A water treatment optimization system and regulation method based on combination of multispectral technology

Abstract

The application discloses a water treatment optimization system and regulation and control method based on a multispectral technology combination, and belongs to the software technical field.The application aims to solve the problem that in the prior art, in a multispectral combination scene, a state vector construction lacks a systematic decision process, leading to serious redundancy, noise amplification or information idling.The application provides a water treatment optimization system and regulation and control method, which comprises the following steps: constructing a candidate state set comprising multispectral analysis indexes and operation parameters;establishing a three-dimensional evaluation system of prediction contribution, signal stability and control sensitivity; dynamically selecting key state variables through a hierarchical screening strategy; and generating a refined and information-rich target state vector for control decision.Compared with the prior art, the application realizes the refinement and information richness of the state vector through a dynamic screening and derivation construction mechanism, significantly improves the control precision and calculation efficiency, and enhances the noise robustness.

Description

Technical Field

[0001] This invention relates to the field of water treatment technology, and more specifically, to a water treatment optimization system and control method based on the combined use of multispectral technology. Background Technology

[0002] In existing wastewater treatment plants, automated control or intelligent regulation systems typically rely on a small amount of online sensor data and soft-sensor or indirectly estimated indicators such as COD and ammonia nitrogen to form a process state vector. While this traditional method has low dimensionality and is easy to implement with simple PID control or experience-based rule control, its ability to finely characterize pollutant components and microbial community functions is limited.

[0003] In recent years, multispectral technologies such as near-infrared, ultraviolet-visible, three-dimensional fluorescence, and Raman spectroscopy have been gradually introduced into wastewater treatment processes. This allows operators to obtain spectral predictions of conventional indicators such as COD, TN, and TP with high temporal resolution, and to analyze in real time the protein components, humic acid components, and aromaticity of dissolved organic matter (DOM). It can even indirectly reflect the metabolic activity of functional microbial communities such as nitrifying bacteria through specific Raman or fluorescence signals. This significantly increases the observable dimensions of the process and greatly enhances state perception capabilities. However, this also brings new challenges: at each control time step, the system simultaneously obtains a large amount of candidate information from the multispectral analysis module and the process operation module, including various spectral characteristics, DOM sub-component ratios, NADH activity index, current DO value and its historical changes, influent flow rate Q and its load fluctuations, aeration intensity and its historical adjustment trajectory, forming a high-dimensional candidate state set with complex correlations between variables.

[0004] The existing technology has the following shortcomings:

[0005] In existing technologies, there are two main ways to utilize these candidate state variables: The first is to perform virtually no screening, using all available spectral analysis results and operational data as input to a certain prediction or control model. This approach leads to an expansion of the state vector dimension, with a large amount of redundancy and strong correlations among variables, making model training difficult, increasing the computational burden of real-time operation, and making it highly sensitive to noise and outliers. For example, when more than 50 spectral features and operational parameters are all input into a neural network model, the model converges slowly and is prone to overfitting, resulting in unstable control strategies.

[0006] The second type still uses an experience-based selection method, where process engineers subjectively choose a few seemingly important indicators from a large pool of candidates based on experience. For example, they might only retain DO, Q, and a few spectroscopically predicted COD and ammonia nitrogen (NH4-N) as control inputs, ignoring information such as DOM structure evolution and microbial activity. While this method maintains model simplicity, a large amount of potentially useful information is left unused, failing to fully leverage the monitoring advantages of multispectral technology. For instance, when carbon sources are insufficient in the anoxic zone, changes in DOM structure may indicate a decrease in denitrification efficiency. However, if this information is ignored, the carbon source dosage cannot be adjusted in a timely manner, leading to fluctuations in TN removal rates.

[0007] To address the above problems, this invention proposes a solution. Summary of the Invention

[0008] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a water treatment optimization system and control method based on multispectral technology to solve the problems mentioned in the background art.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A water treatment optimization system and control method based on multispectral technology includes the following steps:

[0011] The multispectral analysis results output by the multispectral analysis module and the operating data provided by the process control parameter module are obtained to form a candidate state set containing the multispectral analysis results and the operating data.

