Wastewater Chemical Dosage Decision-Making Method and System Integrating Multi-Source Data

Abstract

This invention discloses a method and system for determining wastewater treatment reagent dosage by integrating multi-source data, relating to the field of wastewater treatment technology. The method includes: acquiring multi-source dynamic data from the wastewater treatment end; tracing effluent deviations and correlating reagent contributions to obtain a first reagent dosing guidance vector; identifying pollution impact trends and predicting reagent demand feedforward to obtain a second reagent dosing guidance vector; acquiring a reagent dosing adjustment space; acquiring a target reagent dosing scheme; acquiring a reagent dosing correction scheme; and executing reagent dosing adjustment at the wastewater treatment end. This invention solves the technical problems of existing wastewater reagent dosing technologies, such as reliance on experience-based adjustment, insufficient dosing accuracy, susceptibility to influent impacts and equipment disturbances leading to unstable effluent and significant reagent waste. It achieves precise, stable, and adaptive adjustment of wastewater treatment reagent dosage, improving effluent quality compliance rates and reducing reagent consumption.

Description

Technical Field

[0001] This invention relates to the field of wastewater treatment technology, specifically to a method and system for determining wastewater reagent dosage by integrating multi-source data. Background Technology

[0002] In the current operation of wastewater treatment processes, the dosage of chemicals is mostly controlled by manual experience or fixed thresholds, making it difficult to dynamically adapt to fluctuations in influent water quality, real-time deviations in effluent, and equipment operating status. Existing dosing decision-making methods lack multi-source data fusion analysis, making it impossible to effectively predict pollution shock loads. At the same time, they do not consider disturbance factors such as response delays of dosing equipment and fluctuations in pipeline pressure, which easily leads to overdosing and waste of chemicals, or underdosing and effluent quality exceeding standards. Overall, they suffer from low dosing accuracy, poor stability, and weak self-adaptability, making it difficult to meet the actual needs of refined and intelligent operation of wastewater treatment.

[0003] Existing technologies for wastewater treatment have problems such as reliance on experience-based adjustments for dosing, insufficient dosing precision, susceptibility to influent impacts and equipment disturbances leading to unstable effluent and significant waste of chemicals. Summary of the Invention

[0004] This application provides a wastewater chemical dosage decision-making method and system that integrates multi-source data, which is used to address the technical problems of existing wastewater chemical dosage that rely on experience-based adjustment, have insufficient dosage accuracy, are susceptible to influent impact and equipment disturbance leading to unstable effluent and serious chemical waste.

[0005] In view of the above problems, this application provides a wastewater treatment dosage decision method and system that integrates multi-source data.

[0006] A first aspect of this application provides a wastewater chemical dosage decision-making method that integrates multi-source data, the method comprising: The process involves acquiring multi-source dynamic data from the wastewater treatment plant, including real-time effluent data, real-time influent data, process operation data, and equipment operation data. Based on the process operation data, the process data is used to trace effluent deviations and correlate chemical contributions in the real-time effluent data to obtain a first chemical dosing guidance vector. Based on the real-time influent data, pollution impact trends are identified and chemical demand is predicted forward to obtain a second chemical dosing guidance vector. Big data mining is performed on the first and second chemical dosing guidance vectors to obtain a chemical dosing adjustment space. A chemical action scoring model is used to dynamically optimize the chemical dosing adjustment space to obtain a target chemical dosing scheme. Based on the equipment operation data, the target chemical dosing scheme is corrected for response offset and disturbances are suppressed to obtain a chemical dosing correction scheme. Finally, the chemical dosing adjustment at the wastewater treatment plant is executed according to the chemical dosing correction scheme.

[0007] A second aspect of this application provides a wastewater chemical dosage decision system that integrates multi-source data, the system comprising: A multi-source dynamic data acquisition module is used to acquire multi-source dynamic data from the wastewater treatment end, including real-time effluent data, real-time influent data, process operation data, and equipment operation data; a first vector acquisition module is used to trace effluent deviations and correlate chemical contributions based on the process operation data to obtain a first vector for chemical dosing guidance; a second vector acquisition module is used to identify pollution impact trends and predict chemical demand based on the real-time influent data to obtain a second vector for chemical dosing guidance; a big data mining module is used to... The first and second drug dosing guidance vectors are subjected to big data mining to obtain the drug dosing adjustment space; the target drug dosing scheme acquisition module is used to perform dynamic neighborhood optimization on the drug dosing adjustment space through a drug action scoring model to obtain the target drug dosing scheme; the drug dosing correction scheme acquisition module is used to perform response offset correction and disturbance suppression on the target drug dosing scheme based on the equipment operation data to obtain the drug dosing correction scheme, and to perform drug dosing adjustment at the wastewater treatment end according to the drug dosing correction scheme.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages: This process involves acquiring multi-source dynamic data from the wastewater treatment plant; performing effluent deviation tracing and chemical contribution correlation to obtain a first chemical dosing guidance vector; identifying pollution impact trends and predicting chemical demand based on real-time influent data to obtain a second chemical dosing guidance vector; performing big data mining based on the first and second chemical dosing guidance vectors to obtain the chemical dosing adjustment space; performing dynamic neighborhood optimization to obtain a target chemical dosing scheme; correcting response offset and suppressing disturbances in the target chemical dosing scheme based on equipment operation data to obtain a chemical dosing correction scheme; and adjusting the chemical dosing at the wastewater treatment plant according to the correction scheme. This achieves precise, stable, and adaptive adjustment of wastewater treatment chemical dosing, improving effluent quality compliance rates and reducing chemical consumption. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1This is a schematic flowchart of a wastewater reagent dosage decision-making method that integrates multi-source data, provided in an embodiment of this application. Figure 2 A schematic diagram of the wastewater reagent dosage decision system that integrates multi-source data, provided in an embodiment of this application.

