A method and system for precise control of dissolved oxygen for aerobic treatment of sewage

Abstract

The application discloses a kind of for sewage aerobic treatment dissolved oxygen precision control method and system, it is related to sewage treatment field, the dissolved oxygen precision control method for sewage aerobic treatment, it includes the following steps: S1, obtains data, and carries out pretreatment;S2, preset strategy, and extract characteristic parameter;S3, carry out time sequence division, obtain time sequence set;S4, to time sequence set is analyzed, and set associated rule;S5, will time sequence set according to associated rule generate matrix, and carry out reasoning generation optimal dissolved oxygen set value;S6, using model predictive control algorithm constructs model, and generates optimal aeration control sequence;S7, generates sewage aerobic treatment system dissolved oxygen control operation evaluation report.The present application is realized through the global data acquisition and multidimensional feature fusion extraction of perception layer multi-source sensor, the panoramic portrayal of oxidation-reduction potential, microbial metabolic activity, water condition fluctuation and historical aeration law of DO concentration in aerobic tank.

Description

Technical Field

[0001] This invention relates to the field of wastewater treatment technology, and more specifically, to a method and system for precise control of dissolved oxygen in aerobic wastewater treatment. Background Technology

[0002] Dissolved oxygen concentration, as a core parameter of aerobic wastewater treatment systems, directly determines microbial activity, organic matter degradation efficiency, and nitrogen and phosphorus removal effects. It has become a key technological tool for ensuring the stable operation of aerobic wastewater treatment processes, improving treatment efficiency, and avoiding effluent exceeding standards and energy waste. A precise dissolved oxygen control system based on multi-source data sensing, multi-variable coupled analysis, and intelligent algorithm collaboration is not only a key means to solve the problems of slow response and weak anti-interference ability of traditional control methods, but also provides technical support for the state perception, model building, and collaborative regulation of smart water treatment systems, avoiding the risks of decreased treatment efficiency, increased energy consumption, and effluent exceeding standards caused by inaccurate control and delayed regulation.

[0003] However, existing methods and systems for precise dissolved oxygen control in aerobic wastewater treatment often employ fixed PID regulation or feedback control based on offline detection in practical applications. These methods suffer from prominent issues such as lag in dynamic response and weak anti-interference capabilities. Furthermore, single DO sensor feedback control cannot accurately match the actual oxygen demand because it cannot perceive changes in microbial metabolic state in real time. Although existing intelligent control technologies can partially optimize control strategies, they lack in-depth analysis of multivariate coupling relationships, resulting in insufficient robustness of the system under complex operating conditions. This makes it unable to adapt to the control requirements of complex scenarios such as fluctuating influent load and changes in microbial activity.

[0004] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0005] To address the problems in related technologies, this invention proposes a method and system for precise control of dissolved oxygen in aerobic wastewater treatment, thereby overcoming the aforementioned technical problems in existing related technologies.

[0006] To achieve the above objectives, the specific technical solution adopted by the present invention is as follows: According to one aspect of the present invention, a method for precise control of dissolved oxygen in aerobic wastewater treatment includes the following steps: S1. Acquire multi-source sensing data, influent operating condition data, control cycle parameters and historical aeration operation data of the aerobic wastewater treatment system. The multi-source sensing data includes core data of the aerobic tank water environment and microbial metabolic state data, and preprocess the core data of the water environment and microbial metabolic state data. S2. Based on a multi-source data fusion feature extraction strategy, extract the core feature parameters for dissolved oxygen control and supplementary feature parameters for microbial activity from the preprocessed data. S3. Based on the control cycle parameters, the core characteristic parameters, supplementary characteristic parameters, influent operating condition data and historical aeration operation data are divided into time series to obtain four types of time series sets: core control characteristics, microbial activity, influent operating conditions and historical aeration. As a preferred embodiment, the step of dividing the core characteristic parameters, supplementary characteristic parameters, influent operating condition data, and historical aeration operation data into four time series sets based on control cycle parameters to obtain four types of time series sets: core control characteristics, microbial activity, influent operating conditions, and historical aeration, includes the following steps: S31. Using the control cycle parameter as a unified time reference, perform time synchronization and alignment on various parameters, match the timing sampling frequency with the control cycle, and determine the time window for dividing the control cycle. S32. After synchronization and alignment, various parameters are segmented periodically according to the time window, and four types of time series sets are generated according to the data type, and the time dimension of each type of time series set remains consistent. As a preferred embodiment, the step of periodically segmenting various parameters after synchronization and alignment according to a time window, generating four types of time series sets according to data type, and ensuring that the time dimension of each type of time series set remains consistent includes the following steps: S321. Clearly define the start and end timestamps of the control cycle time window and match them with the sampling frequency of various parameters after synchronization and alignment; S322. Based on the matching results, segment and extract various parameters periodically to ensure that each segment of data corresponds to a unique control period. S323. The extracted parameters are categorized into four time series sets, and the output is generated after verifying that the time dimension is consistent.

[0007] S33. Verify and adjust the timing length and data integrity of the four types of timing sets to complete the timing partitioning.

[0008] S4. Based on the attention-based long short-term memory network, a deep analysis of the multivariate coupling relationship of four types of time series sets is carried out, and multivariate temporal coupling association rules are set among the four types of time series sets. As a preferred embodiment, the method of using a long short-term memory network with an attention mechanism to perform deep analysis of multivariate coupling relationships on four types of time series sets and setting multivariate temporal coupling association rules among the four types of time series sets includes the following steps: S41. Combining the characteristics of multivariate time series data of four types of time series sets with the requirements of aerobic wastewater treatment process, configure the model input layer, attention weight allocation layer and long short-term memory network hidden layer to construct a long short-term memory network model with attention mechanism. S42. After normalizing and preprocessing the four types of time series sets, they are used as multi-channel input data and fed into the long short-term memory network model. S43. The attention weight allocation layer is used to assign feature importance weights to the four types of time series sets. The hidden layer is used to capture the temporal dependencies, and the deep analysis of multivariate coupling relationships is carried out to output the coupling analysis results. As a preferred embodiment, the steps of assigning feature importance weights to the four types of time series sets through an attention weight allocation layer, capturing temporal dependencies using hidden layers, conducting deep analysis of multivariate coupling relationships, and outputting coupling analysis results include the following steps: S431. Initialize the attention weight allocation layer, using the influence of the four types of time series on dissolved oxygen control as the core basis for weight allocation, so that the weight allocation fits the process control requirements. S432. Input the four types of time series sets into the attention weight allocation layer, perform feature differentiation weight allocation and complete the assignment; S433. Input the four types of time series sets after weight assignment into the hidden layer, capture the changing patterns of feature parameters periodically, explore the correlation and temporal dependency relationships among the four types of time series sets, and obtain preliminary analysis results of coupling relationships. S434. Integrate and organize the preliminary analysis results to form coupled analysis results.