[0012] For each candidate quantity in the candidate state set, evaluation is performed from three dimensions: prediction contribution, signal stability, and control sensitivity, to form a multi-criteria evaluation result;

[0013] Based on the multi-criteria evaluation results, a finite number of key state quantities are selected from the candidate state set to form a basic state subset using a hierarchical screening and sorting method.

[0014] Based on the historical time series and spectral characteristics of the state quantities in the basic state subset, derived state quantities reflecting the dynamic characteristics of the process are constructed.

[0015] The derived state variables are merged with the basic state subset to generate a target state vector that is of appropriate dimension and rich in information for the current control objective, and this vector is used as the input to the computational decision module.

[0016] In a preferred embodiment, the multispectral analysis results include conventional water quality indicators and spectral characteristics obtained or predicted by near-infrared and ultraviolet-visible spectra, information on each component of dissolved organic matter obtained by three-dimensional fluorescence analysis, and the NADH activity index calculated by Raman spectroscopy.

[0017] In a preferred embodiment, the operating data includes online dissolved oxygen, influent flow rate, historical aeration intensity, valve opening degree, and carbon source dosage.

[0018] In a preferred embodiment, the evaluation step of the predicted contribution includes: analyzing the impact of introducing or removing a candidate quantity on the prediction error of short-term targets such as dissolved oxygen or total nitrogen, in order to determine the magnitude of the contribution of the candidate quantity to the current control target.

[0019] In a preferred embodiment, the signal stability evaluation step includes: statistically analyzing the variance, outlier ratio, and drift trend of the candidate quantity over a recent period to assess its measurement noise and stability.

[0020] In a preferred embodiment, the control sensitivity assessment step includes: reviewing the relationship between historical control actions and process responses, and analyzing the sensitivity correlation between changes in the candidate quantity and changes in the controlled object.

[0021] In a preferred embodiment, the hierarchical filtering and sorting method includes:

[0022] First, candidates with almost no substantial impact on the target are eliminated based on their predicted contributions;

[0023] Secondly, candidate quantities susceptible to noise interference are eliminated based on signal stability;

[0024] Finally, the remaining set is sorted according to control sensitivity, and a finite number of key state variables are selected to form the basic state subset.

[0025] In a preferred embodiment, the derived state quantity includes:

[0026] The dissolved oxygen recovery time or the dissolved oxygen response intensity caused by a unit aeration adjustment is constructed based on the decay trajectory of dissolved oxygen deviation over time.

[0027] The structural abrupt change amplitude of soluble organic compounds is constructed based on the short-term changes in the fluorescence peak intensity and E2 / E3 ratio of soluble organic compounds; where E2 represents the absorbance of the ultraviolet-visible spectrum at a wavelength of 254 nm, and E3 represents the absorbance of the ultraviolet-visible spectrum at a wavelength of 365 nm.

[0028] The duration of activity inhibition is constructed based on the length of time that the NADH activity index remains below the baseline value.

[0029] A water treatment optimization and control system based on multispectral technology includes:

[0030] The multispectral analysis module is used to output multispectral analysis results;

[0031] The process control parameter module is used to provide operating data;

[0032] The computational decision module includes an internal state vector filtering and derivation construction unit, which is connected to the multispectral analysis module and the process control parameter module. This unit receives the multispectral analysis results and the operating data and executes the aforementioned method.

[0033] In a preferred embodiment, the computational decision module further includes a rule layer and a model layer. The target state vector serves as a unified input to the rule layer and the model layer, and is used to achieve coordinated optimization control of the aeration fan frequency, aeration valve opening degree, and carbon source dosage.

[0034] The technical effects and advantages of the water treatment optimization system and control method based on multispectral technology of the present invention are as follows:

[0035] This invention reduces the dimensionality of the state vector from over 50 to 8-12 core variables through a dynamic filtering mechanism, a 70% reduction. This significantly reduces model inference time and improves computational efficiency, thereby meeting the requirements of real-time control and reducing the computational burden on the decision-making module.

[0036] This invention reduces DO control deviation and narrows the fluctuation range of TN removal rate by constructing a refined and information-rich target state vector. This indicates that the invention can more accurately reflect the process state, thereby achieving more precise control and improving the efficiency and stability of water treatment.