[0011] Figure labeling: Multi-source dynamic data acquisition module 10, first vector acquisition module 20, second vector acquisition module 30, big data mining module 40, target agent dosing scheme acquisition module 50, agent dosing correction scheme acquisition module 60. Detailed Implementation

[0012] This application provides a wastewater chemical dosage decision-making method and system that integrates multi-source data, which is used to address the technical problems of existing wastewater chemical dosage that rely on experience-based adjustment, have insufficient dosage accuracy, are susceptible to influent impact and equipment disturbance leading to unstable effluent and serious chemical waste.

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0014] Example 1, as Figure 1 As shown, this application provides a wastewater treatment dosage decision method that integrates multi-source data, the method comprising: Step S100: Obtain multi-source dynamic data from the wastewater treatment end, including real-time effluent data, real-time influent data, process operation data, and equipment operation data.

[0015] Specifically, multi-dimensional, real-time updated, multi-source dynamic data is collected and acquired from the wastewater treatment site. This data includes real-time effluent data to characterize the treated water quality compliance status, real-time influent data to reflect the characteristics of raw water quality and pollution load, process operation data to reflect the biochemical treatment process and reaction conditions, and equipment operation data to characterize the working status of actuators such as dosing pumps, valves, and pipelines, providing a complete data foundation for subsequent chemical dosing decisions.

[0016] Step S200: Based on the process operation data, trace the effluent deviation and correlate the chemical contribution of the real-time effluent data to obtain the first chemical dosing guidance vector.

[0017] Specifically, real-time effluent data is compared with preset effluent targets to obtain current effluent deviation data and identify the corresponding pollutants exceeding the standard. Based on process operation data, a time-lag correlation is established between historical effluent deviations and historical reagent dosing behavior. Based on this time-lag correlation, the historical reagent action range corresponding to the current effluent deviation data is traced. The contribution of multiple reagents to pollutants exceeding the standard is identified. Combining the pollutants exceeding the standard with the current effluent deviation data, the first reagent dosing guidance vector is finally generated.

[0018] Step S300: Based on the real-time influent data, identify the pollution impact trend and predict the reagent demand in advance to obtain the second vector for reagent dosing guidance.

[0019] Specifically, the pollution load changes are analyzed based on real-time influent data to generate pollution shock trend identification results. The identification results are then verified by combining influent water quality disturbance characteristics to confirm the pollution shock status. Based on this, the pollution load changes in the future reaction cycle are predicted. Then, based on the prediction results, the reagent demand is predicted and a feedforward reagent demand sequence is obtained. Finally, the pollution shock confirmation results, pollution load change results and feedforward reagent demand sequence are integrated to generate a second reagent dosing guidance vector.

[0020] Step S400: Perform big data mining based on the first drug dosing guidance vector and the second drug dosing guidance vector to obtain the drug dosing adjustment space.

[0021] Specifically, the first and second guidance vectors for pesticide dosing are preprocessed by feature concatenation, missing value imputation, and normalization. The Extreme Gradient Boosting (XGBoost) algorithm is then used for big data mining. This algorithm uses an additive ensemble model as its core structure, composed of multiple regression decision trees. Each tree learns the residuals of the preceding model and controls overfitting through regularization, resulting in high accuracy and strong generalization ability. During the training phase, the two guidance vectors from historical operating conditions are used as input features, and the historically optimal pesticide dosing scheme is used as the label to complete model fitting and validation. During the inference phase, the real-time fused guidance vectors are input into the trained XGBoost model. Multiple candidate dosing parameters are output through weighted summation of the leaf nodes of multiple decision trees. These parameters are then filtered based on effluent compliance constraints, pesticide dosing limits, and process safety boundaries, ultimately generating a set of multiple pesticide dosing adjustment schemes that meet the operating conditions, forming a pesticide dosing adjustment space that can be used for subsequent optimization.

[0022] Step S500: The target drug dosing scheme is obtained by dynamically optimizing the drug dosing adjustment space through the drug action scoring model.

[0023] Specifically, a chemical action scoring model is used to perform multiple rounds of dynamic neighborhood competition optimization in the chemical dosing adjustment space. First, the future effluent is predicted for each dosing scheme within the adjustment space. The predicted sequence is input into the model to obtain the chemical action score. The first chemical dosing candidate domain is selected according to the preset score constraints and iteratively optimized to obtain the first potential scheme. Then, based on the first candidate domain, the trend of the dosing parameters is identified, and the search domain and neighborhood range are established. After evaluation and selection, the second candidate domain and the optimal scheme are obtained. The second potential scheme is generated through competition with the first potential scheme. This iterative logic is followed to continuously perform neighborhood screening, scoring evaluation, and competitive optimization until the preset number of optimizations is reached. Finally, the target chemical dosing scheme with the best comprehensive effect, which meets the requirements of effluent compliance and chemical efficiency, is output.

[0024] Step S600: Based on the equipment operation data, perform response offset correction and disturbance suppression on the target reagent dosing scheme to obtain a reagent dosing correction scheme, and execute the reagent dosing adjustment at the wastewater treatment end according to the reagent dosing correction scheme.

[0025] Specifically, based on equipment operation data, features such as dosing pump response delay, valve opening deviation, chemical pipeline pressure fluctuation, chemical reserve status, and actual dosing flow rate are extracted. Based on these features, the timing and flow deviations of the target chemical dosing scheme in actual execution are predicted. At the same time, chemical dosing disturbance characteristics are identified. Then, based on the response deviation and disturbance information, the target chemical dosing scheme is collaboratively optimized and corrected, and on-site disturbances are suppressed. Finally, a chemical dosing correction scheme adapted to the on-site equipment operating conditions is generated, and the precise adjustment of chemical dosing at the wastewater treatment end is automatically completed according to the correction scheme.

[0026] In one possible implementation, step S200 further includes: Step S210: Compare the real-time effluent data with the preset effluent target to obtain the current effluent deviation data, and determine the pollutants exceeding the standard based on the current effluent deviation data.

[0027] Step S220: Based on the process operation data, establish a time-delay correlation between multiple historical effluent deviations and multiple historical reagent dosing behaviors.