[0009] S44. Based on the coupling analysis results and process requirements, clarify the correlation threshold and timing response relationship of the characteristic parameters of each time series set, and set multivariate timing coupling correlation rules among the four types of time series sets.

[0010] S5. Based on the coupling association rule, generate a multivariate coupling prediction matrix from the four types of time series sets, perform time series dynamic reasoning on the matrix, and generate the optimal dissolved oxygen setpoint within the future control step. As a preferred embodiment, the step of generating a multivariate coupled prediction matrix from the four types of time series sets according to the coupling association rule, and performing time-series dynamic inference on the matrix to generate the optimal dissolved oxygen setpoint within the future control step includes the following steps: S51. Based on the coupling association rule, clarify the association dimension of the feature parameters of the four types of time series sets, determine the row and column dimensions of the multivariate coupling prediction matrix, the row dimension corresponds to the control step size, and the column dimension corresponds to the core feature parameters of the four types of time series sets. S52. Integrate the feature parameters of the four types of time series sets according to the coupling association rules, preset the corresponding dimension empty matrix and fill the feature parameters to generate a multivariate coupled prediction matrix. As a preferred embodiment, the step of integrating the feature parameters of the four types of time series sets according to the coupling association rules, pre-setting an empty matrix of the corresponding dimension and filling it with feature parameters to generate a multivariate coupled prediction matrix includes the following steps: S521. Clarify the correlation relationship of the four types of time series feature parameters in the coupling association rule, and determine the feature parameter integration rule by combining the row and column dimensions of the prediction matrix; S522. Extract all feature parameters of the four types of time series sets and classify and organize them according to the integration rules; S523. Set a coupling empty matrix based on the row and column dimensions of the prediction matrix. The row dimension corresponds to the control step size, and the column dimension corresponds to the core feature parameters of the four types of time series sets. S524. Align and fill the rearranged feature parameters one by one into the coupling empty matrix, so that each element in the matrix corresponds to a unique feature parameter and time sequence node, and generate a multivariate coupling prediction matrix.

[0011] S53. Combining the control cycle parameters and the coupling analysis results, determine the future control step size and time-series dependency, perform time-series dynamic deduction on the prediction matrix, and predict the changing trends of various characteristic parameters within the future control step size; S54. Based on the simulation results and the process's required dissolved oxygen threshold, the optimal dissolved oxygen setpoint within the future control step is determined.

[0012] S6. A closed-loop control model for an aeration system with feedforward compensation, feedback correction and disturbance suppression capabilities is constructed using a model predictive control algorithm. The optimal dissolved oxygen setpoint is converted into a real-time control target parameter input model. The optimal operating control parameters for the variable frequency fan, air diffusion device and liquid level regulating valve are calculated to generate the optimal aeration control sequence. As a preferred embodiment, the method of constructing a closed-loop control model for the aeration system with feedforward compensation, feedback correction, and disturbance suppression capabilities using a model predictive control algorithm, converting the optimal dissolved oxygen setpoint into a real-time control target parameter input model, calculating the optimal operating control parameters for the variable frequency fan, air diffuser, and liquid level regulating valve, and generating the optimal aeration control sequence includes the following steps: S61. Based on the operating rules of the aeration system reflected by the control cycle parameters and historical aeration operation data, a closed-loop control model of the aeration system is constructed by combining the model predictive control algorithm. S62. The optimal dissolved oxygen setpoint is taken as the core controlled target. The feedforward compensation term and feedback correction term are constructed by combining the influent operating data and historical aeration operation data, and the setpoint is converted into real-time control target parameters. S63. Substitute the feedforward compensation term, feedback correction term, and real-time control target parameters into the model, calculate the optimal operating control parameters for the three types of actuators, and integrate them to generate the optimal aeration control sequence.

[0013] As a preferred embodiment, the step of substituting the feedforward compensation term, feedback correction term, and real-time control target parameters into the model to calculate the optimal operating control parameters for the three types of actuators and integrating them to generate the optimal aeration control sequence includes the following steps: S631. Take the real-time control target parameters as the core controlled reference of the model, substitute them with feedforward compensation terms and feedback correction terms, and clarify the priority and correlation of the coupling operation of the three types of parameters in the model. S632. Based on the control cycle parameters and the operating law of the aeration system, the three types of coupled parameters are calculated step by step through the model to determine the optimal operating frequency of the variable frequency fan, the optimal opening degree of the air diffusion device and the optimal adjustment amount of the liquid level regulating valve. S633. Integrate the three types of optimal operating control parameters to generate the optimal aeration control sequence.

[0014] S7. Perform real-time verification and dynamic optimization of the execution effect and dissolved oxygen control accuracy of the optimal aeration control sequence, verify the robustness of the system under complex disturbance conditions, and generate a dissolved oxygen control operation evaluation report based on the verified and optimized operating control parameters.

[0015] According to another aspect of the present invention, a precise dissolved oxygen control system for aerobic wastewater treatment is provided, the system comprising: a data acquisition module, a characteristic parameter module, a time series parameter module, an analysis and correlation module, a matrix setting module, a control sequence module, and an evaluation report module; The data acquisition module is used to acquire multi-source sensing data, influent operating condition data, control cycle parameters and historical aeration operation data of the aerobic wastewater treatment system. The multi-source sensing data includes core data of the aerobic tank water environment and microbial metabolic state data, and preprocesses the core data of the water environment and microbial metabolic state data. The feature parameter module is used to preset and extract core feature parameters for dissolved oxygen control and supplementary feature parameters for microbial activity from the preprocessed data based on a multi-source data fusion feature extraction strategy. The timing parameter module is used to perform time-series partitioning of core feature parameters, supplementary feature parameters, influent operating condition data and historical aeration operation data based on control cycle parameters, resulting in four types of time-series sets: core control features, microbial activity, influent operating conditions and historical aeration. The analysis and association module is used to perform in-depth analysis of multivariate coupling relationships on four types of time series based on a long short-term memory network with attention mechanism, and to set multivariate temporal coupling association rules among the four types of time series; The matrix setting module is used to generate a multivariate coupled prediction matrix from four types of time series sets according to the coupling association rules, perform time series dynamic inference on the matrix, and generate the optimal dissolved oxygen setpoint within the future control step. The control sequence module is used to construct a closed-loop control model of the aeration system with feedforward compensation, feedback correction and disturbance suppression capabilities using model predictive control algorithms. It converts the optimal dissolved oxygen setpoint into the real-time control target parameter input model, calculates the optimal operating control parameters of the variable frequency fan, air diffusion device and liquid level regulating valve, and generates the optimal aeration control sequence. The evaluation report module is used to verify and dynamically optimize the execution effect of the optimal aeration control sequence and the accuracy of dissolved oxygen control in real time, verify the robustness of the system under complex disturbance conditions, and generate a dissolved oxygen control operation evaluation report based on the verified and optimized operating control parameters.

