Slurry dehydration-tail water treatment-free method and system based on dual-stage neural network model

Abstract

The invention discloses a slurry dehydration-tail water treatment-free method and system based on a dual-stage neural network model. The method comprises the steps that initial water quality parameters before slurry tail water dehydration are collected; the target tail water quality standard is input into the mixing amount inversion model to be processed, and the zeolite mixing amount is output; splicing the initial water quality parameter and the zeolite doping amount into an input vector, inputting the input vector into a tail water quality forward prediction model for prediction, and outputting the water quality parameter of the dehydrated tail water to obtain predicted water quality; judging whether the predicted water quality meets the emission standard or not, if not, dynamically adjusting the zeolite doping amount and predicting again until the emission standard is met, outputting the recommended zeolite doping amount, and applying the recommended zeolite doping amount to field tail water treatment operation. According to the method, the zeolite feeding precision is remarkably improved, the required zeolite mixing amount is reversely deduced from the target water quality index by adopting the inversion model, and the problem of excess or insufficient feeding caused by empirical feeding is effectively avoided.

Description

Technical Field

[0001] This invention relates to the field of intelligent construction and environmental engineering technology, and in particular to a method and system for mud dewatering and tailwater treatment-free treatment based on a two-stage neural network model. Background Technology

[0002] In municipal pipeline repair, river dredging, and tunnel boring projects, large quantities of high-moisture-content mud are often generated, and the effluent often contains high concentrations of ammonia nitrogen (NH4+). + Pollutants include nitrogen (N), total nitrogen (TN), and chemical oxygen demand (COD, CODMn). Traditional sludge dewatering processes, such as natural sedimentation, mechanical pressure filtration, or polymer flocculation, typically focus on the dewatering effect of the sludge cake, making it difficult to simultaneously achieve effective removal of pollutants from the effluent. This results in the effluent requiring additional treatment before it can be discharged in compliance with standards, increasing operating costs and environmental risks.

[0003] In recent years, modified zeolite has been widely studied for its application in sludge treatment due to its excellent adsorption capacity for cations such as ammonia nitrogen. However, most existing zeolite dosing strategies rely on empirical rules or experimental proportions, lacking a response mechanism to the initial effluent quality and making it difficult to accurately predict the effluent indicators after treatment. This often leads to overdosing or ineffective treatment. Furthermore, most related studies focus on static batch treatment or laboratory simulations, lacking intelligent prediction and feedback mechanisms that can be deployed on-site.

[0004] Currently, there is no mature method to achieve a two-way modeling and control process that "derives the required zeolite dosage from the target emission standard and predicts the treated water quality." Traditional methods have the following shortcomings:

[0005] Weak predictive capability: Most treatment processes cannot predict the final water quality indicators based on the initial effluent characteristics, resulting in a lag in system response;

[0006] Unscientific dosage decisions: lack of data-driven optimization mechanisms, unstable dosage, and serious waste of resources;

[0007] Lack of feedback control: No adaptive adjustment mechanism for dosage based on prediction error has been established, making it difficult to cope with complex water quality fluctuations;

[0008] Lack of an integrated sensing-prediction-execution platform: Most studies have not achieved system integration of pollutant sensing, model prediction and intelligent feeding, resulting in poor field adaptability.

[0009] For example, using a fixed dose of zeolite to treat tailwater from tunnel boring machines can reduce COD by about 60% under laboratory conditions. However, in actual engineering projects, the mud fluctuates greatly, making this method difficult to promote, and there is a lack of predictive control models for the treatment effect. Summary of the Invention

[0010] To address the technical problems existing in the prior art, this invention proposes a mud dewatering-effect water treatment-free method and system based on a dual-stage neural network model. By combining the collection of effect water indicators by a field sensing system, a dual network structure of "inversion model + forward model" is constructed to achieve target-oriented zeolite dosage decision-making and effect water quality prediction, ultimately forming a closed-loop, automated effect water treatment control system.