[0028] Step S230: Based on the time lag correlation, trace back the historical reagent action range corresponding to the current effluent deviation data.

[0029] Step S240: Identify the contribution of the pollutants exceeding the standard based on the historical reagent action range, obtain the multi-reagent contribution identification results, and generate the first reagent dosing guidance vector by combining the pollutants exceeding the standard and the current effluent deviation data.

[0030] Specifically, the real-time collected effluent water quality data is compared with the preset effluent standard target value item by item, and the difference between the actual monitored value and the standard limit is calculated to obtain the current effluent deviation data. Then, the deviation data of each water quality indicator is judged by threshold. Water quality indicators whose deviation value exceeds the preset allowable range and causes the effluent to fail to meet the standard are identified as the pollutants exceeding the standard under the current operating conditions.

[0031] First, the process operation data is standardized and timestamped. The historical dosage, duration, and type of reagents are arranged on a unified time axis with the historical effluent deviation data within the corresponding time lag period. Then, the correlation between the dosage behavior and the effluent deviation under different time delays is calculated using the time-series cross-correlation function CCF to determine the inherent reaction time lag of the wastewater treatment system. Subsequently, using this time lag as a window, the matching dosage data and deviation data are extracted by a sliding time window. The dynamic influence relationship between multiple sets of historical dosage behaviors and historical effluent deviations is fitted using a vector autoregression model (VAR) to finally establish a quantifiable and traceable time lag correlation.

[0032] Based on the time of occurrence of the current effluent deviation, and according to the inherent reaction time delay of the system determined by the established time delay correlation, the start and end points of the corresponding time window are calculated backward and backward to locate the historical time period that will have a lagging impact on the current effluent quality. Then, based on the process operation data and time sequence markers, all reagent dosing data within this time window are extracted from the historical dosing records, so as to accurately trace the historical reagent action interval that is causally corresponding to the current effluent deviation.

[0033] Within the established historical pesticide application range, a random forest regression algorithm is used to identify the contribution of multiple pesticides to pollutants exceeding the standard. This algorithm is an ensemble learning structure, consisting of multiple independent regression decision trees in parallel. Bootstrap sampling is used to construct the training dataset for each decision tree, and a feature randomization strategy is employed to reduce the risk of overfitting in individual trees. Finally, the average of the prediction results from multiple decision trees is used to output the final conclusion. During application, dosage and frequency of various pesticides within the historical pesticide application range are extracted as input features for the algorithm, and the contribution of pollutants exceeding the standard is then considered. The concentration change within this range is used as the algorithm's prediction target. Historical data within the range is loaded to complete the algorithm's training and validation. Then, the trained random forest regression algorithm is used to calculate the feature importance value corresponding to each agent's dosing parameter, thereby quantifying the contribution ratio of each agent to the removal effect of the pollutants exceeding the standard, and obtaining the multi-agent contribution identification result. Finally, the type of pollutant exceeding the standard, the concentration exceeding the standard value, the current effluent deviation data, and the contribution quantification results of each agent are normalized and feature vectors are concatenated to finally generate the first agent dosing guidance vector containing complete control guidance information.

[0034] In one possible implementation, step S300 further includes: Step S310: Analyze the pollution load changes based on the real-time influent data to generate pollution impact trend identification results.

[0035] Step S320: Determine the influent water quality disturbance characteristics based on the real-time influent water data, and perform disturbance verification on the pollution impact trend identification results based on the influent water quality disturbance characteristics to obtain the pollution impact confirmation results.

[0036] Step S330: Based on the pollution shock confirmation results, predict the pollution load changes within the future response cycle.

[0037] Step S340: Based on the pollution load change results, perform feedforward prediction of reagent demand, obtain the feedforward reagent demand sequence, and generate the second reagent dosing guidance vector by combining the pollution shock confirmation results and the pollution load change results.

[0038] Specifically, the influent flow rate and pollutant concentration indicators in the real-time influent data are sampled synchronously over time. The instantaneous pollution load is calculated in real time by multiplying the flow rate and concentration. The pollution load sequence is denoised by using moving average filtering. Then, the load rise rate, abrupt change amplitude and duration are identified by combining slope calculation and differential analysis. In this way, the dynamic change law of pollution load is analyzed, the type and strength of influent impact are determined, and the pollution impact trend identification result is generated.

[0039] The influent water quality disturbance characteristics, such as the amplitude of sudden changes in pollutant concentration, the frequency of flow fluctuations, and the duration of deviations from normal thresholds, are extracted from real-time influent data to construct a quantitative index of disturbance intensity. Then, a sliding window variance test and outlier identification are used to verify the disturbance. Load sudden changes caused by short-term random fluctuations are identified as noise disturbances and are removed, while stable impact trends that last for more than a preset duration and whose disturbance intensity reaches the threshold are retained. By distinguishing between instantaneous random disturbances and continuous water quality impacts, the preliminary identification results are corrected, and finally, accurate pollution impact confirmation results are obtained.

[0040] Based on the pollution impact confirmation results, a Long Short-Term Memory (LSTM) network is used for time series prediction. This algorithm consists of an input layer, a memory cell layer containing a forget gate, an input gate, and an output gate, and an output layer, which can effectively capture the long-term dependencies of time series data. In practice, historical pollution load time series data are first constructed as time series samples with pollution impact intensity and duration characteristics. After normalization, the samples are input into the LSTM model for training. Then, the current pollution impact confirmation results are input into the trained model. Combined with the fixed reaction cycle of the wastewater treatment system, effective time series information is filtered through a gating mechanism to infer the load change trend, and finally, the pollution load change results within the corresponding future reaction cycle are output.

[0041] A multilayer perceptron (MLP) is used for feedforward prediction of pesticide demand. The algorithm consists of an input layer, multiple hidden layers, and an output layer. It fits a nonlinear mapping relationship through a fully connected structure. When in use, the pollution load change results are used as input features. The model is trained using historical pollution load-pesticide dosage corresponding samples. The pollution load of each future time period is input into the trained model and the time-by-time pesticide dosage is output to obtain the feedforward pesticide demand sequence. Then, the pollution impact confirmation results, pollution load change results, and feedforward pesticide demand sequence are normalized and fused to generate a second vector for pesticide dosage guidance.

