Intelligent dosing dynamic optimization system for landfill leachate treatment

Abstract

The present application belongs to the technical field of intelligent control of sewage treatment, and provides an intelligent dosing dynamic optimization system for landfill leachate treatment, which comprises a monitoring module, a fluctuation early warning module, a stable dosing module and a fluctuation dosing module, can monitor multi-source data of landfill leachate in real time, generate a multi-source time sequence database, analyze water quality fluctuation and future water quality fluctuation trend of landfill leachate, judge whether to generate a fluctuation signal, mark fluctuation period and stable period, construct a kinetic equation for chemical reactions in landfill leachate based on dosing, and couple with a material balance equation to establish a complete mechanism model, generate dosing decisions for landfill leachate periodically if it is currently in a stable period, trigger model parameter correction if it is in a fluctuation period, construct and train a parameter correction model based on a time sequence neural network, and periodically correct the complete mechanism model and the dosing decisions.

Description

Technical Field

[0001] This invention belongs to the field of intelligent control technology for wastewater treatment, specifically a dynamic optimization system for intelligent dosing of landfill leachate. Background Technology

[0002] Landfill leachate is a complex organic wastewater with high pollutant concentration and highly volatile water quality generated during landfill or incineration. It contains large amounts of pollutants such as chemical oxygen demand (COD), ammonia nitrogen (NH3-N), and total phosphorus (TP). If not treated properly, it can cause serious harm to soil, groundwater, and the surrounding ecological environment. Chemical treatment is the core link in the purification process of landfill leachate. By adding targeted agents, pollutants can be degraded, precipitated, or transformed. The precision of the chemical dosing and the ability to dynamically control it directly determine the treatment effect and operating cost.

[0003] In the existing technology, there are still many problems to be solved in the chemical control technology for landfill leachate treatment. On the one hand, the data monitoring dimensions are limited and the pretreatment mechanism is imperfect. Existing systems mostly focus only on raw water quality parameters and simple operating conditions, ignoring the indirect impact of environmental factors on leachate quality, resulting in insufficient comprehensiveness of data collection. At the same time, the data collected by sensors are often accompanied by short-term random fluctuations and noise interference. Traditional pretreatment methods are difficult to effectively remove outliers and correct noise, resulting in low reliability of the basic data for subsequent analysis and decision-making.

[0004] On the other hand, water quality fluctuation identification is lagging and lacks forward-looking early warning. Landfill leachate water quality is affected by multiple factors such as waste composition, landfill time, and climate conditions, exhibiting strong volatility. However, existing technologies mostly identify fluctuations through manual inspection or simple threshold judgment, resulting in response delays. This leads to insufficient adaptability and dynamism of dosing decision models. Existing dosing models are mostly based on empirical formulas or simplified mechanisms, failing to fully couple the nonlinear relationship between key reaction conditions such as stirring rate, dissolved oxygen, and pH and the dosage of chemicals. This makes it difficult to accurately reflect the actual chemical reaction process. At the same time, there is a lack of parameter correction and dynamic optimization mechanisms under water quality fluctuations. Faced with drastic water quality fluctuations, traditional systems mostly rely on manual adjustment of dosing parameters based on human experience, lacking data-driven intelligent correction methods. Even if some systems introduce machine learning models, they fail to fully utilize the long-term dependencies of time-series data, making it difficult to accurately capture the dynamic correlation between multi-source data and reaction parameters during fluctuation periods. This results in insufficient parameter correction accuracy and limited dynamic optimization capabilities of dosing decisions.

[0005] To address the aforementioned problems, this invention proposes an intelligent dynamic optimization system for dosing in landfill leachate treatment. Summary of the Invention

[0006] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0007] The technical solution adopted by this invention to solve the technical problem is: an intelligent dosing dynamic optimization system for landfill leachate treatment, comprising:

[0008] Monitoring module: Deploys multiple types of sensors to monitor multi-source data of landfill leachate in real time and generates a multi-source time-series database;

[0009] Fluctuation early warning module: Based on a multi-source time series database, it performs water quality fluctuation analysis on landfill leachate, and constructs a historical fluctuation precursor database to analyze the future water quality fluctuation trend of landfill leachate, determine whether a fluctuation signal is generated, and mark the fluctuation period and stable period based on the fluctuation signal;

[0010] Stable dosing module: Constructs kinetic equations for chemical reactions in landfill leachate based on dosing, and couples them with material balance equations to establish a complete mechanism model. If the situation is stable, a dosing decision cycle is set, and dosing decisions are generated periodically for landfill leachate based on real-time multi-source data. If the situation is fluctuating, the model parameters are corrected.