[0011] On the one hand, to achieve the above objectives, the present invention provides a method for mud dewatering and tailwater treatment-free treatment based on a two-stage neural network model, comprising:

[0012] Initial water quality parameters of the sludge effluent before dewatering were collected, including ammonia nitrogen (NH4). + -N, total nitrogen (TN), chemical oxygen demand (COD), permanganate index (COD) Mn and pH value;

[0013] The target tailwater quality standard is input into the dosage inversion model for processing, and the zeolite dosage is output.

[0014] The initial water quality parameters and the zeolite dosage are concatenated into an input vector, which is then input into the effluent water quality forward prediction model for prediction. The water quality parameters of the dewatered effluent are then output to obtain the predicted water quality.

[0015] Determine whether the predicted water quality meets the discharge standards. If not, dynamically adjust the zeolite dosage and re-predict until the discharge standards are met. Output the recommended zeolite dosage and use it in on-site wastewater treatment operations.

[0016] The dosage inversion model is constructed using a random forest regression algorithm, and the tailwater quality forward prediction model is a multilayer perceptron (MLP) neural network.

[0017] Preferably, the random forest in the dosage inversion model contains no less than 100 decision trees to achieve a nonlinear mapping between the target water quality index and the zeolite dosage;

[0018] The target effluent water quality standard is: NH4 + -N≤1.0mg / L, TN≤15mg / L, COD≤50mg / L, CODMn≤4mg / L, pH≈7.

[0019] Preferably, the doping inversion model is as follows:

[0020]

[0021] In the formula, x zeolite For the predicted amount of zeolite added, Y target For the target water quality, T iLet n represent the prediction function constructed by the i-th regression tree, where n is the total number of trees.

[0022] Preferably, the forward prediction model for effluent quality adopts a two-layer hidden layer structure, and the activation function is selected as ReLU or GELU to simulate the nonlinear relationship between zeolite dosage and effluent quality after dewatering.

[0023] Preferably, the wastewater quality forward prediction model is as follows:

[0024] Y pred =f MLP ([X init ,x zeolite ])=σ (2) (W (2) ·σ (1) (W (1) ·X+b (1) )+b (2) ),

[0025] In the formula, σ (1) With σ (2) W represents the activation functions for the first and second layers, respectively. (1) and W (2) Both are weight matrices, b (1) b (2) All are bias terms; X is the concatenated vector of input variables, Y pred X is the predicted water quality value. init For the initial effluent quality, x zeolite This represents the predicted amount of zeolite to be added.

[0026] Preferably, dynamically adjusting the zeolite content includes:

[0027] Adjust the zeolite dosage within a preset range, and the number of iterations does not exceed a preset number of rounds.

[0028] On the other hand, to achieve the above objectives, the present invention also provides a mud dewatering-effect water treatment-free system based on a two-stage neural network model, comprising:

[0029] Water quality sensing subsystem: used to collect initial water quality parameters of sludge tailwater before dewatering, including ammonia nitrogen (NH4). + -N, total nitrogen (TN), chemical oxygen demand (COD), permanganate index (COD) Mn and pH value;

[0030] Model prediction subsystem: It is used to make predictions using the dosage inversion model and the effluent water quality forward prediction model, respectively, and output the water quality parameters of the dewatered effluent to obtain the predicted water quality.

[0031] Control decision subsystem: used to determine whether the predicted water quality meets the discharge standards; if not, it dynamically adjusts the dosage and re-predicts.

[0032] Execution and Dosing Subsystem: Used to perform automated zeolite dosing operations based on the determined dosage.

[0033] Preferably, the model prediction subsystem is built on the Python platform. In the first stage, a random forest regressor is used for doping inversion, and in the second stage, a multilayer perceptron regressor is used for tailwater index prediction.

[0034] Preferably, the control decision subsystem integrates a SHAP interpretability analysis unit, which is used to evaluate and visualize the contribution of input variables to the model output results.

[0035] Preferably, the system further includes a human-machine interface module, which is used to display tailwater quality monitoring data, model prediction results, recommended zeolite dosage and dosage adjustment suggestions in real time, and supports manual confirmation or manual parameter input to achieve model-human collaborative control.