[0042] In one possible implementation, step S500 further includes: Step S510: Perform optimization analysis on the drug addition adjustment space according to the drug action scoring model to obtain the first drug addition candidate domain and the first drug addition potential scheme.

[0043] Step S520: Based on the drug action scoring model, perform neighborhood competition optimization on the first drug dosing potential scheme according to the first drug dosing candidate domain to obtain the second drug dosing candidate domain and the second drug dosing potential scheme.

[0044] Step S530: Based on the drug action scoring model, perform neighborhood competition optimization on the second drug administration potential scheme according to the second drug administration candidate domain to obtain the third drug administration candidate domain and the third drug administration potential scheme.

[0045] Step S540: Based on the drug action scoring model, continue to perform neighborhood competition optimization on the potential third drug dosing scheme according to the third drug dosing candidate domain until the target drug dosing scheme that satisfies the predetermined number of neighborhood competition optimizations is obtained.

[0046] Specifically, for each chemical dosing adjustment scheme within the chemical dosing adjustment space, the future effluent effect is predicted sequentially to obtain multiple sets of effluent prediction sequences. All effluent prediction sequences are input into a pre-constructed chemical action scoring model to calculate multiple chemical action scores for each scheme. By comparing each score with preset scoring constraints, multiple score verification results are obtained. Based on the score verification results, schemes within the chemical dosing adjustment space are screened and eliminated to form the first chemical dosing candidate domain. Then, the chemical action scores of the schemes within the first chemical dosing candidate domain are iteratively optimized to select the scheme with the highest chemical action score as the first potential chemical dosing scheme.

[0047] Based on a pre-constructed reagent action scoring model, the dosing parameter trends of all schemes within the first reagent dosing candidate domain are first identified, and the changing patterns and correlation characteristics of each dosing parameter are analyzed. Based on this, a first reagent dosing search domain covering a reasonable parameter range is established. Then, based on this first reagent dosing search domain and combined with the wastewater treatment operating conditions, reagent dosing adjustment decisions are made, delineating a first reagent dosing neighborhood centered on the first reagent dosing potential schemes and covering a reasonable parameter fluctuation range. Subsequently, all dosing schemes within the first reagent dosing neighborhood are input into the reagent action scoring model, combined with the pre-constructed model... A scoring constraint is applied to comprehensively evaluate and screen each scheme, eliminating schemes that do not meet the scoring requirements, and generating a second candidate domain for drug administration. The schemes within the second candidate domain are then iteratively optimized through drug action scoring to determine the scheme with the highest drug action score within the second candidate domain as the optimal scheme. Finally, the first potential drug administration scheme and the optimal scheme in the second candidate domain are compared and competed on drug action scores, and the scheme with the highest drug action score among the two is selected as the second potential drug administration scheme.

[0048] Based on the agent effect scoring model, the parameter change trends of each dosing scheme within the second agent dosing candidate domain are first identified to construct the second agent dosing search domain. Based on this search domain and combined with the field conditions, a second agent dosing neighborhood is defined, with the potential second agent dosing scheme as the core. Using the agent effect scoring model and according to predetermined scoring constraints, all schemes within the second agent dosing neighborhood are comprehensively evaluated and screened to obtain the third agent dosing candidate domain. Iterative optimization of agent effect scoring is carried out within the third agent dosing candidate domain, and the scheme corresponding to the maximum agent effect score within the third agent dosing candidate domain is determined as the optimal scheme in the third candidate domain. Then, the potential second agent dosing scheme and the optimal scheme in the third candidate domain are compared and competed on agent effect scores, and the scheme with the highest agent effect score among the two is selected as the potential third agent dosing scheme.

[0049] Based on the drug action scoring model, the dosing parameters within the candidate domain for the third drug dosing are continuously trend-identified, and a corresponding dosing search domain is constructed. A new round of drug dosing neighborhoods is defined around the potential solutions for the third drug dosing. The solutions within the neighborhoods are evaluated and screened using the drug action scoring model combined with predetermined scoring constraints. The candidate domains are iteratively updated, and scoring optimization is performed within the candidate domains. The optimal solution obtained from the new optimization is compared with the potential solutions from the previous round. The potential solutions are continuously updated, and the above neighborhood competition optimization process is repeated until the number of iterations reaches the preset number of neighborhood competition optimizations. Finally, the target drug dosing solution with the best overall performance is output.

[0050] In one possible implementation, step S510 further includes: Step S511: Based on each reagent dosing adjustment scheme within the reagent dosing adjustment space, predict the future effluent and obtain multiple effluent prediction sequences.

[0051] Step S512: Input the multiple effluent prediction sequences into the drug action scoring model to obtain multiple drug action scores.

[0052] Step S513: Determine whether the multiple drug effect scores meet the predetermined scoring constraints, and obtain multiple score verification results.

[0053] Step S514: Based on the multiple scoring test results, filter the drug dosing adjustment space to generate the first drug dosing candidate domain.

[0054] Step S515: Perform iterative optimization of the drug action score for the first drug dosing candidate domain to generate the first drug dosing potential scheme.

[0055] Specifically, for each dosing regulation scheme within the dosing space, a temporal convolutional network (TCN) is used to predict future effluent. This algorithm consists of causal convolutional layers, dilated convolutional layers, residual connections, and activation functions. By expanding the receptive field through dilated convolution, it can accurately capture the temporal correlation between dosing and effluent water quality changes, effectively avoiding the gradient vanishing problem. In practice, the dosing parameters corresponding to each dosing regulation scheme, namely dosing amount, dosing frequency, dosing time sequence, are first fused with historical influent water quality and process operation data to construct a temporal input sample and perform normalization processing. The processed sample is then input into the trained TCN model. Through the model's convolution and residual propagation mechanisms, the changes in effluent water quality indicators in multiple future time periods under the action of the dosing scheme are deduced. Finally, a corresponding effluent prediction sequence is output for each dosing regulation scheme, and multiple effluent prediction sequences are obtained by summarizing them.