[0037] This invention introduces a derived state variable construction mechanism to incorporate key information reflecting process dynamics and microbial community functional states into the state vector, avoiding the problem of a large amount of potentially useful information being left idle in traditional methods. For example, through derived state variables such as DO recovery time, the magnitude of DOM structural mutations in dissolved organic matter, and the duration of activity inhibition, the system can more comprehensively perceive the process dynamic response, providing richer information for control decisions.

[0038] This invention establishes a three-dimensional evaluation system encompassing prediction contribution, signal stability, and control sensitivity, providing a systematic decision-making framework oriented towards control objectives. It addresses the lack of a clear trade-off mechanism in existing state vector construction. This enables a more scientific and rational utilization of multispectral data at the control level, achieving an integrated technical path from deeper and more detailed analysis to more accurate and cost-effective tuning. Attached Figure Description

[0039] Figure 1 This is a schematic diagram of a water treatment optimization system and control method based on the combined use of multispectral technology according to the present invention;

[0040] Figure 2 This is a schematic diagram of a water treatment optimization and control method based on the combined use of multispectral technology according to the present invention;

[0041] Figure reference numerals: 110, Multispectral analysis module; 120, Process control parameter module; 130, Calculation and decision module; 131, State vector screening and derivation construction unit; 132, Rule layer; 133, Model layer. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0043] Example 1: This example provides a water treatment optimization method based on the combined use of multispectral technologies. Its core lies in extracting refined and information-rich state vectors from massive amounts of multispectral and operational data through a dynamic screening and derivation construction mechanism to support the optimized control of the water treatment process. Within each control time step, this method, with the current specific control objective as the core, comprehensively determines which state variables should be included in the comprehensive state vector, which should be temporarily discarded, and whether it is necessary to dynamically supplement derived state variables reflecting the dynamic characteristics of the process.

[0044] Reference Figure 2 This method includes the following steps:

[0045] Step 210: Constructing a candidate state set. At each control time step, the system first obtains multispectral analysis results from the multispectral analysis module (110) and operational data from the process control parameter module (120). These data together constitute the current candidate state set.

[0046] Specifically, the multispectral analysis results include:

[0047] Conventional water quality indicators, such as chemical oxygen demand (COD), total nitrogen (TN), and total phosphorus (TP), are predicted using near-infrared (NIR) and ultraviolet-visible (UV-Vis) spectroscopy. These indicators provide information on the concentration of major pollutants in the water body.

[0048] Information on the components of dissolved organic matter (DOM) obtained by three-dimensional fluorescence spectroscopy, such as humic acids, proteins, and soluble microbial metabolites. This component information reflects the source, structure, and biodegradability of the organic matter.

[0049] The NADH activity index was calculated using Raman spectroscopy. NADH is an important coenzyme in microbial metabolism, and its activity index can indirectly reflect the metabolic intensity and activity of functional microbial communities. Specifically, it is the reduced form of nicotinamide adenine dinucleotide and a key electron transport coenzyme in microbial respiratory metabolism. Its fluorescence or Raman signal intensity can serve as a characterization indicator of the metabolic activity of microbial communities, especially the activity of functional microbial communities such as nitrifying bacteria. The NADH activity index in this invention is a dimensionless activity characterization quantity constructed based on this signal.

[0050] The runtime data includes:

[0051] Current online dissolved oxygen (DO) levels and their historical trends. DO is a key indicator for aeration control.

[0052] Influent flow rate (Q) and its load fluctuations. Changes in influent flow rate directly affect the treatment load;

[0053] Historical aeration intensity and adjustment trajectory. Aeration intensity is a crucial operating parameter affecting dissolved oxygen (DO) and microbial activity.

[0054] Valve opening degree.

[0055] Carbon source addition amount.

[0056] These data come together to form a high-dimensional set of candidate states. For example, in a typical wastewater treatment plant, there may be more than 50 raw spectral features, more than 10 DOM sub-component ratios, 3 NADH activity indices, and about 10 operating parameters, totaling more than 70 candidates.

