motor 61, preventing vibration from being transmitted to the support of the primary screening equipment 2, ensuring the stable operation of the entire system, and reducing noise.

[0144] To achieve the mixing action, the mixing mechanism includes a mixing motor 7 fixedly installed on the upper surface of the mixing and conditioning tank 3. One end of the output shaft of the mixing motor 7 extends into the mixing and conditioning tank 3 and is fixedly fitted with a mixing shaft paddle 71. The upper end of the mixing shaft paddle 71 is mounted on the inner top wall of the mixing and conditioning tank 3 through a bearing. The inlet of the mixing and conditioning tank 3 is connected to one end of the discharge conveyor belt 22 to facilitate the reception of the screened slag into the mixing and conditioning tank 3. The outlet of the mixing and conditioning tank 3 is connected to the inlet of the flocculation conditioning tank 4 through a pipe. A suction pump 72 is installed on the surface of the pipe. The rotation of the output shaft of the mixing motor 7 drives the mixing shaft paddle 71 connected to it to rotate. After the mixture is evenly mixed, the suction pump 72 pumps the homogenized slag slurry from the mixing and conditioning tank 3 to the flocculation conditioning tank 4. The central control unit adjusts the power of the suction pump 72 to match the processing rhythm of the entire system, thereby achieving precise process control.

[0145] To accurately add flocculant, the flocculant addition mechanism includes a connecting pipe 8 fixedly installed on the upper surface of the flocculation conditioning tank 4. A U-shaped pipe 81 is fixedly connected to the upper end of the connecting pipe 8. Solenoid valves 82 are installed on the two branch pipes of the U-shaped pipe 81, and metering pumps 83 are respectively installed on the free ends of the two branch pipes. The central processing unit is electrically connected to the metering pumps 83 and the solenoid valves 82. The solenoid valves 82 control the opening and closing of the U-shaped pipe 81, forming a switchable dual-path dosing system. If one path becomes blocked or malfunctions... When cleaning or maintenance is required, the central processing unit automatically controls the solenoid valve 82 to switch to another path to continue working, ensuring uninterrupted operation of the dosing process and the entire production line. It is particularly suitable for use in engineering environments that require continuous production. At the same time, one path can add conventional flocculant, and the other path can add coagulant aid to adjust the pH value. The central processing unit can intelligently select the dosing path and agent ratio according to the real-time pH value of the slag slurry to realize more complex conditioning processes. Moreover, the metering pump 83 ensures the high accuracy of flocculant dosing and avoids agent waste.

[0146] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A multi-parameter decision deep networked slag resource utilization device, characterized in that: The application relates to a sludge treatment system, which comprises the following components: a sludge inlet (1) for receiving sludge generated by a shield tunneling machine; a primary screening device (2) connected with the sludge inlet (1) and used for separating large stones and sundries; a stirring conditioning tank (3) connected with a discharge port of the primary screening device (2) and used for fully mixing the sludge by mechanical stirring to form a slurry with uniform components; a flocculation conditioning tank (4) connected with a discharge port of the stirring conditioning tank (3) and used for adding a flocculating agent to the slurry to make fine particles gather to form flocculation; a filter press (5) connected with a discharge port of the flocculation conditioning tank (4) and used for mechanically filtering and dewatering the slurry after flocculation to form a mud cake; a sensor module integrated with multiple sensors and used for monitoring key parameters of the whole sludge treatment process in real time; a data acquisition and transmission module electrically connected with the sensor module, used for acquiring sensor data and transmitting the data through a wireless network, and simultaneously having edge computing capability and cloud collaborative communication, and performing local data caching and preliminary processing when the network is interrupted; a central processing unit in communication connection with the data acquisition and transmission module, used for running a deep learning model and a multi-objective optimization algorithm, realizing multi-parameter decision and generating optimal process parameters, and running the following algorithms: A1, cleaning, normalizing and feature extracting of multi-source data; A2, establishing a mapping relationship from multiple parameters to product performance by using a deep learning model, the model can learn the complex relationship between multiple parameters and ecological material performance, and realize the prediction of the ecological material performance, and is used for predicting the performance of regenerated products and environmental protection risks; A3, finding optimal process parameters by using a multi-objective optimization algorithm; A4, converting the optimized process parameters into control instructions; an execution module connected with the central processing unit and each processing device, and used for receiving the control instructions and adjusting the running parameters of the devices, and used for constructing a distributed control network; and a man-machine interaction interface connected with the central processing unit and used for displaying real-time data, prediction results, device states and alarm information, and supporting a manual override mode of process parameters, and allowing an operator to manually input process parameters when the automatic decision system is maintained or has a major deviation. ​ The execution module comprises an adjustable vibration mechanism arranged on the surface of the primary screening device (2), a flocculating agent adding mechanism arranged on the surface of the flocculation conditioning tank (4) and a stirring mechanism arranged on the surface of the stirring conditioning tank (3), the adjustable vibration mechanism is used for controlling the screening efficiency and particle size, the flocculating agent adding mechanism is used for accurately controlling the flocculating agent dosage, and the stirring mechanism is used for adjusting the stirring speed and stirring time. The sensor module comprises the following components: a three-dimensional laser scanner installed on a feeding conveyor belt (21) of the primary screening device (2) and used for monitoring the sludge discharge volume in real time; a dielectric constant sensor installed on the side wall of the stirring conditioning tank (3) and used for measuring the dielectric constant of the slurry and calculating the overall water content; and a machine vision camera installed on a discharge conveyor belt (22) of the primary screening device (2) and equipped with a stable light source system and used for identifying the coarse aggregate gradation of the sludge. ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ 2. The multi-parameter decision deep networked sludge resource utilization device according to claim 1, characterized in that: ​ 3. The multi-parameter decision deep networked sludge resource utilization device according to claim 1, characterized in that: ​ ​ ​ ​ A laser particle size analyzer is installed on a pipeline of the filter press (5) close to a feed pump (51) to analyze the particle size distribution of fine particles on line. A heavy metal ion concentration sensor is installed in the flocculation conditioning tank (4) to detect the concentration of dissolved heavy metal ions in the slurry. A PH sensor is installed in the flocculation conditioning tank (4) to monitor the PH value during flocculation. A water content sensor is installed at the mud outlet of the filter press (5) to detect the final water content of the mud cake.