[0056] A gradient boosting tree (GBDT) model was constructed to evaluate the effectiveness of chemical agents. This model consists of multiple regression decision trees arranged in a sequential iterative structure. The prediction accuracy is improved by continuously fitting the residuals through forward step-by-step addition. During training, historical chemical dosing parameters, corresponding effluent prediction sequences, actual effluent compliance rates, chemical costs, and process stability are used as multi-dimensional features, and manually calibrated comprehensive chemical agent effectiveness scores are used as labels. Feature normalization, tree depth and learning rate parameter tuning, and model convergence training are completed. The obtained multiple effluent prediction sequences are used as model inputs. Through layer-by-layer residual fitting and feature weighting calculation of the GBDT model, a corresponding quantitative chemical agent effectiveness score is output for each chemical dosing adjustment scheme, resulting in multiple complete chemical agent effectiveness scores.

[0057] The obtained multiple agent action scores are compared one by one with the preset scoring thresholds, lower limits of compliance, upper limits of process constraints, etc., to determine whether each score meets the effluent compliance requirements, agent cost limits, dosing stability conditions and process safety boundaries. A binary judgment of qualified or unqualified is formed for each agent action score, and finally multiple score test results corresponding to each dosing scheme are generated.

[0058] Based on the multiple scoring test results obtained, all drug dosing adjustment schemes within the drug dosing adjustment space are reviewed one by one. All dosing schemes that are deemed qualified and meet the predetermined scoring constraints are retained, while dosing schemes that are unqualified or do not meet the constraints are directly eliminated. Through this screening rule of retaining qualified schemes and eliminating unqualified schemes, a first drug dosing candidate domain composed of high-quality and feasible schemes is formed from the original drug dosing adjustment space.

[0059] Within the first candidate domain for drug administration, an iterative ranking and optimization process is conducted with the drug effect score as the optimization objective. First, the drug effect scores of all qualified schemes within the domain are ranked from high to low. The scheme with the highest current score is selected as the initial optimal solution. Then, small-scale parameter fine-tuning is performed around this optimal solution, and the score is recalculated. The scores of the new and old schemes are continuously compared, and higher-scoring schemes are retained while lower-scoring schemes are eliminated. This iterative comparison process is repeated until the score no longer improves. Finally, the first potential drug administration scheme with the highest comprehensive score and the best control effect is output.

[0060] In one possible implementation, step S520 further includes: Step S521: Based on the first drug dosing candidate domain, identify the trend of dosing parameters and establish the first drug dosing search domain.

[0061] Step S522: Make drug dosing adjustment decisions based on the first drug dosing search domain and establish the first drug dosing neighborhood.

[0062] Step S523: Based on the drug action scoring model, evaluate and screen the first drug dosing neighborhood according to the predetermined scoring constraints to obtain the second drug dosing candidate domain.

[0063] Step S524: Perform iterative optimization of the drug action score for the second drug dosing candidate domain to obtain the optimal solution for the second candidate domain.

[0064] Step S525: Compete to find the best drug action score between the first potential drug administration scheme and the second candidate domain optimal scheme to generate the second potential drug administration scheme.

[0065] Specifically, statistical analysis is performed on parameters such as dosage, proportion, and frequency of all qualified schemes within the first agent dosing candidate domain. By calculating the mean, variance, distribution range, and slope of the parameters, the trend of the dosing parameters is identified, and the reasonable fluctuation range, optimal value direction, and constraint boundary of each parameter are determined. In this way, a first agent dosing search domain containing all feasible parameter combinations is constructed.

[0066] Using the potential first agent dosing scheme as the core benchmark, and combining the feasible parameter boundaries of the first agent dosing search domain, the agent dosing adjustment decisions are made in small steps and within a controllable range for parameters such as dosing amount, dosing ratio, and dosing sequence. Under the premise of not exceeding the search domain constraints, neighboring dosing combinations with small parameter fluctuations are generated around the benchmark scheme, thereby defining the local fine search range and establishing the first agent dosing neighborhood with the optimal potential scheme as the center and the parameter changes being stable.

[0067] All reagent dosing schemes within the first reagent dosing neighborhood are sequentially input into the reagent effect scoring model to calculate the corresponding reagent effect score. Then, according to predetermined scoring constraints such as effluent compliance, reagent cost, and process stability, each scheme is judged to see if it meets the requirements. Only dosing schemes with qualified scores and that meet all constraints are retained, while unqualified schemes are eliminated. Finally, these qualified neighboring schemes form the second reagent dosing candidate domain.

[0068] The agent effect scores of all qualified schemes in the second agent addition candidate domain are sorted from high to low. The maximum agent effect score is used as the optimization target. Through multiple iterations of comparison, screening and updating, high-scoring schemes are continuously retained and low-scoring schemes are eliminated until the addition parameter combination with the highest score and the best overall effect in the domain is found. Finally, the optimal scheme of the second candidate domain is determined and output.

[0069] The drug action scores of the first potential drug administration scheme and the optimal scheme in the second candidate domain are directly compared. The drug action scores of the two schemes are read respectively, and the scheme with the higher score and better overall effect is selected as the winner. This is used to generate the updated second potential drug administration scheme.

[0070] In one possible implementation, step S600 further includes: Step S610: Extract the dosing pump response delay characteristics, valve opening deviation characteristics, chemical pipeline pressure fluctuation characteristics, chemical remaining quantity status characteristics, and actual dosing flow characteristics based on the equipment operation data.

[0071] Step S620: Based on the response delay characteristics of the dosing pump and the valve opening deviation characteristics, predict the dosing offset characteristics of the target dosing scheme.

[0072] Step S630: Based on the pressure fluctuation characteristics of the reagent pipeline, the remaining reagent status characteristics, and the actual dosing flow characteristics, the execution disturbance identification of the target reagent dosing scheme is performed to obtain the reagent dosing disturbance characteristics.