[0036] Compared with the prior art, the present invention has the following advantages and technical effects:

[0037] (1) This invention significantly improves the accuracy of zeolite addition. By using an inversion model to deduce the required amount of zeolite from the target water quality index, it effectively avoids the problem of excessive or insufficient addition caused by empirical addition.

[0038] (2) The present invention introduces a multilayer perceptron neural network to predict the indicators of the treated effluent, so that the average prediction error is controlled within ±0.2mg / L, thereby improving the accuracy of effluent water quality prediction.

[0039] (3) The control module in this invention automatically adjusts the dosage based on model feedback to avoid non-compliance with discharge standards caused by tailwater disturbance or slurry characteristic fluctuations. It can dynamically adjust the dosage according to the actual prediction error to achieve closed-loop adaptive control. By integrating the SHAP interpretability analysis module, it can output different indicators (such as NH4). + The weight distribution of -N and pH in the prediction results provides a basis for subsequent operation and maintenance personnel to adjust the processing parameters;

[0040] (4) The model of this invention can be periodically fine-tuned according to the actual processing data to maintain long-term adaptability to different mud sources and tailwater fluctuations, and reduce the risk of "model aging"; compared with the traditional "dosing-sampling-detection-dosing" mode, this invention can complete the prediction of dosage and standard judgment before the first mud treatment dosing, which greatly shortens the treatment process and drainage waiting time.

[0041] (5) This invention can simultaneously meet multiple types of wastewater indicators, significantly enhancing environmental friendliness and intelligent decision-making capabilities. It is also adaptable to various sludge treatment scenarios and has good versatility and system expansion capabilities. Attached Figure Description

[0042] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0043] Figure 1 This is a flowchart of a mud dewatering-tailwater treatment-free method based on a two-stage neural network model according to an embodiment of the present invention.

[0044] Figure 2 This is a flowchart of the mud treatment control based on dual-model prediction and threshold closed-loop adjustment according to an embodiment of the present invention. Detailed Implementation

[0045] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0046] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0047] like Figures 1-2 This embodiment proposes a mud dewatering-effect water treatment-free method based on a two-stage neural network model, including:

[0048] Initial water quality parameters of the sludge effluent before dewatering were collected, including ammonia nitrogen (NH4). + -N, total nitrogen (TN), chemical oxygen demand (COD), permanganate index (COD) Mn and pH value;

[0049] The target tailwater quality standard is input into the dosage inversion model for processing, and the zeolite dosage is output.

[0050] The initial water quality parameters and the zeolite dosage are concatenated into an input vector, which is then input into the effluent water quality forward prediction model for prediction. The water quality parameters of the dewatered effluent are then output to obtain the predicted water quality.

[0051] Determine whether the predicted water quality meets the discharge standards. If not, dynamically adjust the zeolite dosage and re-predict until the discharge standards are met. Output the recommended zeolite dosage and use it in on-site wastewater treatment operations.

[0052] The dosage inversion model is constructed using a random forest regression algorithm, and the tailwater quality forward prediction model is a multilayer perceptron (MLP) neural network.

[0053] Furthermore, the random forest in the dosage inversion model contains no fewer than 100 decision trees, which are used to achieve a nonlinear mapping between the target water quality index and the zeolite dosage.

[0054] The target effluent water quality standard is: NH4 + -N≤1.0mg / L, TN≤15mg / L, COD≤50mg / L, CODMn≤4mg / L, pH≈7.

[0055] Specifically, the doping inversion model is as follows:

[0056]

[0057] In the formula, x zeolite For the predicted amount of zeolite added, Y target For the target water quality, T i Let n represent the prediction function constructed by the i-th regression tree, where n is the total number of trees.

[0058] Furthermore, the forward prediction model for tailwater quality adopts a two-layer hidden layer structure with 64 and 32 nodes respectively, and the activation function is ReLU or GELU to simulate the nonlinear relationship between zeolite dosage and tailwater quality after dewatering.