[0011] Fluctuation dosing module: If model parameter correction is triggered, set the parameter update cycle, build and train a parameter correction model based on a time-series neural network to periodically correct the model parameters of the complete mechanism model, and dynamically correct the dosing decision;

[0012] The method for analyzing water quality fluctuations is as follows:

[0013] Set a flag value and assign it an initial value of 0. Set a sliding fluctuation analysis window for landfill leachate with the current time as the endpoint. Obtain multi-source data within the current sliding fluctuation analysis window from the multi-source time series database. The multi-source data includes raw water quality parameters. Calculate the coefficient of variation of each type of raw water quality parameter within the current sliding fluctuation analysis window.

[0014] If one or more coefficients of variation are greater than the preset coefficient of variation standard at the current moment, the flag value is incremented once; otherwise, the flag value is reset to 0. If the flag value is greater than or equal to the preset flag value threshold, it is determined that water quality fluctuation has occurred.

[0015] The historical fluctuation precursor database is obtained as follows:

[0016] Get the flag value, mark the moment when the flag value is equal to the flag value threshold as the fluctuation moment, set the historical time period with the current time as the end, get all the fluctuation moments within the historical time period, set the buffer period with each fluctuation moment as the end, and extract a fluctuation precursor window with the start of the buffer period as the end for each fluctuation moment.

[0017] Based on a multi-source time series database, multi-source data within each fluctuation precursor window are obtained, integrated into fluctuation precursor sequences, and summarized to obtain a historical fluctuation precursor database.

[0018] The method for analyzing future water quality fluctuation trends is as follows:

[0019] Acquire multi-source data within the current sliding fluctuation analysis window and integrate them according to time series to obtain a precursor analysis sequence. In the historical fluctuation precursor database, calculate the cosine similarity between the precursor analysis sequence and each fluctuation precursor sequence.

[0020] If the cosine similarity is greater than the preset similarity standard, it is determined that the landfill leachate will experience water quality fluctuations after the buffer period; otherwise, the landfill leachate water quality trend is determined to be stable.

[0021] The method for marking the fluctuation period and the stable period is as follows:

[0022] Set a fluctuation anchor point value and assign it an initial value of 0;

[0023] If water quality fluctuations occur at the current moment, or if it is determined that water quality fluctuations will occur after the buffer period of landfill leachate, a fluctuation signal is generated. If a fluctuation signal is generated and the fluctuation anchor value is 0, the fluctuation anchor value is changed to 1.

[0024] If no water quality fluctuations occur, and the trend of landfill leachate water quality is determined to be stable, and the fluctuation anchor value is 1, a fluctuation buffer signal is generated. If the duration of continuously generated fluctuation buffer signals exceeds the preset buffer threshold, the fluctuation anchor value is changed to 0.

[0025] Periods with consecutive fluctuation anchor values ​​of 1 are marked as fluctuation periods, and periods with consecutive values ​​of 0 are marked as stable periods;

[0026] The complete mechanism model is established as follows:

[0027] In the process of landfill leachate treatment, kinetic equations describing the relationship between chemical reaction rates and influencing factors are constructed for chemical reactions based on chemical dosing. The kinetic equations include COD degradation kinetic equations, ammonia nitrogen removal kinetic equations, and total phosphorus removal reaction equations, which are coupled with material balance equations to construct a complete mechanism model for landfill leachate chemical dosing treatment.

[0028] The coupled equations of the complete mechanism model contain the COD degradation reaction rate constant, ammonia nitrogen reaction rate constant, and total phosphorus reaction rate constant corresponding to the kinetic equations. These are all labeled as model parameters of the complete mechanism model. The initial values ​​of the model parameters are obtained by fitting experimental data.

[0029] The method for making decisions on the periodic generation and dosing of landfill leachate is as follows:

[0030] A dosing decision cycle is set for landfill leachate. The starting point of each dosing decision cycle is marked as the decision time. If the current time is the decision time, multi-source data of landfill leachate at the current time is obtained. Combining the multi-source data, the Newton-Raphson iterative method is used to solve the coupled equations of the complete mechanism model. By continuously iterating to approximate the solution of the coupled equations, the dosage of COD treatment, ammonia nitrogen treatment, and total phosphorus treatment is calculated in each iteration, and the dosing vector is obtained.

[0031] Calculate the Euclidean distance between two dosing vectors obtained in adjacent iterations. If it is less than the preset deviation distance threshold, stop the iteration. The dosing vector obtained in the last iteration is the dosing decision for landfill leachate in the current decision cycle.

[0032] The construction and training method of the complete mechanism model is as follows:

[0033] All fluctuation periods within the historical time period are obtained. Combined with a multi-source time series database, the multi-source data within each fluctuation period are integrated and normalized according to the time series to obtain a normalized multi-source time series sequence. The sliding window method is used to segment the data to obtain a sample sequence. Label data including the model parameter correction values ​​of the complete mechanism model are added to each sample sequence to obtain a training data set.