[0016] The beneficial effects of this invention are as follows: 1. This invention achieves a panoramic and precise characterization of DO concentration, oxidation-reduction potential, microbial metabolic activity, influent operating condition fluctuations, and historical aeration patterns in aerobic tanks through full-domain data acquisition and multi-dimensional feature fusion extraction from multi-source sensors in the sensing layer, combined with a refined data governance mechanism of time synchronization alignment and time-series standardization. This overcomes the shortcomings of traditional fixed PID regulation, offline detection feedback control, and single DO sensor control methods, which suffer from single data sources, inability to perceive microbial metabolic status in real time, and fuzzy representation of coupling relationships. At the same time, through the precise division of labor between long short-term memory networks with attention mechanisms and multivariate time-series coupling analysis, it achieves adaptive differentiation of control feature importance, in-depth mining of time-series dependencies, and scientific definition of multivariate association rules. This replaces the shortcomings of existing intelligent control technologies such as shallow feature extraction of traditional single models, static association settings, fuzzy control, and neural networks, improving the targeting and robustness of the control feature set.

[0017] 2. This invention relies on the deep integration of multivariate coupled prediction matrix and time-series dynamic reasoning, and improved model predictive control algorithm. Combined with the hierarchical design of feedforward compensation, feedback correction and disturbance suppression, it achieves accurate prediction of the optimal DO setpoint for future control step size and collaborative optimization of actuator parameters. It forms a control logic of dynamic weight allocation, predictive feedforward guidance and closed-loop real-time correction, which solves the problems of high or low DO concentration, large fluctuations and low control accuracy caused by strong lag and weak anti-disturbance ability of offline feedback control. It replaces the empirical aeration adjustment and fixed parameter control mode, and achieves millisecond-level precise control of DO concentration. The generated optimal aeration control sequence can accurately match the dissolved oxygen control requirements of different influent loads, different microbial states and different process stages, improve control stability, response speed and operating efficiency, and enhance the robustness of the system under complex operating conditions.

[0018] 3. This invention utilizes real-time monitoring of the optimal aeration control sequence execution effect and DO concentration control accuracy, combined with a two-dimensional collaborative evaluation system of control accuracy verification, robustness verification, and dynamic optimization adjustment, to construct a full-process closed-loop iterative mechanism. This enables dynamic adaptive updating of the dissolved oxygen control strategy, overcoming the shortcomings of traditional fixed PID regulation and single-sensor control that only focus on instantaneous adjustment effects, ignore long-term operating condition drift, model mismatch, and disturbance accumulation leading to control quality degradation. Simultaneously, through time series set verification, parameter archiving, and continuous update mechanisms for control rules, it continuously improves multi-variable coupling correlation rules and the aeration system operation feature library, forming a data-driven self-optimizing control capability. This ensures the stable adaptation of the dissolved oxygen control strategy to complex disturbance conditions such as influent load fluctuations, changes in microbial activity, and abnormal operation of aeration equipment throughout the entire operating cycle. 4. This invention employs an end-to-end control loop through a three-tiered control architecture of multi-source data fusion, dynamic model prediction, and intelligent execution optimization. Combined with the collaborative design of modular functional units, it ensures the portability of the technical solution, the flexibility of engineering implementation, and the efficiency of control. Furthermore, through a refined control optimization mechanism involving time sequence integrity verification, correlation conflict identification, and parameter adaptation correction, it avoids potential risks such as time sequence mismatch, feature conflicts, and uncoordinated actuator adjustments. This comprehensively improves the accuracy, stability, and safety of DO concentration control in aerobic wastewater treatment systems, preventing problems such as sludge aging, energy waste, and excessive ammonia nitrogen, ensuring wastewater meets discharge standards, and enhancing the stability and energy efficiency of aerobic treatment processes. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the 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.

[0020] Figure 1 This is a flowchart of a method for precise control of dissolved oxygen in aerobic wastewater treatment according to an embodiment of the present invention; Figure 2 This is a system block diagram of a dissolved oxygen precision control system for aerobic wastewater treatment according to an embodiment of the present invention.

[0021] In the picture: 1. Data acquisition module; 2. Feature parameter module; 3. Time series parameter module; 4. Analysis and correlation module; 5. Matrix setting module; 6. Control sequence module; 7. Evaluation report module. Detailed Implementation

[0022] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0023] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0024] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, the method for precise control of dissolved oxygen in aerobic wastewater treatment according to an embodiment of the present invention includes the following steps: S1. Acquire multi-source sensing data, influent operating condition data, control cycle parameters and historical aeration operation data of the aerobic wastewater treatment system. The multi-source sensing data includes core data of the aerobic tank water environment and microbial metabolic state data, and preprocess the core data of the water environment and microbial metabolic state data. Specifically, by integrating electrochemical DO sensors, fluorescence dissolved oxygen probes, and microbial metabolic thermal imagers into the sensing layer of the aerobic wastewater treatment system, multi-source sensing data, influent operating condition data, control cycle parameters, and historical aeration operation data are collected simultaneously. The multi-source sensing data covers core data of the aerobic tank water environment (dissolved oxygen concentration, oxidation-reduction potential, etc.) and microbial metabolic state data (microbial heat production signals, etc.). Influent operating condition data includes key water quality parameters such as influent COD and ammonia nitrogen concentration. Control cycle parameters are preset according to process requirements. Historical aeration operation data are retrieved from the system's historical database. After collection, the core data of the aerobic tank water environment and the microbial metabolic state data are preprocessed to remove interference and redundant information, correct outliers, and complete missing data. At the same time, all types of data are standardized to a unified dimension to ensure data integrity and consistency.

[0025] S2. Based on a multi-source data fusion feature extraction strategy, extract the core feature parameters for dissolved oxygen control and supplementary feature parameters for microbial activity from the preprocessed data. Specifically, based on the requirements of aerobic wastewater treatment processes, the characteristics of multi-source sensing data, and the goal of precise dissolved oxygen control, a hierarchical fusion extraction strategy with principal component analysis (PCA) and mutual information feature selection as the core is pre-set. The fusion principles of dimensionality reduction and redundancy removal, high correlation screening, and retention of time-series features are clarified. The feature screening criteria are set as follows: mutual information correlation degree ≥ 0.7 as the core feature screening threshold, mutual information correlation degree ≥ 0.5 as the supplementary feature screening threshold, and feature redundancy degree ≤ 0.8 as the retention threshold. A three-level extraction process of initial extraction of single-source features, fusion of multi-source features, and collaborative verification of features is determined.

[0026] During the pre-processing process, considering the heterogeneity of core data on the aerobic pool water environment (dissolved oxygen concentration, oxidation-reduction potential, etc.) and microbial metabolic state data (microbial heat production signals, etc.), this hierarchical fusion extraction mode was determined to be adopted. First, single-source features of the two types of data were initially extracted using algorithms. Then, multi-source data fusion was achieved through feature concatenation. At the same time, redundant and irrelevant features were removed by the above-mentioned quantitative screening threshold to ensure extraction efficiency and feature targeting.