[0059] Specifically, the wastewater quality forward prediction model is as follows:

[0060] Y pred =f MLP ([X init ,x zeolite ])=σ (2) (W (2) ·σ (1) (W (1) ·X+b (1) )+b (2) ),

[0061] In the formula, σ (1) With σ (2) W represents the activation functions for the first and second layers, respectively. (1) and W (2) Both are weight matrices, b (1) b (2) All are bias terms; X is the concatenated vector of input variables, Y pred X is the predicted water quality value. init For the initial effluent quality, x zeoliteThis represents the predicted amount of zeolite to be added.

[0062] Furthermore, dynamically adjusting the zeolite content includes:

[0063] Adjust the zeolite dosage within a preset range, and the number of iterations does not exceed a preset number of rounds.

[0064] Specifically, in this embodiment, if there are any substandard items in the forward prediction, the model automatically feeds back to the inversion stage, adjusts the doping level by ±0.5%, and performs iterative optimization. Generally, all indicators can be achieved within 2 to 3 rounds.

[0065] This embodiment also provides a mud dewatering-effect water treatment-free system based on a two-stage neural network model, including:

[0066] Water quality sensing subsystem: used to collect initial water quality parameters of sludge tailwater before dewatering, including ammonia nitrogen (NH4). + -N, total nitrogen (TN), chemical oxygen demand (COD), permanganate index (COD) Mn and pH value;

[0067] Model prediction subsystem: It is used to make predictions using the dosage inversion model and the effluent water quality forward prediction model, respectively, and output the water quality parameters of the dewatered effluent to obtain the predicted water quality.

[0068] Control decision subsystem: used to determine whether the predicted water quality meets the discharge standards; if not, it dynamically adjusts the dosage and re-predicts.

[0069] Execution and Dosing Subsystem: Used to perform automated zeolite dosing operations based on the determined dosage.

[0070] Specifically, the data sampling cycle of the water quality sensing subsystem is no more than once every 10 seconds, and the measurement error is controlled within ±5%.

[0071] The execution and dosing subsystem is linked with the programmable logic controller (PLC) system to realize automatic metering, conveying and precise dosing control of zeolite.

[0072] Furthermore, the model prediction subsystem is built on the Python platform. In the first stage, a random forest regressor is used for doping inversion, and in the second stage, a multilayer perceptron regressor is used for tailwater index prediction.

[0073] Specifically, the model prediction subsystem has a dynamic update mechanism, which can collect and learn historical prediction errors based on the monitoring results of the actual treated tailwater, and periodically update the parameters of the inversion model and the forward prediction model to improve the model prediction accuracy and system adaptability.

[0074] Furthermore, the control decision subsystem integrates a SHAP interpretability analysis unit, which is used to evaluate and visualize the contribution of input variables to the model output results.

[0075] Specifically, the system also includes a human-machine interface module, which is used to display tailwater quality monitoring data, model prediction results, recommended zeolite dosage and dosage adjustment suggestions in real time, and supports manual confirmation or manual parameter input to achieve model-human collaborative control.

[0076] Furthermore, in this embodiment, the system also includes a tailwater quality sensing subsystem, which includes an ion-selective electrode, an ultraviolet-visible absorption spectrometer, a conductivity meter, and a multi-parameter water quality sensing module. This subsystem is used to collect key pollutant indicators of the tailwater in real time and input the collected data into a two-stage neural network model mud dewatering-tailwater treatment-free system for processing.

[0077] The system described in this embodiment is applicable to the control and treatment of sludge tailwater pollutants generated in engineering scenarios such as river dredging, port silt removal, shield tunneling, and sludge pond treatment.

[0078] To more clearly illustrate the technical solution of the present invention, specific embodiments are provided below for description:

[0079] Example 1

[0080] (1) Preparation of raw data:

[0081] Samples of sludge effluent before dewatering were collected from a city river dredging project, and the main water quality indicators were measured, including: ammonia nitrogen (NH4). + -N), total nitrogen (TN), chemical oxygen demand (COD), permanganate index (COD) Mn And pH value.