4. The multi-parameter decision deep networked sludge resource utilization device according to claim 1, characterized in that: The deep learning model run by the central processing unit is a multi-task learning network, and the model structure includes: A long short-term memory network layer is used to extract time sequence features of multi-source sensor data. A convolutional neural network layer is used to extract spatial features between multiple parameters. An attention mechanism layer is used to focus on key parameters and time steps. A multi-task output layer simultaneously outputs the classification of the slag type, the prediction of the compressive strength of the recycled product, the prediction of the heavy metal leaching risk, and the prediction of the chloride ion diffusion coefficient.

5. The multi-parameter decision deep networked sludge resource utilization device according to claim 4, characterized in that: The loss function of the multi-task learning network is a weighted multi-task loss function, and the expression is: a cross-entropy loss function representing the slag type classification task, which is calculated as: where, is the number of samples, is the measured slag soil sample, is the total number of slag soil type classes, is the measured slag soil class, is the true label of the sample in the class , and is the probability that the model predicts the sample belongs to the class . The mean squared error loss function representing the recycled product compressive strength prediction task is calculated as: wherein, is the true compressive strength value of the sample is the model predicted compressive strength value;​ a mean squared error loss function representing the heavy metal leaching rate prediction task, which is calculated as: wherein, is the true heavy metal leaching rate of the sample is the true heavy metal leaching rate of the sample is the model predicted leaching rate; a log mean squared error loss function representing the chloride ion diffusion coefficient prediction task, which is calculated as: wherein, is the true chloride diffusion coefficient of the sample , is the model predicted chloride diffusion coefficient; L2 regularization term applied on the entire deep learning model parameters; , , , , are configurable hyperparameters used to balance the weights of the classification task loss, the compressive strength loss, the leaching rate loss, the diffusion coefficient loss, and the regularization term in the total loss, respectively.

6. The multi-parameter decision deep networked sludge resource utilization device according to claim 5, characterized in that: The multi-objective optimization algorithm run by the central processing unit is an improved multi-objective evolutionary algorithm based on NSGA-II, and the objective function includes: C1. Maximizing the compressive strength of the recycled product whose expression is: wherein, is the compressive strength of the recycled product predicted by the deep learning model of claim 4; C2, minimization of heavy metal leaching whose expression is: wherein, is the heavy metal leaching rate predicted by the deep learning model according to claim 4; C3, minimize the overall processing cost whose expression is: wherein, is the total treatment cost, represents the energy cost, represents the drug cost, represents the disposal cost; is the decision variable vector, including process parameters such as stirring speed, flocculant addition amount, and filter pressure.

7. The multi-parameter decision deep networked sludge resource utilization device according to claim 2, characterized in that: The adjustable vibration mechanism includes a vibrating screen (6) arranged in an inclined manner inside the support of the preliminary screening device (2), a vibrating motor (61) is installed at the bottom of the vibrating screen (6), the discharge conveyor belt (22) is arranged below the vibrating screen (6), a material collecting box (62) is arranged below one side of the preliminary screening device (2), and the lower surface of the vibrating screen (6) is connected to the inner surface of the support of the preliminary screening device (2) through a plurality of shock-absorbing springs (63) arranged in a rectangular array.

8. The multi-parameter decision deep networked sludge resource utilization device according to claim 7, characterized in that: The stirring mechanism includes a stirring motor (7) fixedly installed on the upper surface of the stirring conditioning tank (3), one end of the output shaft of the stirring motor (7) extends into the stirring conditioning tank (3) and is fixedly sleeved with a stirring shaft paddle (71), the upper end of the stirring shaft paddle (71) is installed on the inner top wall of the stirring conditioning tank (3) through a bearing, the feed inlet of the stirring conditioning tank (3) is in communication with one end of the discharge conveyor belt (22), the discharge outlet of the stirring conditioning tank (3) is in communication with the feed inlet of the flocculation conditioning tank (4) through a pipeline, and the surface of the pipeline is provided with a suction pump (72).

9. The multi-parameter decision deep networked sludge resource utilization device according to claim 2, characterized in that: The flocculant adding mechanism includes a connecting pipe (8) fixedly installed on the upper surface of the flocculation conditioning tank (4), the upper end of the connecting pipe (8) is fixedly communicated with a U-shaped pipe (81), two branch pipes of the U-shaped pipe (81) are respectively provided with electromagnetic valves (82), and the free end faces of the two branch pipes of the U-shaped pipe (81) are respectively provided with metering pumps (83).

10. The multi-parameter decision deep networked sludge resource utilization device according to claim 1, characterized in that: The central processing unit is preconfigured with a model trained based on slag samples in nine major geological regions nationwide, which can be quickly migrated and adapted according to the geological type of a new project.

Images