[0027] Based on this pre-set strategy, feature extraction is first performed on the core data of the aerobic pool water environment after pretreatment. The correlation between the data and dissolved oxygen control is used as the core criterion. The PCA algorithm is used to reduce the dimensionality and remove redundancy of the core water environment data, retaining the time-series features with a principal component contribution rate of ≥85%. The focus is on capturing the core features directly related to dissolved oxygen control, and four core feature parameters of dissolved oxygen control are selected: the time-series change rate of dissolved oxygen concentration, the steady-state fluctuation value of oxidation-reduction potential (ORP), the time-series change of water pH value, and the gradient difference of dissolved oxygen concentration. These parameters directly determine the accuracy and response speed of dissolved oxygen regulation. Then, supplementary feature extraction is performed on the pretreated microbial metabolic state data, focusing on the correlation features between microbial activity and dissolved oxygen demand.

[0028] Then, the correlation between microbial metabolic data and microbial oxygen demand is calculated using a mutual information feature selection algorithm. Four supplementary microbial activity feature parameters are extracted: the temporal fluctuation coefficient of microbial heat production signal, the duration of peak microbial metabolic activity, the rate of change of microbial heat production gradient, and the sludge respiration rate. These parameters are used to assist in determining the optimal dissolved oxygen setting. During the extraction process, a preset feature concatenation and fusion rule is used to concatenate the core feature parameters extracted from single sources with the supplementary feature parameters along the time dimension to form a fused feature set. The fused feature set is then normalized. Simultaneously, the Pearson correlation coefficient between features is calculated for collaborative verification, and overlapping features with redundancy ≥0.8 are removed to ensure the accuracy of the core feature parameters and the correlation of the supplementary feature parameters. Finally, a feature parameter set with clear dimensions, no redundancy, and high correlation is obtained.

[0029] S3. Based on the control cycle parameters, the core characteristic parameters, supplementary characteristic parameters, influent operating condition data and historical aeration operation data are divided into time series to obtain four types of time series sets: core control characteristics, microbial activity, influent operating conditions and historical aeration. In this embodiment of the invention, the step of dividing the core feature parameters, supplementary feature parameters, influent operating condition data, and historical aeration operation data into four time series sets based on control cycle parameters to obtain four types of time series sets: core control features, microbial activity, influent operating conditions, and historical aeration, includes the following steps: S31. Using the control cycle parameter as a unified time reference, perform time synchronization and alignment on various parameters, match the timing sampling frequency with the control cycle, and determine the time window for dividing the control cycle. Specifically, using the preset control cycle parameters as a unified time reference, the core characteristic parameters of dissolved oxygen control, the characteristic parameters of microbial activity replenishment, the influent operating condition data, and the historical aeration operation data are synchronized and the sampling frequency is adapted to determine the time window for dividing the control cycle.

[0030] First, retrieve the original timestamp information of the four types of data. Using the time unit (such as minutes or hours) corresponding to the control cycle parameter as the standard, calibrate the time axis of all data. Align all types of parameters at different collection time points to the same time scale to eliminate time deviations caused by differences in collection equipment and collection frequency, and ensure that the four types of data correspond one-to-one at the same time node.

[0031] Subsequently, the original timing sampling frequencies of various parameters were analyzed and compared with the control cycle parameters. Data with excessively high sampling frequencies were downsampled, and data with excessively low sampling frequencies were interpolated to complete the data. This ensured that the timing sampling frequencies of the four types of parameters were perfectly matched with the control cycle, guaranteeing that there was complete and corresponding data for each type of parameter in each control cycle.

[0032] Finally, based on the four types of data after synchronization and frequency matching, and combined with the control cycle parameters, the start and end timestamps of the control cycle time window are determined, and the duration and interval of each time window are made consistent with the control cycle.

[0033] S32. After synchronization and alignment, various parameters are segmented periodically according to the time window, and four types of time series sets are generated according to the data type, and the time dimension of each type of time series set remains consistent. As a preferred embodiment, the step of periodically segmenting various parameters after synchronization and alignment according to a time window, generating four types of time series sets according to data type, and ensuring that the time dimension of each type of time series set remains consistent includes the following steps: S321. Clearly define the start and end timestamps of the control cycle time window and match them with the sampling frequency of various parameters after synchronization and alignment; Specifically, firstly, based on the preset control cycle parameters and the operating rhythm of the aerobic wastewater treatment process, the baseline starting point for dividing the time window is determined. The start time of the first complete control cycle after the system has been operating stably is used as the initial timestamp. The start and end timestamps of the time window for each control cycle are calculated sequentially to ensure that the duration of each time window is strictly equal to the control cycle, and that adjacent time windows do not overlap or have gaps, thus ensuring the continuity and standardization of the time window division.

[0034] Subsequently, the actual sampling frequencies of the four types of data after time synchronization and alignment are retrieved and compared with the standard sampling frequencies corresponding to the control cycle to verify that the sampling frequencies of each type of data match the control cycle. If there are slight deviations, the start and end timestamps of the time windows are fine-tuned so that each time window contains exactly one complete set of sampling data for each type of parameter, ensuring that the time window division is highly adapted to the data sampling frequency. Through the above operations, the specific time range of the time window division for each control cycle is determined, achieving precise matching between the time window and the sampling frequencies of the four types of data.

[0035] S322. Based on the matching results, segment and extract various parameters periodically to ensure that each segment of data corresponds to a unique control period. Specifically, the time window and start and end timestamps are divided according to the established control cycle. The four types of data are traversed according to a unified time scale. The time axis is divided sequentially according to the length of the control cycle, so that each segment strictly corresponds to an independent control cycle.

[0036] During the segmentation process, the sampling frequency matching results serve as constraints, and each type of data is synchronously truncated according to the same time window boundary. This ensures that the time span, time sequence length, and number of sampling points of the four types of data within the same period are completely consistent. For high-frequency acquired data, valid time sequences are extracted periodically, and for low-frequency data, time sequence points are supplemented periodically, ensuring that the segmented data has no overlap, omissions, or misalignments. After truncating, a period uniqueness check is performed to verify that the timestamp interval of each data segment corresponds one-to-one with the control period number, avoiding cross-period mixing and missing data. Through the above-mentioned precise period-by-period truncating and uniqueness constraints, segmented data with regular time sequence and clear periodicity is formed.

[0037] S323. The extracted parameters are categorized into four time series sets, and the output is generated after verifying that the time dimension is consistent.

[0038] Specifically, all the extracted core characteristic parameters of dissolved oxygen control are collected sequentially according to the control cycle to form a core control characteristic time series set. Then, the extracted supplementary characteristic parameters of microbial activity are collected in a unified manner to form a microbial activity time series set. The influent operating condition data and historical aeration operation data are collected into an influent operating condition time series set and a historical aeration time series set, respectively. All four types of time series sets are arranged according to the control cycle number, retaining complete data information within each cycle.