[0057] Step 220: Perform multi-criteria evaluation. For each candidate variable in the candidate state set constructed in Step 210, perform quantitative or semi-quantitative evaluation from three dimensions: prediction contribution, signal stability, and control sensitivity, to form a multi-criteria evaluation result. Specifically, this includes:

[0058] 221. Predictive Contribution Assessment: This assessment aims to determine the predictive or explanatory power of a candidate variable for the current control objective, such as maintaining stable DO in the aerobic phase, ensuring no decrease in TN removal rate, and avoiding short-term carbon deficiency in the anoxic phase. A specific method can be ridge regression analysis to assess the impact of introducing or removing a candidate variable on the prediction error of short-term DO, TN, and other targets. Assume we have a prediction model M, whose input is a state vector. The output is the control objective Y. For each candidate variable... We can calculate its marginal contribution. .

[0059] One calculation method is to compare the models in which they include and not included The change in prediction error over time. For example, using mean squared error (MSE) as a metric for prediction error: ;

[0060] ;

[0061] but .if The larger, the more The greater the contribution to prediction.

[0062] In the formula, is the total number of samples; j is the index of the j-th sample; For the j-th sample, the actual observed value is such as pollutant concentration, dissolved oxygen level, carbon source demand, etc. The complete input feature vector for the j-th sample contains all features from multispectral data, process operation data, microbial activity indicators, etc. For example, the i-th feature in the feature vector, such as the intensity of a certain band, a certain spectral ratio, or the operating condition value of a certain sensor; Let be the feature vector of the j-th sample, but remove or mask the i-th feature xi. Masking methods typically include: setting it to the mean; setting it to zero; replacing it with noise, etc.; M(·) is the prediction model function, which can be a linear model, a tree model, or a neural network, and will not be elaborated here. This represents the mean squared prediction error when using the complete feature set. The mean square value of the prediction error after removing the i-th feature; The feature contribution score is given; a score greater than 0 indicates that the feature has made a contribution.

[0063] In addition, the predictive contribution can also be measured by calculating the mutual information between features and the target.

[0064] ;

[0065] in, This represents the mutual information between random variables X and Y. It is their joint probability distribution function. and These are their marginal probability distribution functions. The greater the mutual information, the greater the contribution of the candidate variable to the target prediction; The input feature variable can be a single feature, such as... The ratio can also be the set of discretized feature values; Discretized intervals for target variables, such as DO, NH4-N, TN, carbon source addition, etc. It is a value of the random variable X, such as a certain absorbance range; Let Y be a value of the random variable Y, such as a certain interval of DO.

[0066] 222. Signal Stability Assessment: This assessment aims to measure the measurement noise level, outlier frequency, and drift trend of candidate quantities to filter out quantities with excessive noise or frequent anomalies. Specific methods include:

[0067] Variance or standard deviation: Statistical analysis of the variance or standard deviation of the candidate quantity over a recent period, such as the past 24 hours or 100 sampling points. The larger the variance, the greater the signal fluctuation and the worse the stability.

[0068] Outlier ratio: Outliers are identified using statistical methods, such as the 3σ criterion or box plot method, and the proportion of outliers within the statistical period is calculated. The higher the proportion, the lower the signal reliability.

[0069] Drift Trend: Analyze whether there is a significant trend of drift in the signal using methods such as linear regression or moving average. For example, for candidate quantities... Its stability score Both the coefficient of variation (CV) and the outlier ratio (OR) can be considered: but ,in It is a decreasing function, for example ,in As weight.

[0070] In the formula, Features The standard deviation is used to characterize volatility; Features The mean; The coefficient of variation; The number of samples identified as outliers; This represents the total number of samples. The proportion of outliers; This is the comprehensive stability evaluation function.

[0071] Control sensitivity assessment: This assessment aims to analyze the sensitivity association between changes in candidate quantities and changes in controlled objects, such as DO and TN, filtering out quantities that are difficult to effectively influence through control measures or whose influence is unstable. Specific methods include reviewing historical control actions and process responses, such as adjustments to aeration intensity and changes in carbon source dosage, and the relationship between these historical control actions and process responses such as changes in DO and TN. For example, if a certain spectral characteristic is highly correlated with the trend of DO changes, and this correlation shows a clear causal relationship after aeration intensity adjustments, then the control sensitivity of this characteristic is high. Granger causality tests or cross-correlation analysis can be used to quantify this sensitivity. For candidate quantities... and control objects Calculate their cross-correlation coefficients at different lag times. This is used to describe the correlation between a feature change that lags behind or leads the target variable.