[0034] The parameter correction model consists of an input layer, a hidden layer, and an output layer. The input layer contains 12 feature dimensions, the hidden layer consists of two LSTM layers, both of which use the ReLU function as the activation function, and the output layer has 3 neurons that output the model parameter correction values ​​of the complete mechanism model. The parameter correction model is obtained by training the training dataset.

[0035] The method for periodically correcting the model parameters of the complete mechanism model is as follows:

[0036] Set parameter update cycles for landfill leachate, and mark the start of each parameter update cycle as the update time;

[0037] At the update time, a time period with the current time as the end point and the same duration as the time period corresponding to the sample sequence length is extracted. Multi-source data within the time period is obtained and normalized in the same way as the normalized multi-source time series sequence to obtain the current input sequence input parameter correction model, output the model parameter correction value, and use the model parameter correction value to correct the complete mechanism model.

[0038] The method for dynamically correcting dosing decisions is as follows:

[0039] At each update time, multi-source data of landfill leachate is acquired. Combining the multi-source data, the Newton-Raphson iterative method is used to solve the coupled equations of the corrected complete mechanism model. The iterative calculation yields the dosing vectors, including the dosing amounts for COD treatment, ammonia nitrogen treatment, and total phosphorus treatment, which are the dosing decisions for landfill leachate in the current parameter update cycle.

[0040] The beneficial effects of this invention are as follows:

[0041] 1. This invention generates a multi-source time-series database by deploying multiple types of sensors, which can accurately analyze water quality fluctuations, predict future trends and mark time periods in advance, adopt differentiated strategies for different time periods, generate dosing decisions periodically during stable periods to ensure treatment stability, trigger model parameter correction during fluctuating periods, and use time-series neural networks for dynamic optimization to effectively respond to water quality changes, improve system adaptability, and ensure continuous and stable treatment results.

[0042] 2. This invention constructs a complete mechanistic model, coupling chemical reaction kinetics with material balance equations to provide a scientific basis for dosing decisions. At the same time, by combining real-time monitoring data, it enables dynamic adjustment of dosing decisions. During periods of fluctuation, the model parameters are periodically corrected to ensure that dosing decisions closely follow changes in water quality, avoiding over-dosing or under-dosing, effectively reducing treatment costs, improving resource utilization, and achieving efficient, economical, and environmentally friendly treatment of landfill leachate. Attached Figure Description

[0043] The invention will now be further described with reference to the accompanying drawings.

[0044] Figure 1 This is a module architecture diagram of the intelligent dosing dynamic optimization system for landfill leachate treatment according to an embodiment of the present invention;

[0045] Figure 2 This is a flowchart illustrating the overall steps of the intelligent chemical dosing dynamic optimization system for landfill leachate treatment as described in this embodiment of the invention. Detailed Implementation

[0046] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0047] Example 1

[0048] Please see Figure 1 and Figure 2 As shown in the embodiment of the present invention, the intelligent chemical dosing dynamic optimization system for landfill leachate treatment includes the following modules:

[0049] Monitoring module: Deploys multiple types of sensors to monitor multi-source data of landfill leachate in real time and generates a multi-source time-series database;

[0050] Multiple types of sensors are deployed at key nodes of the landfill leachate treatment system to monitor multi-source data of landfill leachate in real time. The multi-source data includes raw water quality parameters, operating condition data and environmental data.

[0051] The raw water quality parameters include chemical oxygen demand (COD) concentration, ammonia nitrogen (NH3-N) concentration, total phosphorus (TP) concentration, dissolved oxygen concentration, pH, and conductivity. COD reflects the degree of organic pollution in landfill leachate, ammonia nitrogen concentration characterizes the level of nitrogen-containing organic pollution, total phosphorus concentration is used to assess phosphorus pollution status, pH reflects the acidity or alkalinity of the water body, and conductivity reflects the ion concentration in the water. The raw water quality parameters are collected by an online water quality analyzer installed at the inlet of the equalization tank.

[0052] The operating data includes influent flow rate, dosing pump frequency, and valve opening. Influent flow rate is collected by an electromagnetic flow meter installed in the influent pump room, dosing pump frequency is collected by the dosing device control cabinet, and valve opening is recorded by a valve positioner.

[0053] The environmental data includes rainfall, temperature, and humidity. Rainfall is collected by tipping bucket rain gauges installed in the treatment plant area, while temperature and humidity are collected by temperature and humidity sensors installed in the air environment above the regulating tank.

[0054] For the collected multi-source data, a reasonable range of values ​​is set for the multi-source data. If the multi-source data exceeds the reasonable range of values, it is judged as an outlier and the outlier is removed. The moving average filter is used to smooth the multi-source data after removing the outlier to reduce the impact of short-term random fluctuations. Kalman filter is used to correct the random noise in the multi-source data. The corrected data is used to update the multi-source data to obtain the updated multi-source data.