[0039] After the data collection is completed, the key verification is to ensure that the time dimensions of the four types of time series sets are consistent. By checking the number of control cycles, the timestamp range of each cycle, the time series length and the number of sampling points of each type of time series set, it is confirmed that the time axes of the four types of time series sets are completely synchronized, and the four types of data corresponding to each control cycle are one-to-one, without any time misalignment, missing cycles or inconsistent lengths. If any deviation in the time dimension is found, the collection process is traced and adjusted in time to ensure that the time dimension of the time series sets is consistent. After the verification is passed, the four types of time series sets are output synchronously.

[0040] S33. Verify and adjust the timing length and data integrity of the four types of timing sets to complete the timing partitioning.

[0041] Specifically, the time series lengths of the four types of time series sets are checked item by item to verify whether the total time series span, single period segment length, and number of sampling points are completely consistent. The checks are conducted to identify any deviations such as missing periods, overlapping time series, or uneven lengths, ensuring that the four types of time series sets are strictly aligned in the time dimension. At the same time, data integrity checks are performed period by period to check whether there are any issues such as missing data, outliers, null values, or duplicate sampling within each time series set, ensuring that each segment of time series data is continuous, valid, and distortion-free.

[0042] For the time series length deviations discovered during verification, uniform corrections were made using truncation, interpolation, or resampling methods. For issues such as missing or abnormal data, repairs were completed through reasonable interpolation, anomaly removal, and valid value replacement. This ensured that the lengths of the four types of time series sets matched and the data was standardized and complete after adjustment. After multiple rounds of verification and correction, it was confirmed that the time series lengths of the four types of time series sets were uniform, the data was complete and valid, the time axes were synchronized and aligned, and there were no deviations, missing data, or anomalies. This fully met the requirements for subsequent multivariate coupling analysis and model input, thus completing the entire time series partitioning process.

[0043] S4. Based on the attention-based long short-term memory network, a deep analysis of the multivariate coupling relationship of four types of time series sets is carried out, and multivariate temporal coupling association rules are set among the four types of time series sets. In this embodiment of the invention, the method of performing deep analysis of multivariate coupling relationships on four types of time series based on a long short-term memory network with attention mechanism, and setting multivariate temporal coupling association rules among the four types of time series, includes the following steps: S41. Combining the characteristics of multivariate time series data of four types of time series sets with the requirements of aerobic wastewater treatment process, configure the model input layer, attention weight allocation layer and long short-term memory network hidden layer to construct a long short-term memory network model with attention mechanism. Specifically, a deep analysis was first conducted on four types of time series sets to extract variable dimensions, time series length, sampling frequency, data distribution characteristics, and time series correlation characteristics among multiple variables. Long-term and short-term dependence patterns and key influencing factors were identified. At the same time, in combination with the aerobic treatment process operation requirements, dissolved oxygen control targets, control cycle parameters, and tolerance range for complex disturbances, the input constraints, output indicators, and accuracy requirements of the model were clarified. Among them, the core control targets of the model were clarified as follows: dissolved oxygen steady-state control deviation ≤ ±0.3 mg / L, dynamic response lag time ≤ 1 control cycle, and overshoot ≤ 10% under complex disturbance conditions.

[0044] In the model structure configuration, the input layer adopts a 4-channel parallel design, with the core control feature time series set, microbial activity time series set, water intake condition time series set, and historical aeration time series set as independent input channels. The single-channel input dimension matches the number of feature parameters of the corresponding time series set, and the input layer time series length is consistent with the single window time series length of the control cycle division, so that the input dimension and time series length accurately match the data characteristics. The attention weight allocation layer adopts the Bahdanau additive attention mechanism, setting 4 weight branches that correspond one-to-one with the input channels, and the output dimension matches the hidden layer dimension. This is used to adaptively weight the feature importance of different time series sets and different time series nodes, and strengthen the representation ability of key control features.

[0045] The Long Short-Term Memory (LSTM) network hidden layers employ a two-layer bidirectional LSTM structure. The first hidden layer has 64 neurons, and the second hidden layer has 32 neurons. The activation function is the Tanh function, and the recurrent activation function is the Sigmoid function. The initial value of the forget gate bias is set to 1.0 to avoid gradient vanishing in the early stages of training and to efficiently capture temporal variation patterns and multivariate coupling relationships. The model output layer is a fully connected layer with the same number of neurons as the core feature parameters, outputting the quantitative results of multivariate coupling strength and temporal dependence. Finally, the input layer, attention weight allocation layer, LSTM network hidden layers, and output layer are cascaded in an ordered manner according to data flow logic. The Adam optimizer is used to initialize the parameters, with an initial learning rate of 0.001, a batch size of 32, and 200 iterations. An early stopping mechanism is also set to prevent overfitting. After structural integration, an attention-based LSTM network model adapted for dissolved oxygen control in aerobic wastewater treatment is constructed.

[0046] S42. After normalizing and preprocessing the four types of time series sets, they are used as multi-channel input data and fed into the long short-term memory network model. Specifically, data statistics and feature analysis were first performed on the four types of time series sets to clarify the numerical range, distribution characteristics, and temporal patterns of each time series parameter. To address the issues of inconsistent data units and significant numerical differences, the Z-score standardization and normalization method was adopted, according to the formula... This method maps all feature parameters to a unified interval with a mean of 0 and a variance of 1, eliminating the interference of differences in units and magnitudes on model calculations, while preserving time-series trends and multivariate correlation features. In the formula, The standardized feature parameter values, i.e., the standardized feature data input to the LSTM model with attention mechanism after Z-score normalization, are the input after Z-score normalization. For the four time series sets corresponding to core control features, microbial activity, influent conditions, and historical aeration, the standardized results of each feature parameter (such as dissolved oxygen concentration, oxidation-reduction potential (ORP), influent COD, microbial heat generation signal, and blower operating frequency) at the corresponding time series node are finally mapped to a unified interval with a mean of 0 and a variance of 1.

[0047] The original measured values ​​of characteristic parameters before standardization, i.e., the original collected / calculated values ​​of each characteristic parameter under the corresponding control cycle and time node in the four types of time series. For example, the on-site measured value of dissolved oxygen concentration, the laboratory value of influent COD concentration, the fluctuation coefficient of microbial heat generation signal, the historical fan operating frequency and other unprocessed raw data within a certain control cycle.

[0048] The historical mean of the corresponding characteristic parameter is the arithmetic mean of all valid measured data for that characteristic parameter within the historical operating cycle of the aerobic wastewater treatment system. This value is pre-calibrated based on the historical operating data of the system's hierarchical classification and is periodically updated as the system operates. Its core function is to eliminate the baseline offset of the characteristic parameter and ensure that the standardized data is not affected by long-term operating condition drift.

[0049] The historical standard deviation of the corresponding characteristic parameter is the standard deviation of all valid measured data for that characteristic parameter within the historical operating cycle of the aerobic wastewater treatment system. It characterizes the historical fluctuation range of this characteristic parameter and is a core parameter in Z-score standardization for measuring data dispersion; it is calibrated and updated synchronously with the historical mean μ.