[0072] ;

[0073] in, The lag time is used. A higher cross-correlation coefficient indicates stronger sensitivity. For example, the i-th feature in a time series, such as the intensity of a certain spectral channel; Time series with target variables, such as D0, effluent NH4-N, etc.; and The mean of their respective time series; and The standard deviation of each time series; The expected value is calculated by averaging through a sliding window.

[0074] It should be noted that in this specification, all instances of t represent discrete sampling time t=n·Δt, where n is a non-negative integer.

[0075] Step 230: Hierarchical Screening and Ranking. Based on the multi-criteria evaluation results formed in Step 220, a hierarchical screening and ranking method is used to select a finite number of key state quantities from the candidate state set to form a basic state subset. This hierarchical screening avoids the information loss that may be caused by simple weighted scoring. The specific screening process is as follows:

[0076] a. Predicted Contribution Screening: First, eliminate candidate quantities whose predicted contribution to the target is lower than a preset threshold. For example, predicted contribution... These quantities have almost no practical effect on the current control objective and can be safely discarded to reduce redundancy.

[0077] b. Signal stability screening: From the remaining candidates, further eliminate candidates with signal stability scores below a preset threshold. For example, stability score... or coefficient of variation or the proportion of outliers These quantities are susceptible to noise interference or frequent anomalies. Incorporating them into the state vector introduces uncertainty and affects the stability of the control strategy.

[0078] c. Control Sensitivity Ranking and Selection: After the first two rounds of screening, the remaining candidate variables contribute to the control objective and provide stable signals. At this point, they are ranked according to their control sensitivity, and the top N candidate variables are selected to form a subset of the basic states. For example, N = 8-12. Specifically, the value of N can be adjusted according to the complexity of the actual control system and computational resources.

[0079] Step 240: Constructing Derived State Quantities. After obtaining the subset of basic states, this invention further introduces a mechanism for constructing derived state quantities. These derived state quantities are characteristic quantities that can reflect the dynamic characteristics of the process and compensate for the deficiencies of static measurements. Specifically, the derived state quantities include:

[0080] DO recovery time or DO response intensity caused by a unit aeration adjustment: Constructed based on the decay trajectory of DO deviation over time. For example, the time required for DO to recover from its current value to the target range after a change in aeration intensity, or the rate of change of DO caused by a unit aeration intensity adjustment. This characterizes the dynamic response capability of the current aeration system. For example, defining DO recovery time... for:

[0081] ;

[0082] in, Adjusting the aeration time, For the target D0 value, This represents the allowable deviation. This refers to the sampling time step, i.e., the system time resolution, such as 1 min / 5 min; This is dissolved oxygen monitoring data.

[0083] The magnitude of structural abrupt changes in DOM of soluble organic compounds: based on the intensity of DOM fluorescence peaks and Construction of short-term changes in ratios.

[0084] It should be noted that, among them, This indicates the absorbance of the ultraviolet-visible spectrum at a wavelength of 254 nm. This indicates the absorbance of the ultraviolet-visible spectrum at a wavelength of 365 nm. The ratio is used to characterize the average molecular weight and aromaticity of dissolved organic matter, reflecting the degree of humification.

[0085] For example, the intensity of the DOM fluorescence peak or The maximum rate of change of the ratio. This can be used to indicate the degree of change in the properties and biodegradability of organic matter, and to predict potential carbon source insufficiency or changes in microbial activity.

[0086] For example, defining the magnitude of DOM structural mutations. for:

[0087] ;

[0088] in, This is the time window, the length of the time window used to detect sudden changes in the DOM structure, such as 1 hour or 1 day. This is the current moment in the analysis.

[0089] Duration of activity inhibition: Constructed based on the length of time the NADH activity index remains below a baseline value. For example, if the NADH activity index is below the average value for normal operation for N consecutive sampling periods, the duration is recorded. This can be used to reflect the degree of inhibition of functional microbial communities such as nitrifying bacteria and to provide early warning of potential declines in nitrification efficiency.