[0055] Set a historical time period ending at the current time. Within the historical time period, integrate all multi-source data according to time series to generate a multi-source time series database.

[0056] It should be noted that the function of this module is to collect raw water quality parameters, operating data and environmental data of landfill leachate in real time by deploying multiple types of sensors. After outlier removal, moving mean filtering and smoothing, and Kalman filtering for noise reduction, the data is integrated into a multi-source time series database, providing high-quality basic data support for subsequent water quality analysis, model building and chemical dosing decisions. The multi-dimensional data collection covers the key factors affecting landfill leachate treatment, avoiding the limitations of a single data type.

[0057] Fluctuation early warning module: Based on a multi-source time series database, it performs water quality fluctuation analysis on landfill leachate, and constructs a historical fluctuation precursor database to analyze the future water quality fluctuation trend of landfill leachate, determine whether a fluctuation signal is generated, and mark the fluctuation period and stable period based on the fluctuation signal;

[0058] Based on a multi-source time-series database, water quality fluctuation analysis of landfill leachate was performed.

[0059] Specifically, a flag value is set and initialized to 0. A sliding fluctuation analysis window is set for the landfill leachate with the current time as the endpoint. The sliding fluctuation analysis window slides with the current time and the duration remains constant. The raw water quality parameters in the current sliding fluctuation analysis window are obtained from the multi-source time series database. The standard deviation and mean of each type of raw water quality parameter in the current sliding fluctuation analysis window are calculated and the ratio is processed to obtain the coefficient of variation of each type of raw water quality parameter in the current sliding fluctuation analysis window.

[0060] If one or more coefficients of variation are greater than the preset coefficient of variation standard at the current time, the flag value is incremented; otherwise, the flag value is reset to 0.

[0061] If the current flag value is greater than or equal to the preset flag value threshold, it is determined that water quality fluctuation has occurred at the current time, and the time when the flag value is equal to the flag value threshold is marked as the fluctuation time.

[0062] If the current flag value is less than the preset flag value threshold, it is determined that there is no water quality fluctuation at the current time, and the future water quality fluctuation trend of the landfill leachate is analyzed.

[0063] Specifically, based on a multi-source time series database, all fluctuation moments within a historical period are obtained. A buffer period of the same duration is set with each fluctuation moment as the endpoint. A fluctuation precursor window is extracted for each fluctuation moment, with the starting point of the buffer period as the endpoint. The duration of all fluctuation precursor windows is the same as that of the sliding fluctuation analysis window.

[0064] In the multi-source time series database, multi-source data at each time point within all fluctuation precursor windows are obtained, and according to the time series, the corresponding multi-source data sequences are integrated within each fluctuation precursor window and marked as fluctuation precursor sequences. All fluctuation precursor sequences are summarized to construct a historical fluctuation precursor database.

[0065] The multi-source data at each moment within the current moment's sliding fluctuation analysis window are acquired and integrated according to the time series to obtain the current moment's precursor analysis sequence. In the historical fluctuation precursor database, the cosine similarity between the current moment's precursor analysis sequence and each fluctuation precursor sequence is calculated. The cosine similarity value ranges from [-1, 1]. The closer the value is to 1, the higher the similarity between the two sequences.

[0066] If the calculated cosine similarity is greater than the preset similarity standard, the current moment's precursor analysis sequence is considered to be highly similar to the corresponding fluctuation precursor sequence, and it is determined that the landfill leachate will experience water quality fluctuations after the buffer period. Otherwise, the landfill leachate water quality trend is determined to be stable.

[0067] Set a fluctuation anchor point value and assign it an initial value of 0;

[0068] If water quality fluctuations occur at the current moment or if it is determined that water quality fluctuations will occur after the buffer period, a fluctuation signal is generated. If a fluctuation signal is generated and the fluctuation anchor value is 0, the fluctuation anchor value is changed to 1, and the period in which the fluctuation anchor value is continuously 1 is marked as a fluctuation period.

[0069] If there is no water quality fluctuation at the current moment, and the trend of landfill leachate water quality is determined to be stable, and the fluctuation anchor value is 1, a fluctuation buffer signal is generated. The moments when the fluctuation buffer signal is generated continuously are merged into a fluctuation buffer period. If the duration of the fluctuation buffer period is greater than the preset buffer threshold, the fluctuation anchor value is changed to 0, and the period when the fluctuation anchor value is continuously 0 is marked as a stable period.

[0070] It should be noted that the purpose of the fluctuation buffer period is to determine the end of the fluctuation period and to provide a stable buffer time for the subsequent intelligent dosing to switch from the fluctuation period to the stable period.