[0082] All water quality tests were conducted using national standard methods (GB 7481-87, HJ 535-2009, etc.), and were calibrated online using an ultraviolet spectrometer, conductivity meter, and multi-parameter water quality electrode. A total of 120 sets of sample data were collected and divided into training and test sets at an 8:2 ratio.

[0083] (2) Neural network modeling process:

[0084] Phase 1: Dosage Inversion Model (with target water quality as input);

[0085] Model type: Random Forest Regressor;

[0086] Input characteristics: Target discharge water quality index (NH4) +-N≤1.0mg / L, TN≤15mg / L, COD≤50mg / L, CODMn≤4mg / L, pH≈7);

[0087] Output: Recommended zeolite addition amount (unit: % mass fraction);

[0088] The model parameters are as follows:

[0089] Random Forest Regressor(n_estimators=300, max_depth=10, random_state=42)

[0090] Mean absolute error (MAE): 0.018;

[0091] Coefficient of determination (R) 2 ): 0.934.

[0092] Phase Two: Forward Prediction Model for Wastewater Quality

[0093] Model type: Multilayer perceptron neural network (MLP Regressor + Multi Output Regressor);

[0094] Input characteristics: initial water quality indicators + zeolite dosage;

[0095] Output result: Processed NH4 + -N, TN, COD, COD Mn pH value;

[0096] Neural network structure:

[0097] Input layer dimension: 6;

[0098] Hidden layer: 64 → 32 (activation function is ReLU);

[0099] Output layer dimension: 5;

[0100] Optimizer: Adam, learning rate 0.001, training epochs 500.

[0101] The model's performance on the test set after training is shown in Table 1.

[0102] Table 1

[0103]

[0104] (3) Control logic and feedback mechanism:

[0105] A threshold-based closed-loop strategy is implemented: if any forward predictions fail to meet the targets, the model automatically feeds back to the inversion stage, adjusting the doping level by ±0.5% for iterative optimization. Generally, all indicators can be achieved within 2-3 rounds.

[0106] (4) Practical application testing:

[0107] The target effluent NH4 requirement + Taking -N≤1.0mg / L as an example, the inversion model suggests a dosage of 5.5% (mass fraction). After actual addition, the measured NH4+ level after treatment... + -N is 0.87 mg / L, which meets the national Class V surface water discharge standard.

[0108] Compared with the traditional empirical dosing method: the average dosage is reduced by 19.3%; the water quality failure rate is reduced from 26.7% to 4.2%; and the dosing process time is shortened by 38%.

[0109] This embodiment demonstrates that by constructing a two-stage neural network model, the optimal zeolite dosage and tailwater compliance level can be quickly predicted without conducting pre-experiments, realizing intelligent, quantitative, and adjustable control of the tailwater treatment process, which has extremely high engineering promotion value.

[0110] Significantly improves the accuracy of zeolite dosing. By using an inversion model to deduce the required zeolite dosage from the target water quality indicators, the problem of over- or under-dosing caused by empirical dosing is effectively avoided.

[0111] Example 2

[0112] This embodiment aims to verify the accuracy of the forward prediction model for effluent quality under different zeolite dosage conditions in response to the indicators of the treated effluent, as well as the model's adaptability and optimization capability under low / high dosage conditions.

[0113] Select initial tailwater sample (NH4) + -N=3.5mg / L, TN=21.6mg / L, COD=110mg / L, CODMn=7.2mg / L, pH=6.7) were used as test inputs.

[0114] Five different zeolite dosage conditions were set up, as shown in Table 2.

[0115] Table 2

[0116]

[0117] The predicted values ​​and actual measured values ​​obtained by inputting the above five dosages into the effluent water quality forward prediction model are shown in Table 3.

[0118] Table 3

[0119]

[0120] Mean absolute error (MAE) = 0.04 mg / L.