[0050] During preprocessing, outlier removal, missing value interpolation and completion, and data smoothing are performed simultaneously. The 3σ criterion is used to remove outliers exceeding ±3 standard deviations, linear interpolation is used to complete missing values, and moving averages are used for data smoothing to ensure the continuity, integrity, and validity of each time series set. Subsequently, the four normalized time series sets are encapsulated into independent channels, each corresponding to a parallel input channel. Each channel maintains complete consistency in time series length, timestamp, and sampling frequency, forming a synchronously aligned multi-channel parallel input structure. Finally, following the interface specifications of the model input layer, the multi-channel input data is synchronously fed into the Long Short-Term Memory (LSTM) network model with an attention mechanism in temporal order. This ensures that the four time series sets enter the model simultaneously for subsequent processing, providing high-quality, standardized data input for adaptive attention weight allocation, in-depth mining of temporal dependencies, and multivariate coupling analysis.

[0051] S43. The attention weight allocation layer is used to assign feature importance weights to the four types of time series sets. The hidden layer is used to capture the temporal dependencies, and the deep analysis of multivariate coupling relationships is carried out to output the coupling analysis results. In this embodiment of the invention, the steps of assigning feature importance weights to four types of time series sets through an attention weight allocation layer, capturing temporal dependencies using hidden layers, conducting deep analysis of multivariate coupling relationships, and outputting coupling analysis results include the following steps: S431. Initialize the attention weight allocation layer, using the influence of the four types of time series on dissolved oxygen control as the core basis for weight allocation, so that the weight allocation fits the process control requirements. Specifically, during the initialization phase, the basic parameters of the attention weight allocation layer, such as the weight matrix, bias term, activation function, and learning rate, are first initialized with default values ​​and random constraints. The weight matrix is ​​initialized using a uniform distribution of Xavier, the bias term is initialized to 0, and the activation function is the Softmax function, to ensure that the initial state is stable and facilitates subsequent iterative optimization.

[0052] Simultaneously, the core criteria for weight allocation were clarified. Combining prior knowledge of aerobic wastewater treatment processes and biochemical reaction control mechanisms, the contribution priorities and initial prior weights of four types of time series sets to dissolved oxygen concentration regulation were quantitatively set: The first priority is the core control characteristic time series set (including direct control parameters such as dissolved oxygen (DO) concentration and oxidation-reduction potential (ORP), which has the highest direct impact on dissolved oxygen control, with an initial prior weight set of 0.45; the second priority is the influent operating condition time series set (including disturbance parameters such as influent COD, ammonia nitrogen, and water volume), which is the core preceding influencing factor of dissolved oxygen supply and demand fluctuations, with an initial prior weight set of 0.25; the third priority is the microbial activity time series set (including metabolic parameters such as microbial heat production signals and sludge respiration rate), which directly reflects the actual oxygen demand of microorganisms, with an initial prior weight set of 0.20; the fourth priority is the historical aeration time series set (including execution parameters such as historical fan frequency and valve opening), which is used to characterize the response characteristics of the aeration system, with an initial prior weight set of 0.10; the sum of the initial prior weights of the four types of time series sets is 1.

[0053] Building upon this foundation, the dissolved oxygen control target, control cycle, disturbance tolerance range, and biochemical reaction constraints of aerobic wastewater treatment are integrated, transforming process operation requirements into weighted allocation constraints: For the aerobic nitrification stage, the weight of dissolved oxygen concentration should not increase by more than 0.2, and the weight of influent ammonia nitrogen should not increase by more than 0.15; for the organic matter degradation stage, the weight of microbial metabolic activity should not increase by more than 0.15, and the weight of influent COD should not increase by more than 0.1; the lower limit for single-time-series set weight adjustment should not be less than 50% of the initial prior weight, and the upper limit should not exceed twice the initial prior weight, ensuring that the weight distribution always tilts towards the feature set that is sensitive to dissolved oxygen changes and strongly correlated with oxygen demand, thus mitigating the impact of redundant interference features. After initialization, the rationality of the weights is verified to ensure that the weight allocation is consistent with the actual control logic and highly adapted to process requirements.

[0054] S432. Input the four types of time series sets into the attention weight allocation layer, perform feature differentiation weight allocation and complete the assignment; Specifically, the core control feature time series set, microbial activity time series set, influent operating condition time series set, and historical aeration time series set, which have been normalized and preprocessed and aligned with multiple channels, are used as four independent time series sets to be synchronously input into the attention weight allocation layer of the long short-term memory network model with attention mechanism. Feature-differentiated weight allocation is carried out and standardized weight assignment is completed.

[0055] The core allocation criteria are the degree of influence of four types of time series sets on dissolved oxygen regulation and the strength of their correlation with microbial oxygen demand. Combined with prior knowledge of aerobic wastewater treatment processes, control objectives, and disturbance constraints, the importance of features in each time series set is quantitatively assessed, distinguishing between dominant control features, auxiliary supporting features, and disturbance reference features. The attention weight allocation layer, based on initialized prior weights and process constraint rules, is calculated according to the Bahdanau attention mechanism formula. ; in, The attention score at time t is the core intermediate output value of the attention weight allocation layer. It is used to characterize the importance of the feature vectors of the four time series sets at time t to the precise control of dissolved oxygen. The higher the score, the higher the contribution of the corresponding feature at that time to the current dissolved oxygen control. Subsequently, the score will be normalized by the Softmax function to obtain the final attention weight coefficients of the four time series sets and each time series node.

[0056] This is the transpose of the weight vector output by the hidden layer of the attention mechanism. It represents the trainable parameters of the model, and its dimension perfectly matches the output dimension of the tanh activation function. Its core function is to map the nonlinear features output by the tanh function into a single-dimensional attention score. During model training, it iterative optimization is performed in conjunction with the dissolved oxygen control accuracy target for aerobic wastewater treatment to ensure that the weight allocation always aligns with the process control requirements.

[0057] The hyperbolic tangent activation function is a non-linear activation function in the attention mechanism. Its core function is to perform a non-linear transformation on the linear combination features of the input, mapping the output value to the (-1, 1) interval, thus solving the problem of insufficient feature representation ability of pure linear transformation and capturing the non-linear correlation between the four types of time-series features and the dissolved oxygen control target.

[0058] The weight matrix is ​​the time-series feature vector and the trainable parameters of the model. Its core function is to perform a linear transformation on the input time-series feature vector ht at time t to adapt to the dimensionality requirements of the LSTM hidden layer. During the model training process, iterative optimization is performed by combining the influence of the four types of time-series sets on dissolved oxygen control to enhance the weight mapping effect of the core control features.