[0071] It should be noted that this module is based on a multi-source time-series database. It calculates the coefficient of variation of raw water quality parameters through a sliding fluctuation analysis window to identify current water quality fluctuations. Combined with a historical fluctuation precursor database and cosine similarity analysis, it predicts future water quality fluctuation trends. Finally, it marks fluctuation periods, stable periods, and fluctuation buffer periods by fluctuation anchor values ​​and buffer periods, providing a time basis for the dynamic adjustment of subsequent dosing strategies. It identifies current water quality fluctuations in real time and predicts future fluctuation trends, realizing dual prevention and control of current anomaly monitoring and future trend early warning. This avoids the deterioration of treatment effect due to sudden changes in water quality. The setting of fluctuation buffer periods provides a stable buffer for the determination of the end point of fluctuation periods and the switching of dosing strategies, avoiding misjudgment of periods due to short-term small fluctuations and improving the accuracy of period division.

[0072] Stable dosing module: Constructs kinetic equations for chemical reactions in landfill leachate based on dosing, couples the kinetic equations with the material balance equations to establish a complete mechanism model. If the current period is stable, sets the dosing decision cycle and generates dosing decisions periodically for landfill leachate by combining real-time monitoring of multi-source data. If the current period is fluctuating, it triggers model parameter correction.

[0073] In the process of landfill leachate treatment, kinetic equations describing the relationship between chemical reaction rates and influencing factors are constructed for chemical reactions based on chemical dosing. These kinetic equations include COD degradation kinetic equations, ammonia nitrogen removal kinetic equations, and total phosphorus removal reaction equations.

[0074] Specifically, considering the diffusion effect after the drug is added, a COD degradation kinetic equation is constructed, which is as follows:

[0075] ;

[0076] in, Let C represent the COD degradation rate constant, C represent the COD concentration, and t represent time. This represents the rate of change of COD concentration over time. This represents the dosage of COD treatment, 'a' represents the COD reaction order, and 'v' represents the stirring rate. A coefficient representing the effect of stirring rate on the reaction;

[0077] By introducing dissolved oxygen concentration (DO) as a regulating term, an ammonia nitrogen removal kinetic equation is constructed as follows:

[0078] ;

[0079] in, The expression represents the ammonia nitrogen reaction rate constant, where N represents the ammonia nitrogen concentration. This indicates the rate of change of ammonia nitrogen concentration over time. denoted by , b represents the dosage of ammonia nitrogen treatment chemicals, b represents the correction coefficient for the ammonia nitrogen-chemical coupling term, pH represents the pH of the landfill leachate, c represents the correction coefficient for the ammonia nitrogen-pH coupling term, and D0 represents the dissolved oxygen concentration. The adjustment parameter indicating the effect of dissolved oxygen;

[0080] The total phosphorus removal kinetic equation is constructed as follows:

[0081] ;

[0082] in, The total phosphorus reaction rate constant is represented by P, which represents the ammonia nitrogen concentration. This indicates the rate of change of ammonia nitrogen concentration over time. d represents the total phosphorus treatment dosage, e represents the total phosphorus-dosage coupling term correction coefficient, and e represents the total phosphorus-pH coupling term correction coefficient.

[0083] The constructed COD degradation kinetic equation, ammonia nitrogen removal kinetic equation, and total phosphorus removal reaction equation were coupled with the material balance equation to construct a complete mechanism model for landfill leachate chemical treatment. The coupled equation set of the complete mechanism model is as follows:

[0084]

[0085] Where Q represents the influent flow rate and V represents the effective volume of the reaction tank. and These represent the influent COD concentration and the target effluent COD concentration, respectively. and These represent the influent ammonia nitrogen concentration and the target effluent ammonia nitrogen concentration, respectively. and These represent the influent total phosphorus concentration and the target effluent total phosphorus concentration, respectively.

[0086] Among them, the COD degradation reaction rate constant ammonia nitrogen reaction rate constant Rate constant of reaction with total phosphorus All of these are model parameters of a complete mechanism model. The initial values ​​of the model parameters were obtained through experiments and by fitting experimental data.

[0087] During stable periods, the dosing decision is determined by combining calculations from the complete mechanism model with actual engineering constraints. The actual engineering constraints include the maximum output limit of the dosing pump and the pH adjustment range.

[0088] Set up a dosing decision cycle for landfill leachate, mark the start of each dosing decision cycle as the decision time, the start of the current dosing decision cycle is also the end of the previous decision cycle, and the start of the current stable period is the start of the first dosing decision cycle within the stable period;

[0089] If the current moment is the decision moment, acquire multi-source data of landfill leachate at the current moment. Combine the multi-source data and use the Newton-Raphson iterative method to solve the coupled equations of the complete mechanism model. By continuously iterating to approximate the solution of the coupled equations, the dosage of COD treatment, ammonia nitrogen treatment, and total phosphorus treatment calculated in each iteration are sorted to obtain the dosing vector. The Euclidean distance between two dosing vectors in adjacent iterations is calculated. If the Euclidean distance is less than the preset deviation distance threshold, the iteration stops, and the dosing vector obtained in the last iteration is used as the dosing decision for landfill leachate in the current decision cycle.