[0090] For example, defining the duration of activity inhibition. for:

[0091] ;

[0092] in, This is an indicator function; it returns 1 if the condition is met, and 0 otherwise. The baseline NADH activity value is given, and L is the number of historical sampling points. It also refers to the length of time that microbial activity remains below the baseline value.

[0093] Step 250: Generate the target state vector. The derived state variables constructed in Step 240 are merged with the subset of basic states selected in Step 230 to generate a target state vector that is appropriate in dimension and rich in information for the current control objective. This target state vector is then used as the unified input to the rule layer (132) and model layer (133) in the computational decision module (130) to achieve subsequent coordinated optimization control of the aeration fan frequency, aeration valve opening, and carbon source dosage.

[0094] Through the above method, this invention no longer simply feeds all the multispectral analysis results into the model or uses only a few empirical variables. Instead, within each control cycle, it focuses on a clear, small objective, utilizing the inherent relationship between multispectral analysis data and operational data to complete a selective and structured comprehensive analysis and construction of state variables. This avoids high-dimensional redundancy and noise amplification while fully incorporating key information reflecting the dynamic characteristics of the process and the functional state of the microbial community. The resulting target state vector, combined with the subsequent interpretable decision model, enables the water treatment optimization system and control method based on multispectral technology to truly achieve an integrated technical path from seeing more deeply and in greater detail to adjusting more accurately and economically.

[0095] Example 2: This example provides a water treatment optimization system based on multispectral technology to implement the above method. (Refer to...) Figure 1 The system includes a multispectral analysis module (110), a process control parameter module (120), and a calculation and decision module (130). The calculation and decision module (130) internally includes a state vector filtering and derivation construction unit (131), a rule layer (132), and a model layer (133). In engineering implementation, the multispectral analysis module (110) can collect and input raw spectral data through an external spectral sensor; the process control parameter module (120) can connect to on-site online instruments or control systems to obtain operating data. The calculation and decision module (130) can receive the multispectral analysis results and operating data through a communication interface, and issue control commands to the on-site actuators through a control command output interface; historical data can be saved by an external storage device and called by the calculation and decision module (130). The above external devices / interfaces are engineering configurations and are not considered as part of the overall system. Figure 1 The system components shown.

[0096] Specifically, the system consists of the following components:

[0097] Multispectral analysis module (110): It is responsible for receiving raw spectral data collected by external spectral sensors, such as NIR, UV-Vis, three-dimensional fluorescence, and Raman spectrometers, and processing and analyzing the raw spectral data through built-in chemometrics or machine learning models, and outputting conventional water quality indicators, information on each component of DOM, such as the fluorescence intensity and proportion of humic acid and protein substances, as well as the NADH activity index.

[0098] Conventional water quality indicators include chemical oxygen demand (COD), total nitrogen (TN), total phosphorus (TP), and nitrate nitrogen, which are used to characterize the load levels of organic and nitrogenous pollutants in wastewater.

[0099] Process control parameter module (120): This module is responsible for collecting various operational data from the wastewater treatment process, including but not limited to online DO sensor data, influent flow meter data, aeration fan frequency, aeration valve opening, carbon source dosage, pH value, and temperature. This module transmits this data to the calculation and decision module in real time.

[0100] The computation and decision-making module (130) is the core control unit of the system, responsible for receiving all input data, constructing state vectors, generating control strategies, and outputting instructions. This module includes an internal state vector filtering and derivation construction unit (131).

[0101] State Vector Filtering and Derivation Construction Unit (131): This unit is the innovation of the present invention. It receives the multispectral analysis results output from the multispectral analysis module (110) and the operating data provided by the process control parameter module (120). At each control time step, this unit performs the steps described in Embodiment 1 as constructing a candidate state set, performing multi-criteria evaluation, hierarchical filtering and sorting, and constructing derived state quantities, based on the current control task, and finally generates a refined and information-rich target state vector.

[0102] Rule layer (132): Receives the target state vector output by the state vector filtering and derivation building unit (131), and makes preliminary judgments or corrections on the control actions based on preset expert experience rules or fuzzy control rules. For example, when DO is below the threshold and the DO recovery time is too long, the rule layer may suggest increasing the aeration intensity.