[0121] When the doping concentration is 5.0%, the model predicts a value of 0.89 mg / L, which meets the target setting (NH4). + With a concentration of -N≤1.0mg / L, the dosage is economically viable. If the optimal cost is the target, the recommended dosage is 4.7% ± 0.2%.

[0122] This embodiment verifies that the model has high prediction accuracy under different dosage levels, and can maintain an error of less than ±0.1 mg / L even near the low dosage critical point. This indicates that the model has good stability and generalizability, and can be used for dosage control and prediction in multiple scenarios.

[0123] By introducing a multilayer perceptron neural network to predict the parameters of the treated effluent, the average prediction error is controlled within ±0.2 mg / L, and the root mean square error of the prediction is reduced by 62% compared with the traditional linear model. The dosage is automatically adjusted based on model feedback to avoid non-compliance with discharge standards caused by effluent disturbances or fluctuations in sludge characteristics. The system can dynamically adjust the dosage according to the actual prediction error, achieving closed-loop adaptive control.

[0124] Example 3

[0125] This embodiment provides a mud dewatering-effect water treatment-free system based on a two-stage neural network model. The specific system deployment is as follows.

[0126] The hardware and software modules of this system were deployed on an actual mud dewatering production line, including:

[0127] Water quality sensing subsystem: equipped with a conductivity meter, ultraviolet spectrophotometer (UV1800), and multi-parameter electrodes, with a sampling frequency of 10 seconds / time.

[0128] Model prediction subsystem: An embedded industrial control computer (Advantech IPC) is used to run a two-stage neural network model in a Python environment.

[0129] Control decision subsystem: Based on a set threshold closed loop, it dynamically judges whether the model prediction meets the target and automatically issues adjustment signals.

[0130] Execution and Dosing Subsystem: Executed by a screw feeder controlled by a PLC, receiving control commands with an accuracy of ±1%.

[0131] The control logic flow is as follows:

[0132] (1) Real-time sensing of effluent indicators (including NH4) + -N, TN, COD, pH, etc.

[0133] (2) Set target emission requirements (e.g., TN≤15mg / L).

[0134] (3) Input the target into the inversion model and output the recommended zeolite content.

[0135] (4) The data is transmitted to the forward prediction model and the pretreated water quality is output.

[0136] (5) If any index exceeds the limit, the system will provide feedback to adjust the dosage and perform iterative optimization.

[0137] (6) After determining the final recommended value, the control module sends a command to the PLC.

[0138] (7) The dosing machine adds the medicine according to the instructions to achieve automatic proportioning.

[0139] The entire process takes no more than 20 seconds, data latency is controlled within 500ms, and the system runs stably for 12 hours without any abnormalities.

[0140] System performance:

[0141] On-site processing capacity: approximately 10m³ 3 / h; Prediction accuracy: NH4 + -N average error ±0.08mg / L; dosing stability: within ±1.2%. Compared with manual, experience-based dosing, treatment efficiency is increased by 32%; the pass rate is increased to 95.6% (an increase of approximately 23 percentage points); and dosing materials are saved by 18-22%.

[0142] This embodiment demonstrates the complete deployment process and intelligent dosing performance of the system in a real-world sludge tailwater treatment site, verifying that the system possesses excellent industrial adaptability, automation level, and energy-saving and emission-reduction efficiency, demonstrating broad application value. The system can be linked with a PLC automation control system to achieve unattended continuous dosing operations, reducing labor costs by approximately 40%.

[0143] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for mud dewatering and tailwater treatment-free treatment based on a two-stage neural network model, characterized in that, include: Initial water quality parameters of the sludge effluent before dewatering were collected, including ammonia nitrogen (NH4). + -N, total nitrogen (TN), chemical oxygen demand (COD), permanganate index (COD) Mn and pH value; The target tailwater quality standard is input into the dosage inversion model for processing, and the zeolite dosage is output. The initial water quality parameters and the zeolite dosage are concatenated into an input vector, which is then input into the effluent water quality forward prediction model for prediction. The water quality parameters of the dewatered effluent are then output to obtain the predicted water quality. Determine whether the predicted water quality meets the discharge standards. If not, dynamically adjust the zeolite dosage and re-predict until the discharge standards are met. Output the recommended zeolite dosage and use it in on-site wastewater treatment operations. The dosage inversion model is constructed using a random forest regression algorithm, and the tailwater quality forward prediction model is a multilayer perceptron (MLP) neural network.