[0073] Step S640: Based on the drug dosing offset characteristics and the drug dosing perturbation characteristics, the target drug dosing scheme is jointly optimized to generate the drug dosing correction scheme.

[0074] Specifically, the collected equipment operation data is analyzed and quantitatively extracted. The response delay characteristics of the dosing pump from the issuance of the command to the actual action are calculated from the runtime sequence data. The valve opening deviation characteristics between the set opening and the actual opening are calculated from the valve control data. The pressure fluctuation characteristics of the chemical pipeline as the pressure changes over time are extracted from the pipeline monitoring data. The remaining amount and consumption status characteristics of the chemical are obtained from the chemical storage device data. The actual output chemical dosing flow characteristics are extracted from the flow meter data. All five types of equipment characteristics are accurately extracted.

[0075] A timing deviation fitting algorithm is used to quantify the response delay characteristics of the dosing pump and the valve opening deviation characteristics. First, the time difference between the actual start time and the set time of the target agent dosing scheme is estimated based on the response delay characteristics to obtain the dosing timing offset. Then, the error relationship between the set dosing flow rate and the actual output flow rate is fitted based on the valve opening deviation characteristics to calculate the dosing flow rate offset. The two types of offsets are fused to finally predict the complete agent dosing offset characteristics of the target agent dosing scheme.

[0076] Based on the characteristics of pressure fluctuations in the reagent pipeline, the characteristics of the reagent balance, and the characteristics of the actual dosing flow rate, a multi-dimensional disturbance matching and anomaly identification is performed on the field execution process of the target reagent dosing scheme. By comparing the pressure fluctuation amplitude, the rate of balance decrease, and the deviation between the actual flow rate and the set flow rate, it is determined whether there are execution disturbances such as pipeline blockage, insufficient reagent, and unstable flow. The identified disturbance types, disturbance intensity, and disturbance impact range are quantitatively characterized, and finally, reagent dosing disturbance characteristics that can be used for correction and compensation are obtained.

[0077] Based on the characteristics of reagent dosing offset and reagent dosing disturbance, a collaborative optimization is carried out. First, the dosing time and theoretical dosing amount of the target reagent dosing scheme are pre-compensated according to the time offset and flow offset. Then, combined with disturbance information such as pressure fluctuation, reagent balance, and flow anomaly, the dosing acceleration rate and valve opening are dynamically adjusted to suppress execution interference. The time correction, flow compensation and disturbance suppression are executed simultaneously to eliminate equipment response deviation and on-site execution error, and finally form a reagent dosing correction scheme that is adapted to actual working conditions and can be executed stably and accurately.

[0078] In one possible implementation, step S640 further includes: The drug dosing offset features include dosing timing offset and dosing flow rate offset.

[0079] Specifically, the agent dosing deviation features include dosing timing deviation and dosing flow rate deviation. The dosing timing deviation is used to quantify the time difference between the dosing start time and the set time caused by the response delay of the dosing pump. The dosing flow rate deviation is used to quantify the flow rate error between the actual dosing flow rate and the set dosing flow rate caused by the valve opening deviation. The two types of deviations together completely characterize the timing and flow rate deviations of the target agent dosing scheme in actual execution.

[0080] In one possible implementation, step S210 further includes: Based on the current water discharge deviation data, a water discharge deviation early warning signal is generated.

[0081] Specifically, based on the current effluent deviation data obtained in real time, the deviation value is compared with the preset warning threshold. When the deviation exceeds the normal allowable range, the warning judgment is triggered immediately. The effluent deviation warning signal, which includes the warning level, deviation index and abnormal time, is generated by combining the deviation type, the degree of exceeding the standard and the impact level. This is used to remind the system and operators to carry out timely dosing adjustment and process intervention.

[0082] Example 2, based on the same inventive concept as the wastewater reagent dosage decision method integrating multi-source data in the aforementioned examples, such as... Figure 2 As shown, this application provides a wastewater treatment dosage decision system that integrates multi-source data. The system and method embodiments in this application are based on the same inventive concept. The system includes: The multi-source dynamic data acquisition module 10 is used to acquire multi-source dynamic data from the wastewater treatment end. The multi-source dynamic data includes real-time effluent data, real-time influent data, process operation data, and equipment operation data.

[0083] The first vector acquisition module 20 is used to trace the effluent deviation and associate the chemical contribution with the real-time effluent data based on the process operation data, and to obtain the first chemical dosing guidance vector.

[0084] The second vector acquisition module 30 is used to identify pollution impact trends and predict reagent demand based on the real-time influent data, and to acquire a second reagent dosing guidance vector.

[0085] The big data mining module 40 is used to perform big data mining based on the first drug dosing guidance vector and the second drug dosing guidance vector to obtain the drug dosing adjustment space.

[0086] The target drug dosing scheme acquisition module 50 is used to obtain the target drug dosing scheme by dynamically optimizing the drug dosing adjustment space through a drug action scoring model.

[0087] The reagent dosing correction scheme acquisition module 60 is used to perform response offset correction and disturbance suppression on the target reagent dosing scheme based on the equipment operation data, acquire the reagent dosing correction scheme, and execute the reagent dosing adjustment at the wastewater treatment end according to the reagent dosing correction scheme.

[0088] Furthermore, the system is also used to implement the following functions: The real-time effluent data is compared with the preset effluent target to obtain the current effluent deviation data, and the pollutants exceeding the standard are determined based on the current effluent deviation data. Based on the process operation data, a time-lag correlation is established between multiple historical effluent deviations and multiple historical reagent dosing behaviors. The historical reagent action interval corresponding to the current effluent deviation data is traced according to the time-lag correlation. The contribution of the pollutants exceeding the standard is identified according to the historical reagent action interval, and the multi-reagent contribution identification results are obtained. The first reagent dosing guidance vector is generated by combining the pollutants exceeding the standard and the current effluent deviation data.