[0103] Model layer (133): Receives the target state vector output by the state vector filtering and derivation building unit (131), and uses a control model built based on machine learning or optimization algorithms, such as reinforcement learning or predictive control models, to generate optimal control commands. For example, it predicts the changes in DO and TN over a future period based on the target state vector, and calculates the optimal aeration fan frequency and carbon source dosage. Historical operating data, spectral data, control commands, and state vectors can be saved by an external storage device for use by the computation and decision-making module (130).

[0104] In one optional example, the system workflow is as follows:

[0105] An external spectral sensor continuously collects water quality spectral information and inputs it into a multispectral analysis module (110) for analysis, generating multispectral analysis results.

[0106] The process control parameter module (120) collects online DO, influent flow rate and other operating data in real time.

[0107] The multispectral analysis results and running data are simultaneously transmitted to the state vector filtering and derivation construction unit (131) in the computational decision module (130).

[0108] The state vector filtering and derived building block (131) performs the following operations based on the current control objective:

[0109] a. Construct a candidate state set containing multispectral analytical indicators and operational data.

[0110] b. Perform a three-dimensional evaluation of each candidate variable in the set, including prediction contribution, signal stability, and control sensitivity.

[0111] c. Employ a hierarchical screening and sorting strategy to select key state variables from the candidate set to form a subset of basic states.

[0112] d. Based on the historical time series and spectral characteristics of the basic state variables, construct derived state variables that reflect the dynamic characteristics of the process.

[0113] e. Combine the derived state variables with the subset of basic states to generate the target state vector.

[0114] The generated target state vector serves as the unified input to the rule layer (132) and model layer (133) in the computational decision module (130).

[0115] The rule layer (132) and the model layer (133) work together to calculate the optimal control commands such as the aeration fan frequency, aeration valve opening degree and carbon source addition amount based on the target state vector and the preset control logic.

[0116] The calculation and decision module (130) sends these control commands to the field actuators through the control command output interface to realize real-time optimization control of the water treatment process.

[0117] Through this system, the present invention has effectively implemented multispectral technology in water treatment optimization control. The state vector screening and derivation building unit (131) ensures that the state vector input to the control model is neither excessively redundant nor lacks key dynamic process information, thereby significantly improving the intelligence level and operating efficiency of the control system.

[0118] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A water treatment optimization and control method based on multispectral technology, characterized in that, Includes the following steps: The multispectral analysis results output by the multispectral analysis module and the operating data provided by the process control parameter module are obtained to form a candidate state set. The multispectral analysis results include conventional water quality indicators and spectral characteristics obtained or predicted by near-infrared and ultraviolet-visible spectra, information on each component of dissolved organic matter obtained by three-dimensional fluorescence analysis, and NADH activity index calculated by Raman spectroscopy. The operating data includes online dissolved oxygen, influent flow rate, historical aeration intensity, valve opening degree, and carbon source dosage. For each candidate quantity in the candidate state set, evaluation is performed from three dimensions: prediction contribution, signal stability, and control sensitivity, to form a multi-criteria evaluation result. The prediction contribution is used to characterize the impact of introducing or removing the candidate quantity on the short-term dissolved oxygen or total nitrogen target prediction error. The signal stability is used to characterize the variance, outlier ratio, and drift trend of the candidate quantity in a recent period. The control sensitivity is used to characterize the sensitivity correlation between the change of the candidate quantity and the change of the controlled object in the relationship between historical control actions and process responses. Based on the multi-criteria evaluation results, a hierarchical screening and sorting method is adopted to select a finite number of key state quantities from the candidate state set to form a basic state subset. The hierarchical screening and sorting includes: firstly, eliminating candidate quantities that have almost no substantial effect on the target according to the predicted contribution; secondly, eliminating candidate quantities that are susceptible to noise interference according to the signal stability; and finally, sorting the remaining set according to control sensitivity to select a finite number of key state quantities to form the basic state subset. Based on the historical time series and spectral characteristics of the state quantities in the basic state subset, derived state quantities reflecting the dynamic characteristics of the process are constructed. These derived state quantities include: dissolved oxygen recovery time or dissolved oxygen response intensity caused by unit aeration adjustment, constructed based on the decay trajectory of dissolved oxygen deviation over time; dissolved organic matter structural mutation amplitude, constructed based on short-term changes in the fluorescence peak intensity and E2 / E3 ratio of dissolved organic matter; and activity inhibition duration, constructed based on the duration of time when the NADH activity index is continuously lower than the baseline value. The derived state variables are merged with the basic state subset to generate a target state vector that is of moderate dimension and rich in information for the current control objective. The target state vector is then used as a unified input to the rule layer and model layer in the computational decision module to achieve coordinated optimization control of aeration fan frequency, aeration valve opening degree, and carbon source dosage.