[0059] The time series feature vector at time t is the multi-channel input data of the input layer of the LSTM model with attention mechanism. It is the fused feature vector of the core control feature time series set, microbial activity time series set, influent operating condition time series set, and historical aeration time series set at time t and under the same control cycle. It contains the time series information of all core features such as dissolved oxygen concentration, ORP, microbial heat production signal, influent water quality and quantity, and historical aeration parameters.

[0060] The weight matrix represents the hidden layer state at the previous time step. It consists of trainable parameters of the model and its core function is to adjust the weights of the LSTM hidden layer state at the previous time step. Linear transformation is performed to achieve cross-cycle transmission of temporal context information; during model training, iterative optimization is combined with the temporal dependence characteristics of dissolved oxygen control to accurately capture the parameter linkage patterns of water inflow-aeration-microbial metabolism-dissolved oxygen changes across control cycles.

[0061] The output state of the LSTM hidden layer at time t-1 (the previous time) contains contextual information of all historical time-series features before the current control cycle, as well as the historical response patterns of dissolved oxygen control. Its core function is to provide a time-dependent reference for the calculation of the attention score at the current time, reflecting the temporal continuity of the changes of various parameters during the aerobic treatment of wastewater.

[0062] The bias term of the attention mechanism is a trainable parameter of the model, and its core function is to correct the baseline offset of the linearly transformed features, thereby improving the model's fitting ability and convergence stability. It is initialized in the model initialization stage according to the Xavier uniform distribution, and it is iteratively optimized with the dissolved oxygen control target during the training process.

[0063] During the weighting process, for the time series set of core control features directly related to dissolved oxygen changes, the weights are dynamically adjusted based on the deviation between the real-time DO concentration and the set value, with a higher weight for a larger deviation; for the time series set of microbial activity reflecting metabolic needs, the weights are adjusted based on the fluctuation amplitude of microbial heat production signals, with a higher weight for a larger fluctuation in metabolic activity; for the time series set of influent operating conditions characterizing influent fluctuations, the weights are adjusted based on the abrupt change amplitude of influent water quality and quantity, with a higher weight for a larger fluctuation amplitude; and for the historical aeration time series set reflecting historical patterns, the weights are adjusted based on the response lag characteristics of the aeration system, achieving adaptive differentiation and differential allocation of feature importance. After weighting, the weight coefficients are normalized and mapped to the [0,1] interval to ensure that the sum of the weights of the four time series sets is 1, so that the weight distribution conforms to the control logic and meets the requirements of model operation. Through differentiated weighting and assignment, the ability to represent key features is strengthened, and the interference of redundant information is weakened, so that the weight allocation is highly aligned with the requirements of precise dissolved oxygen regulation.

[0064] S433. Input the four types of time series sets after weight assignment into the hidden layer, capture the changing patterns of feature parameters periodically, explore the correlation and temporal dependency relationships among the four types of time series sets, and obtain preliminary analysis results of coupling relationships. Specifically, the core control feature time series set, microbial activity time series set, water intake condition time series set, and historical aeration time series set, after completing the differentiated assignment of attention weights, are synchronously input into the hidden layer of the long short-term memory network in a multi-channel parallel structure. Relying on the hidden layer's feature extraction and long-term memory capabilities for time series data, the time series change patterns are captured periodically and the coupling correlations between multiple variables are explored to obtain preliminary analysis results of the multivariate coupling relationships.

[0065] Using the control cycle as the smallest analytical unit, the system iterates through various time series sets along the time axis periodically to accurately extract the fluctuation trends, mutation nodes, stable intervals, and periodic evolution patterns of the characteristic parameters within a single time series set, fully preserving the time series change characteristics and key state information of each parameter. Based on this, through the interactive transmission and linkage operation of hidden layer neurons, the system simultaneously analyzes the collaborative change characteristics and cross-cycle transmission response relationships of the four types of time series sets within the same control cycle. It focuses on exploring the dynamic correlation strength, influence transmission paths, and time series lag dependence patterns between core control characteristics, microbial activity, influent conditions, and historical aeration, and distinguishes strong coupling, weak coupling, and independent action relationships among multiple variables: parameter pairs with an absolute value of Pearson correlation coefficient ≥ 0.7 are judged as strongly coupled, those with an absolute value of correlation coefficient < 0.7 are judged as weakly coupled, and those with an absolute value of correlation coefficient < 0.3 are judged as independent and uncoupled. Then, the internal variation patterns of a single time series set are integrated and summarized with the interrelationships and temporal dependencies among multiple variables to form preliminary analytical results of multivariate coupling relationships covering feature evolution, coupling strength, and dependency characteristics, which comprehensively reflect the inherent laws and interaction mechanisms of the four types of time series sets.

[0066] S434. Integrate and organize the preliminary analysis results to form coupled analysis results.

[0067] Specifically, the internal feature change patterns, inter-variable correlations and temporal dependencies of various time series contained in the preliminary analysis results are comprehensively extracted, classified and collected. The data are then sorted in layers according to dimensions such as single time series feature evolution, cross-variable correlation strength, and time series lag response characteristics. Redundant, redundant and invalid analysis information is removed, and core coupling features that are strongly correlated with dissolved oxygen control are retained.

[0068] Based on this, various relationships were quantitatively calibrated and logically regulated. The coupling strength, influence direction, and transmission path among four time series sets—core control characteristics, microbial activity, influent conditions, and historical aeration—were clarified. Simultaneously, the lag period, response patterns, and dynamic change characteristics of time series dependencies were defined. The response lag time between different parameters was quantified to be 1-3 control cycles, ensuring clear and accurate quantification of various relationships. Combining the constraints of aerobic wastewater treatment processes and dissolved oxygen control targets, the results were rationally verified and corrected to ensure that the coupling relationships conformed to the actual biochemical reaction mechanisms and control logic of aerobic microbial degradation of organic matter and nitrification. After integration, refinement, and verification, a systematic coupling analysis result was formed, encompassing single-time series characteristic patterns, multivariate coupling correlation matrices, and cross-period time series dependency characteristics. This comprehensively and accurately reflects the intrinsic mechanisms and temporal evolution characteristics among multiple variables.

[0069] S44. Based on the coupling analysis results and process requirements, clarify the correlation threshold and timing response relationship of the characteristic parameters of each time series set, and set multivariate timing coupling correlation rules among the four types of time series sets.

[0070] Specifically, firstly, combining the core conclusions from the coupling analysis results, such as the correlation strength, coupling direction, and time-dependent lag period among the characteristic parameters, and overlaying them with actual operational requirements such as process control range, microbial biochemical reaction conditions, influent load fluctuation range, and aeration regulation constraints, the system quantifies and determines the synergistic change amplitude, linkage trigger boundary, and stable operating range among different parameters. It clarifies the threshold boundaries for strongly correlated, weakly correlated, and independent parameters, ensuring that the correlation thresholds align with actual operating conditions and control objectives. The system defines the time-series response relationships between parameters, clarifies positive or negative linkage characteristics, response lag period, duration of sustained stability, and decay patterns, and establishes the corresponding logic for multi-parameter time-series transmission and dynamic response. Subsequently, the correlation thresholds, time-series response relationships, and process constraints are integrated and encapsulated to form multi-variable time-series coupling correlation rules covering parameter linkage, time-series transmission, and operating condition adaptation. The system clarifies the coupling logic and control guidance of four types of time-series sets under different operating conditions. Finally, through historical operating data verification and simulation testing, the rules are validated for rationality and adaptively corrected to ensure accuracy, stability, and adaptability to complex disturbance conditions.