[0090] If the current time is during a period of fluctuation, trigger model parameter correction;

[0091] It should be noted that the role of this module is to construct the kinetic equations for COD degradation, ammonia nitrogen removal, and total phosphorus removal in landfill leachate treatment, and couple them with material balance equations to form a complete mechanism model. During stable periods, dosing decisions are generated based on real-time data at fixed intervals, and model parameter corrections are triggered during fluctuating periods, achieving precise dosing control in different time periods. The construction of the kinetic equations considers the coupling of multiple factors, which is more in line with the actual reaction process than the traditional single-factor kinetic model. The periodic dosing decision ensures the stability of the treatment process and avoids the waste of reagents or substandard treatment caused by frequent adjustments. The proposed time-segmented strategy of periodic decision in stable periods and parameter correction in fluctuating periods breaks the adaptability limitations of a single model under complex working conditions and realizes differentiated control in stable and fluctuating scenarios.

[0092] Fluctuation dosing module: If model parameter correction is triggered, construct and train a parameter correction model based on a time-series neural network, set the parameter update period, use the parameter correction model to periodically correct the model parameters of the complete mechanism model, and dynamically correct the dosing decision;

[0093] If the model parameter correction is triggered, all fluctuation periods in the historical period are obtained. Combined with the multi-source time series database, the multi-source data of each moment in each fluctuation period are integrated according to the time series to obtain the multi-source time series sequence of the fluctuation period. Then, the multi-source data of each type in the multi-source time series sequence is normalized to obtain the normalized multi-source time series sequence containing normalized multi-source data.

[0094] A parameter correction model based on a time-series neural network is constructed. Based on all multi-source time series, the nonlinear compensation relationship between multi-source data and model parameters of the complete mechanism model within the fluctuation period is learned through the time-series neural network, so as to realize the dynamic correction of drug dosing decisions.

[0095] It should be noted that the reason for choosing a long short-term memory network is that the multi-source data of landfill leachate has strong temporal sequence. The long short-term memory network can effectively capture long-term dependencies through the gating mechanism, avoiding the gradient vanishing problem of traditional recurrent neural networks.

[0096] Specifically, the parameter correction model includes an input layer, a hidden layer, and an output layer. The input layer contains 12 feature dimensions, each corresponding to a normalized multi-source data. The hidden layer consists of two LSTM layers. The first LSTM layer has 64 neurons and is responsible for extracting local temporal features from the normalized multi-source data. The second LSTM layer has 32 neurons and abstracts and integrates the output of the first layer to capture global features. Both the first and second LSTM layers use the ReLU function as the activation function to solve the gradient vanishing problem while introducing nonlinear transformation. The output layer has 3 neurons and outputs the corrected values ​​of the model parameters of the complete mechanism model. The corrected values ​​of the model parameters include the corrected values ​​of the COD degradation reaction rate constant, the ammonia nitrogen reaction rate constant, and the total phosphorus reaction rate constant.

[0097] The sliding window method is used to segment each normalized multi-source time series to obtain sample sequences of the same length. Label data including model parameter correction values ​​is added to each sample sequence. The label data is obtained by the actual model parameters of the complete mechanism model in the corresponding fluctuation period. The training data set is then integrated.

[0098] The data training set is divided into a training set and a validation set. The parameter correction model is trained using the training set. The mean squared error is used as the loss function. The Adam optimizer is used to optimize the learning rate adaptively to accelerate convergence. An L2 regularization term is added to prevent overfitting. At the same time, an early stopping strategy is adopted. When the loss function deviation of the parameter correction model in the validation set is continuously less than the preset deviation threshold, the training is stopped, and the parameter correction model is judged to be trained successfully.

[0099] Set parameter update cycles for landfill leachate, mark the start of each parameter update cycle as the update time, the start of the current parameter update cycle is also the end of the previous parameter update cycle, and the start of the current fluctuation period is the start of the first parameter update cycle within the fluctuation period.

[0100] If the current time is the update time, extract a time period with the current time as the end point and the same duration as the time period corresponding to the sample sequence length. Obtain the multi-source data of each time within the time period and perform the same normalization process as the normalized multi-source time series sequence to obtain the current input sequence. Input the current input sequence into the parameter correction model, output the model parameter correction value, and use the model parameter correction value to correct the complete mechanism model.