2. The water treatment optimization and control method based on multispectral technology according to claim 1, characterized in that: The conventional water quality indicators include at least one of chemical oxygen demand, total nitrogen, total phosphorus, and nitrate nitrogen, and the information on each component of dissolved organic matter includes at least one of humic acid substances, protein substances, and soluble microbial metabolites.

3. The water treatment optimization and control method based on multispectral technology according to claim 1, characterized in that: The operational data also includes at least one of the following: historical trend of online dissolved oxygen, influent flow rate load fluctuation, historical aeration intensity adjustment trajectory, pH value, and temperature.

4. The water treatment optimization and control method based on multispectral technology according to claim 1, characterized in that: The evaluation steps for the predicted contribution include using ridge regression analysis to introduce or remove the influence of a candidate quantity on the prediction error of short-term dissolved oxygen or total nitrogen targets, in order to determine the magnitude of the contribution of the candidate quantity to the current control target.

5. The water treatment optimization and control method based on multispectral technology according to claim 1, characterized in that: The signal stability evaluation steps include, The variance or standard deviation, outlier ratio, and drift trend of the candidate quantity over a recent period are statistically analyzed to assess its measurement noise and stability.

6. The water treatment optimization and control method based on multispectral technology according to claim 1, characterized in that: The steps for assessing control sensitivity include reviewing the relationship between historical control actions and process responses, and using Granger causality tests or cross-correlation analysis to quantify the sensitivity association between changes in the candidate variable and changes in the controlled object.

7. The water treatment optimization and control method based on multispectral technology according to claim 6, characterized in that: In the hierarchical screening and sorting process, candidates with a prediction contribution lower than a preset threshold are first eliminated, then candidates with a stability score lower than a preset threshold, or a coefficient of variation higher than a preset threshold, or an outlier ratio higher than a preset threshold are eliminated. Then, the remaining candidates are sorted according to control sensitivity, and the top N key state variables are selected to form the basic state subset, where N is 8 to 12.

8. The water treatment optimization and control method based on multispectral technology according to claim 2, characterized in that: The derived state quantities include: The dissolved oxygen recovery time or the dissolved oxygen response intensity caused by a unit aeration adjustment is constructed based on the decay trajectory of dissolved oxygen deviation over time. The structural abrupt change amplitude of soluble organic compounds is constructed based on the short-term changes in the fluorescence peak intensity and E2 / E3 ratio of soluble organic compounds; where E2 represents the absorbance of the ultraviolet-visible spectrum at a wavelength of 254 nm, and E3 represents the absorbance of the ultraviolet-visible spectrum at a wavelength of 365 nm. The duration of activity inhibition is constructed based on the length of time that the NADH activity index remains below the baseline value.

9. A water treatment optimization and control system based on multispectral technology, used to implement the method described in any one of claims 1-8, characterized in that, include: The multispectral analysis module is used to output the multispectral analysis results; The process control parameter module is used to provide the operating data; The computational decision module includes a state vector filtering and derivation construction unit, which is connected to the multispectral analysis module and the process control parameter module. The state vector filtering and derivation construction unit is used to receive the multispectral analysis results and the operating data, and to execute the method described in any one of claims 1 to 8.

10. A water treatment optimization and control system based on multispectral technology according to claim 9, characterized in that: The computational decision module also includes a rule layer and a model layer. The target state vector serves as a unified input to the rule layer and the model layer, and is used to achieve coordinated optimization control of the aeration fan frequency, aeration valve opening degree, and carbon source dosage.

Images