[0089] Furthermore, the system is also used to implement the following functions: The pollution load change is analyzed based on the real-time influent data to generate a pollution shock trend identification result; the influent water quality disturbance characteristics are determined based on the real-time influent data, and the pollution shock trend identification result is perturbed based on the influent water quality disturbance characteristics to obtain a pollution shock confirmation result; the pollution load change result within the future reaction cycle is predicted based on the pollution shock confirmation result; the reagent demand is predicted based on the pollution load change result to obtain a feedforward reagent demand sequence, and the second reagent dosing guidance vector is generated by combining the pollution shock confirmation result and the pollution load change result.

[0090] Furthermore, the system is also used to implement the following functions: The drug dosing adjustment space is optimized according to the drug action scoring model to obtain a first drug dosing candidate domain and a first drug dosing potential scheme. Based on the drug action scoring model, the first drug dosing potential scheme is optimized through neighborhood competition based on the first drug dosing candidate domain to obtain a second drug dosing candidate domain and a second drug dosing potential scheme. Based on the drug action scoring model, the second drug dosing potential scheme is optimized through neighborhood competition based on the second drug dosing candidate domain to obtain a third drug dosing candidate domain and a third drug dosing potential scheme. Based on the drug action scoring model, the third drug dosing potential scheme is optimized through neighborhood competition based on the third drug dosing candidate domain until the target drug dosing scheme that satisfies a predetermined number of neighborhood competition optimizations is obtained.

[0091] Furthermore, the system is also used to implement the following functions: Based on each dosing regulation scheme within the dosing regulation space, future effluent is predicted to obtain multiple effluent prediction sequences; these multiple effluent prediction sequences are input into the dosing regulation scoring model to obtain multiple dosing regulation scores; it is determined whether the multiple dosing regulation scores meet predetermined scoring constraints to obtain multiple scoring verification results; the dosing regulation space is filtered based on the multiple scoring verification results to generate the first dosing regulation candidate domain; iterative optimization of the dosing regulation scores is performed on the first dosing regulation candidate domain to generate the first dosing regulation potential scheme.

[0092] Furthermore, the system is also used to implement the following functions: Based on the first candidate domain for drug administration, trend identification of administration parameters is performed to establish a first drug administration search domain; based on the first drug administration search domain, drug administration adjustment decisions are made to establish a first drug administration neighborhood; based on the drug action scoring model, the first drug administration neighborhood is evaluated and screened according to the predetermined scoring constraints to obtain a second candidate domain for drug administration; the drug action scoring of the second candidate domain is iteratively optimized to obtain the optimal solution of the second candidate domain; the drug action scoring of the first potential solution for drug administration and the optimal solution of the second candidate domain are competitively optimized to generate the second potential solution for drug administration.

[0093] Furthermore, the system is also used to implement the following functions: Based on the equipment operation data, extract the dosing pump response delay characteristics, valve opening deviation characteristics, chemical pipeline pressure fluctuation characteristics, chemical remaining quantity status characteristics, and actual dosing flow characteristics; based on the dosing pump response delay characteristics and the valve opening deviation characteristics, predict the chemical dosing offset characteristics of the target chemical dosing scheme; based on the chemical pipeline pressure fluctuation characteristics, the chemical remaining quantity status characteristics, and the actual dosing flow characteristics, identify the execution disturbance of the target chemical dosing scheme to obtain the chemical dosing disturbance characteristics; based on the chemical dosing offset characteristics and the chemical dosing disturbance characteristics, perform collaborative optimization of the target chemical dosing scheme to generate the chemical dosing correction scheme.

[0094] Furthermore, the system is also used to implement the following functions: The drug dosing offset features include dosing timing offset and dosing flow rate offset.

[0095] Furthermore, the system is also used to implement the following functions: Based on the current water discharge deviation data, a water discharge deviation early warning signal is generated.

[0096] It should be noted that the order of the embodiments described above is for descriptive purposes only and does not represent the superiority or inferiority of the embodiments. Specific embodiments of this specification have been described above. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0097] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0098] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A wastewater treatment dosage decision method integrating multi-source data, characterized in that, The method includes: Acquire multi-source dynamic data from the wastewater treatment end, including real-time effluent data, real-time influent data, process operation data, and equipment operation data; Based on the process operation data, the real-time effluent data is traced for effluent deviation and correlated with the contribution of reagents to obtain the first vector for reagent dosing guidance. Based on the real-time influent data, pollution impact trend identification and reagent demand feedforward prediction are performed to obtain the second vector for reagent dosing guidance. Big data mining is performed based on the first and second drug dosing guidance vectors to obtain the drug dosing adjustment space; The target drug dosing scheme is obtained by dynamically optimizing the drug dosing adjustment space using a drug action scoring model. Based on the equipment operation data, the target reagent dosing scheme is corrected for response offset and disturbance is suppressed to obtain a reagent dosing correction scheme, and the reagent dosing adjustment at the wastewater treatment end is executed according to the reagent dosing correction scheme.

2. The wastewater treatment dosage decision method integrating multi-source data as described in claim 1, characterized in that, Based on the process operation data, the real-time effluent data is analyzed for effluent deviation tracing and chemical contribution correlation to obtain a first chemical dosing guidance vector, including: The real-time effluent data is compared with the preset effluent target to obtain the current effluent deviation data, and the pollutants exceeding the standard are determined based on the current effluent deviation data. Based on the process operation data, a time-delay correlation relationship is established between multiple historical effluent deviations and multiple historical reagent dosing behaviors; Based on the time lag correlation, the historical reagent action range corresponding to the current effluent deviation data is traced; Based on the historical reagent action range, the contribution of the pollutants exceeding the standard is identified, and the multi-reagent contribution identification results are obtained. Combining the pollutants exceeding the standard and the current effluent deviation data, the first reagent dosing guidance vector is generated.