2. The mud dewatering-effect water treatment-free method based on a two-stage neural network model according to claim 1, characterized in that, The random forest in the dosage inversion model contains no fewer than 100 decision trees, which are used to achieve a nonlinear mapping between the target water quality index and the zeolite dosage. The target effluent water quality standard is: NH4 + -N≤1.0mg / L, TN≤15mg / L, COD≤50mg / L, CODMn≤4mg / L, pH≈7.

3. The mud dewatering-effect water treatment-free method based on a two-stage neural network model according to claim 2, characterized in that, The doping inversion model is as follows: In the formula, x zeolite For the predicted amount of zeolite added, Y target For the target water quality, T i Let n represent the prediction function constructed by the i-th regression tree, where n is the total number of trees.

4. The mud dewatering-effect water treatment-free method based on a two-stage neural network model according to claim 1, characterized in that, The forward prediction model for tailwater quality adopts a two-layer hidden layer structure, and the activation function is ReLU or GELU, which is used to simulate the nonlinear relationship between zeolite dosage and tailwater quality after dewatering.

5. The mud dewatering-effect water treatment-free method based on a two-stage neural network model according to claim 4, characterized in that, The tailwater quality forward prediction model is as follows: Y pred =f MLP ([X init ,x zeolite ])=σ (2) (W (2) ·σ (1) (W (1) ·X+b (1) )+b (2) ), In the formula, σ (1) With σ (2) W represents the activation functions for the first and second layers, respectively. (1) and W (2) Both are weight matrices, b (1) b (2) All are bias terms; X is the concatenated vector of input variables, Y pred X is the predicted water quality value. init For the initial effluent quality, x zeolite This represents the predicted amount of zeolite to be added.

6. The method for mud dewatering and tailwater treatment-free treatment based on a two-stage neural network model according to claim 1, characterized in that, Dynamically adjusting the zeolite content includes: Adjust the zeolite dosage within a preset range, and the number of iterations does not exceed a preset number of rounds.

7. A mud dewatering-effect water treatment-free system based on a two-stage neural network model, characterized in that, include: Water quality sensing subsystem: used to collect initial water quality parameters of sludge tailwater before dewatering, including ammonia nitrogen (NH4). + -N, total nitrogen (TN), chemical oxygen demand (COD), permanganate index (COD) Mn and pH value; Model prediction subsystem: It is used to make predictions using the dosage inversion model and the effluent water quality forward prediction model, respectively, and output the water quality parameters of the dewatered effluent to obtain the predicted water quality. Control decision subsystem: used to determine whether the predicted water quality meets the discharge standards; if not, it dynamically adjusts the dosage and re-predicts. Execution and Dosing Subsystem: Used to perform automated zeolite dosing operations based on the determined dosage.

8. The mud dewatering-effect water treatment-free system based on a two-stage neural network model according to claim 7, characterized in that, The model prediction subsystem is built on the Python platform. In the first stage, a random forest regressor is used for inversion of the amount of pollutants, and in the second stage, a multilayer perceptron regressor is used for prediction of tailwater indicators.

9. The mud dewatering-effect water treatment-free system based on a two-stage neural network model according to claim 7, characterized in that, The control decision subsystem integrates a SHAP interpretability analysis unit, which is used to evaluate and visualize the contribution of input variables to the model output results.

10. The mud dewatering-effect water treatment-free system based on a two-stage neural network model according to claim 7, characterized in that, The system also includes a human-machine interface module, which is used to display tailwater quality monitoring data, model prediction results, recommended zeolite dosage and dosage adjustment suggestions in real time, and supports manual confirmation or manual parameter input to achieve model-human collaborative control.

Images