[0101] The system acquires multi-source data of landfill leachate at the current moment. Combining the multi-source data, the Newton-Raphson iterative method is used to solve the coupled equations of the corrected complete mechanism model. By iteratively approximating the solution of the coupled equations, the dosage of COD treatment, ammonia nitrogen treatment, and total phosphorus treatment calculated in each iteration is sorted to obtain the dosing vector. The Euclidean distance between two dosing vectors in adjacent iterations is calculated. If the Euclidean distance is less than the preset deviation distance threshold, the iteration stops, and the dosing vector obtained in the last iteration is used as the dosing decision for landfill leachate in the current parameter update cycle.

[0102] It should be noted that the function of this module is to build a parameter correction model based on a long short-term memory network during periods of fluctuation. By training the model with historical fluctuation data, it learns the nonlinear relationship between key parameters of multi-source data and the complete mechanism model. It outputs parameter correction values ​​at fixed periods, dynamically corrects the complete mechanism model, and generates dosing decisions, thereby achieving precise dosing optimization in fluctuating scenarios. It applies a temporal neural network to the parameter correction of the complete mechanism model and uses its gating mechanism to solve the gradient vanishing problem of traditional recurrent neural networks. It accurately captures the long-term temporal characteristics of multi-source data of landfill leachate, and periodically updates parameters and corrects dosing decisions, so that the complete mechanism model can adapt to fluctuating conditions in real time and avoid the decline in treatment effect caused by parameter solidification.

[0103] The technical solution of this invention is as follows: Deploy multiple types of sensors to monitor multi-source data of landfill leachate in real time, generate a multi-source time-series database, perform water quality fluctuation analysis on the landfill leachate based on the multi-source time-series database, and construct a historical fluctuation precursor database to analyze future water quality fluctuation trends of the landfill leachate, determine whether a fluctuation signal is generated, mark fluctuation periods and stable periods based on fluctuation signals, construct kinetic equations for chemical reactions in the landfill leachate based on chemical dosing, couple the kinetic equations with the material balance equations to establish a complete mechanism model, if the current period is stable, set a chemical dosing decision cycle, and periodically generate chemical dosing decisions for the landfill leachate based on real-time monitoring of multi-source data, if the current period is fluctuating, trigger model parameter correction, if model parameter correction is triggered, construct and train a parameter correction model based on a time-series neural network, set a parameter update cycle, use the parameter correction model to periodically correct the model parameters of the complete mechanism model, and dynamically correct the chemical dosing decision.

[0104] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A smart dynamic optimization system for landfill leachate treatment, characterized in that: include: Monitoring module: Deploys multiple types of sensors to monitor multi-source data of landfill leachate in real time and generates a multi-source time-series database; Fluctuation early warning module: Based on a multi-source time series database, it performs water quality fluctuation analysis on landfill leachate, and constructs a historical fluctuation precursor database to analyze the future water quality fluctuation trend of landfill leachate, determine whether a fluctuation signal is generated, and mark the fluctuation period and stable period based on the fluctuation signal; The method for analyzing water quality fluctuations is as follows: Set a flag value and assign it an initial value of 0. Set a sliding fluctuation analysis window for landfill leachate with the current time as the endpoint. Obtain multi-source data within the current sliding fluctuation analysis window from the multi-source time series database. The multi-source data includes raw water quality parameters. Calculate the coefficient of variation of each type of raw water quality parameter within the current sliding fluctuation analysis window. If one or more coefficients of variation are greater than the preset coefficient of variation standard at the current moment, the flag value is incremented once; otherwise, the flag value is reset to 0. If the flag value is greater than or equal to the preset flag value threshold, it is determined that water quality fluctuation has occurred. The historical fluctuation precursor database is obtained as follows: Get the flag value, mark the moment when the flag value is equal to the flag value threshold as the fluctuation moment, set the historical time period with the current time as the end, get all the fluctuation moments within the historical time period, set the buffer period with each fluctuation moment as the end, and extract a fluctuation precursor window with the start of the buffer period as the end for each fluctuation moment. Based on a multi-source time series database, multi-source data within each fluctuation precursor window are obtained, integrated into fluctuation precursor sequences, and summarized to obtain a historical fluctuation precursor database. The method for analyzing future water quality fluctuation trends is as follows: Acquire multi-source data within the current sliding fluctuation analysis window and integrate them according to time series to obtain a precursor analysis sequence. In the historical fluctuation precursor database, calculate the cosine similarity between the precursor analysis sequence and each fluctuation precursor sequence. If the cosine similarity is greater than the preset similarity standard, it is determined that the landfill leachate will experience water quality fluctuations after the buffer period; otherwise, the landfill leachate water quality trend is determined to be stable. Stable dosing module: Constructs kinetic equations for chemical reactions in landfill leachate based on dosing, and couples them with material balance equations to establish a complete mechanism model. If the situation is stable, a dosing decision cycle is set, and dosing decisions are generated periodically for landfill leachate based on real-time multi-source data. If the situation is fluctuating, the model parameters are corrected. Fluctuation dosing module: If model parameter correction is triggered, set the parameter update cycle, build and train a parameter correction model based on a time-series neural network to periodically correct the model parameters of the complete mechanism model, and dynamically correct the dosing decision.