3. The wastewater treatment dosage decision method integrating multi-source data as described in claim 1, characterized in that, Based on the real-time influent data, pollution impact trend identification and chemical demand feedforward prediction are performed to obtain a second chemical dosing guidance vector, including: Based on the real-time influent data, the pollution load change is analyzed to generate pollution impact trend identification results; Based on the real-time influent data, the influent water quality disturbance characteristics are determined, and the pollution impact trend identification results are verified based on the influent water quality disturbance characteristics to obtain the pollution impact confirmation results. Based on the pollution shock confirmation results, predict the pollution load changes during the future response cycle; Based on the pollution load change results, feedforward prediction of reagent demand is performed to obtain the feedforward reagent demand sequence. The second vector for reagent dosing guidance is generated by combining the pollution shock confirmation results and the pollution load change results.

4. The wastewater treatment dosage decision method integrating multi-source data as described in claim 1, characterized in that, The target drug dosing scheme is obtained by dynamically optimizing the neighborhood of the drug dosing adjustment space using a drug action scoring model, including: The drug action scoring model is used to perform optimization analysis on the drug dosing adjustment space to obtain the first drug dosing candidate domain and the first drug dosing potential scheme; Based on the drug action scoring model, the first drug dosing candidate domain is used to perform neighborhood competition optimization on the first drug dosing potential scheme to obtain the second drug dosing candidate domain and the second drug dosing potential scheme; Based on the drug action scoring model, neighborhood competition optimization is performed on the second drug dosing candidate domain to obtain the third drug dosing candidate domain and the third drug dosing potential scheme; Based on the drug action scoring model, the potential drug administration schemes are further optimized through neighborhood competition according to the candidate domain of the third drug administration, until the target drug administration scheme that satisfies the predetermined number of neighborhood competition optimizations is obtained.

5. The wastewater treatment dosage decision method integrating multi-source data as described in claim 4, characterized in that, The drug dosage adjustment space is optimized based on the drug action scoring model to obtain the first drug dosage candidate domain and the first drug dosage potential scheme, including: Based on each reagent dosing adjustment scheme within the reagent dosing adjustment space, future effluent prediction is performed to obtain multiple effluent prediction sequences; The multiple effluent prediction sequences are input into the agent effect scoring model to obtain multiple agent effect scores; Determine whether the multiple drug effect scores meet the predetermined scoring constraints, and obtain multiple score verification results; Based on the multiple scoring test results, the drug dosing adjustment space is screened to generate the first drug dosing candidate domain; The drug action score of the first drug dosing candidate domain is iteratively optimized to generate the first drug dosing potential scheme.

6. The wastewater treatment dosage decision method integrating multi-source data as described in claim 5, characterized in that, Based on the drug action scoring model, neighborhood competition optimization is performed on the first drug dosing candidate domain for the first drug dosing potential scheme to obtain the second drug dosing candidate domain and the second drug dosing potential scheme, including: Based on the first candidate domain for drug dosing, the trend of dosing parameters is identified, and a first drug dosing search domain is established. Based on the first drug dosing search domain, a drug dosing adjustment decision is made, and a first drug dosing neighborhood is established; Based on the drug action scoring model, the first drug dosing neighborhood is evaluated and screened according to the predetermined scoring constraints to obtain the second drug dosing candidate domain; The optimal solution for the second candidate domain is obtained by iterative optimization of the drug action score. The first potential drug administration scheme and the second candidate domain optimal scheme are evaluated by drug action score competition to generate the second potential drug administration scheme.

7. The wastewater treatment dosage decision method integrating multi-source data as described in claim 1, characterized in that, Based on the equipment operating data, the target agent dosing scheme is adjusted for response offset and disturbance suppression to obtain a agent dosing correction scheme, including: Based on the equipment operation data, extract the dosing pump response delay characteristics, valve opening deviation characteristics, chemical pipeline pressure fluctuation characteristics, chemical remaining status characteristics, and actual dosing flow characteristics; Based on the response delay characteristics of the dosing pump and the valve opening deviation characteristics, predict the dosing offset characteristics of the target dosing scheme; Based on the characteristics of pressure fluctuation in the reagent pipeline, the characteristics of the remaining reagent status, and the characteristics of the actual dosing flow rate, the execution disturbance of the target reagent dosing scheme is identified to obtain the reagent dosing disturbance characteristics; The target drug dosing scheme is collaboratively optimized based on the drug dosing offset characteristics and the drug dosing perturbation characteristics to generate the drug dosing correction scheme.

8. The wastewater treatment dosage decision method integrating multi-source data as described in claim 7, characterized in that, The drug dosing offset features include dosing timing offset and dosing flow rate offset.

9. The wastewater treatment dosage decision method integrating multi-source data as described in claim 2, characterized in that, Obtain the current water discharge deviation data, including: Based on the current water discharge deviation data, a water discharge deviation early warning signal is generated.

10. A wastewater treatment dosage decision system integrating multi-source data, characterized in that, The system is used to implement the wastewater reagent dosage decision method based on multi-source data as described in any one of claims 1-9, and the system comprises: The multi-source dynamic data acquisition module is used to acquire multi-source dynamic data from the wastewater treatment end, including real-time effluent data, real-time influent data, process operation data, and equipment operation data. The first vector acquisition module is used to trace the effluent deviation and associate the chemical contribution with the real-time effluent data based on the process operation data, and to obtain the first chemical dosing guidance vector. The second vector acquisition module is used to identify the pollution impact trend and predict the reagent demand based on the real-time influent data, and to acquire the second vector for reagent dosing guidance. The big data mining module is used to perform big data mining based on the first drug dosing guidance vector and the second drug dosing guidance vector to obtain the drug dosing adjustment space; The target drug dosing scheme acquisition module is used to perform dynamic neighborhood optimization on the drug dosing adjustment space through a drug action scoring model to obtain the target drug dosing scheme. The reagent dosing correction scheme acquisition module is used to perform response offset correction and disturbance suppression on the target reagent dosing scheme based on the equipment operation data, acquire the reagent dosing correction scheme, and execute the reagent dosing adjustment at the wastewater treatment end according to the reagent dosing correction scheme.

Images