2. The intelligent dosing dynamic optimization system for landfill leachate treatment according to claim 1, characterized in that: The method for marking the fluctuation period and the stable period is as follows: Set a fluctuation anchor point value and assign it an initial value of 0; If water quality fluctuations occur at the current moment, or if it is determined that water quality fluctuations will occur after the buffer period of landfill leachate, a fluctuation signal is generated. If a fluctuation signal is generated and the fluctuation anchor value is 0, the fluctuation anchor value is changed to 1. If no water quality fluctuations occur, and the trend of landfill leachate water quality is determined to be stable, and the fluctuation anchor value is 1, a fluctuation buffer signal is generated. If the duration of continuously generated fluctuation buffer signals exceeds the preset buffer threshold, the fluctuation anchor value is changed to 0. Periods in which the fluctuation anchor value is continuously 1 are marked as fluctuation periods, and periods in which it is continuously 0 are marked as stable periods.

3. The intelligent dosing dynamic optimization system for landfill leachate treatment according to claim 1, characterized in that: The complete mechanism model is established as follows: In the process of landfill leachate treatment, kinetic equations describing the relationship between chemical reaction rates and influencing factors are constructed for chemical reactions based on chemical dosing. The kinetic equations include COD degradation kinetic equations, ammonia nitrogen removal kinetic equations, and total phosphorus removal reaction equations, which are coupled with material balance equations to construct a complete mechanism model for landfill leachate chemical dosing treatment. The coupled equations of the complete mechanism model contain the COD degradation reaction rate constant, ammonia nitrogen reaction rate constant, and total phosphorus reaction rate constant corresponding to the kinetic equations. These are all labeled as model parameters of the complete mechanism model. The initial values ​​of the model parameters are obtained by fitting experimental data.

4. The intelligent dosing dynamic optimization system for landfill leachate treatment according to claim 3, characterized in that: The method for making decisions on the periodic generation and dosing of landfill leachate is as follows: A dosing decision cycle is set for landfill leachate. The starting point of each dosing decision cycle is marked as the decision time. If the current time is the decision time, multi-source data of landfill leachate at the current time is obtained. Combining the multi-source data, the Newton-Raphson iterative method is used to solve the coupled equations of the complete mechanism model. By continuously iterating to approximate the solution of the coupled equations, the dosage of COD treatment, ammonia nitrogen treatment, and total phosphorus treatment is calculated in each iteration, and the dosing vector is obtained. Calculate the Euclidean distance between two dosing vectors obtained in adjacent iterations. If it is less than the preset deviation distance threshold, stop the iteration. The dosing vector obtained in the last iteration is the dosing decision for landfill leachate in the current decision cycle.

5. The intelligent dosing dynamic optimization system for landfill leachate treatment according to claim 1, characterized in that: The construction and training method of the parameter correction model based on the temporal neural network is as follows: All fluctuation periods within the historical time period are obtained. Combined with a multi-source time series database, the multi-source data within each fluctuation period are integrated and normalized according to the time series to obtain a normalized multi-source time series sequence. The sliding window method is used to segment the data to obtain a sample sequence. Label data including the model parameter correction values ​​of the complete mechanism model are added to each sample sequence to obtain a training data set. The parameter correction model consists of an input layer, a hidden layer, and an output layer. The input layer contains 12 feature dimensions, the hidden layer consists of two LSTM layers, both of which use the ReLU function as the activation function, and the output layer has 3 neurons that output the model parameter correction values ​​of the complete mechanism model. The parameter correction model is obtained by training with a training dataset.

6. The intelligent dosing dynamic optimization system for landfill leachate treatment according to claim 5, characterized in that: The method for periodically correcting the model parameters of the complete mechanism model is as follows: Set parameter update cycles for landfill leachate, and mark the start of each parameter update cycle as the update time; At the update time, a time period with the current time as the endpoint and the same duration as the time period corresponding to the sample sequence length is extracted. Multi-source data within the time period is obtained and normalized in the same way as the normalized multi-source time series sequence to obtain the current input sequence input parameter correction model, output the model parameter correction value, and use the model parameter correction value to correct the complete mechanism model.

7. The intelligent chemical dosing dynamic optimization system for landfill leachate treatment according to claim 5, characterized in that: The method for dynamically correcting dosing decisions is as follows: At each update time, multi-source data of landfill leachate is acquired. Combining the multi-source data, the Newton-Raphson iterative method is used to solve the coupled equations of the corrected complete mechanism model. The iterative calculation yields the dosing vectors, including the dosing amounts for COD treatment, ammonia nitrogen treatment, and total phosphorus treatment, which are the dosing decisions for landfill leachate in the current parameter update cycle.

Images