[0071] The multivariate temporal coupling association rules specifically include the following four types of executable quantization rules: Pre-linkage rules for water intake conditions: Using the water intake condition time series as the pre-trigger factor, establish linkage rules between water intake parameters and core control features and weight allocation.

[0072] 1. When the single-cycle fluctuation range of influent COD concentration is ≥20%, or the single-cycle fluctuation range of influent flow rate is ≥15%, a strong disturbance linkage is triggered. The weight of the influent operating condition time series set is increased by 30%-50%, and the weight of the core control characteristic time series set is increased by 20% simultaneously. At the same time, it is specified that for every 100mg / L increase in influent COD, the corresponding dissolved oxygen setpoint must be increased by 0.3-0.5mg / L simultaneously, and the response lag period shall not exceed one control cycle. 2. When the influent ammonia nitrogen concentration fluctuates by ≥30% in a single cycle, the nitrification reaction is triggered. The weight of the influent operating condition time series is increased by 20%-40%, and the weights of DO concentration and ORP parameters in the core control characteristics are increased by 30% simultaneously. It is clear that for every 10 mg / L increase in influent ammonia nitrogen, the corresponding dissolved oxygen setpoint needs to be increased by 0.2-0.4 mg / L simultaneously, and a stable supply needs to be maintained for 3 consecutive control cycles. 3. When the fluctuation range of influent water quality and quantity in a single cycle is less than 10%, it is determined to be a steady-state condition. The weight of the influent condition time series set remains unchanged from the initial prior weight, and no additional linkage adjustment is triggered.

[0073] Microbial activity-dissolved oxygen demand matching association rules: Using the time series of microbial activity as the basis for oxygen demand determination, a matching rule between metabolic characteristics and dissolved oxygen control is established.

[0074] 1. When the fluctuation coefficient of microbial heat production signal is ≥0.2, or the sludge respiration rate increases by ≥20% compared to the baseline value, it is determined that the microbial metabolic activity has increased. The weight of the microbial activity time series set is increased by 30%-40%, and the corresponding dissolved oxygen set value is increased by 0.2-0.4 mg / L to ensure that the oxygen supply matches the microbial oxygen demand. 2. When the fluctuation coefficient of the microbial heat production signal is ≤0.05 and the sludge respiration rate is ≥30% lower than the baseline value, it is determined that the microbial metabolic activity is reduced. The weight of the microbial activity time series set is reduced by 20%-30%, and the corresponding dissolved oxygen set value is reduced by 0.1-0.3mg / L. Under the premise of meeting the biochemical reaction requirements, the aeration energy consumption is avoided. 3. When the peak duration of microbial metabolic activity lasts for ≥2 control cycles, the continuous oxygen supply linkage is triggered, locking the lower limit of dissolved oxygen setpoint to no less than 1.5 mg / L to prevent a sudden drop in DO concentration from inhibiting microbial activity.

[0075] Closed-loop linkage rules for core control features: Using the time series set of core control features as the core of closed-loop control, establish linkage constraint rules for DO and ORP parameters.

[0076] 1. When the real-time DO concentration deviates from the set value by ≥±0.5mg / L, the core weight strong linkage is triggered, and the weight of the core control feature time series set is increased by 40%-60%, prioritizing the closed-loop correction of DO concentration. For every 0.2mg / L increase in deviation, the corresponding aeration adjustment step size is increased by 10%. 2. When the steady-state fluctuation value of ORP exceeds the ±50mV range, the auxiliary linkage is triggered, the weight of ORP parameter is increased by 30%, and the matching between DO concentration and ORP is checked simultaneously. When ORP continues to decrease and DO concentration does not change significantly, it is determined that there is a hidden interference in the system. The weight of the influent operating conditions and microbial activity time series set is increased by 15% simultaneously to investigate the source of disturbance. 3. When the steady-state deviation of DO concentration is ≤ ±0.3 mg / L and the ORP fluctuation value is ≤ ±20 mV, it is determined to be in control steady state. The weights of the time series set of core control features remain stable, and the linkage rules of each parameter enter a low-amplitude fine-tuning mode.

[0077] Historical aeration lag response association rules: Using the historical aeration time series as a reference for system characteristics, establish lag matching rules between aeration execution and dissolved oxygen response.

[0078] 1. Define the inherent lag period of the aeration system as 1-2 control cycles. When the historical aeration adjustment is continuously increased by ≥20% within 2 control cycles, but the DO concentration does not increase significantly, trigger the lag correction, increase the weight of the historical aeration time series by 25%, suspend the continuous increase of aeration, wait for the system response, and avoid overshoot. 2. When the correlation coefficient between historical aeration frequency and DO concentration is ≥0.8, the aeration system is judged to be responding normally, and the weight of the historical aeration time series set remains unchanged from its initial value; when the correlation coefficient is <0.5, the aeration system is judged to be responding abnormally, the weight of the historical aeration time series set is reduced by 30%, and the weight of the core control feature time series set is increased simultaneously, prioritizing the correction control strategy based on real-time DO concentration. 3. Based on historical aeration data, the frequency of the variable frequency fan is calibrated so that for every 1 Hz increase in frequency, the corresponding increase in DO concentration is 0.1-0.2 mg / L. When the actual response amplitude deviates from the benchmark value by ≥30%, the aeration system characteristic correction is triggered, and the feature weights and response benchmarks of the historical aeration time series are updated.

[0079] S5. Based on the coupling association rule, generate a multivariate coupling prediction matrix from the four types of time series sets, perform time series dynamic reasoning on the matrix, and generate the optimal dissolved oxygen setpoint within the future control step. In this embodiment of the invention, the step of generating a multivariate coupled prediction matrix from four types of time series sets according to the coupling association rule, and performing time-series dynamic inference on the matrix to generate the optimal dissolved oxygen setpoint within the future control step includes the following steps: S51. Based on the coupling association rule, clarify the association dimension of the feature parameters of the four types of time series sets, determine the row and column dimensions of the multivariate coupling prediction matrix, the row dimension corresponds to the control step size, and the column dimension corresponds to the core feature parameters of the four types of time series sets. Specifically, based on the coupling strength, influence transmission path, and temporal dependency of each feature parameter in the multivariate temporal coupling correlation rules, the correlation dimensions within and between the four types of time series sets are defined hierarchically. The dominant control features, auxiliary support features, and weakly correlated redundant features are distinguished. The core feature parameters that play a key role in dissolved oxygen prediction and regulation are selected, and irrelevant or redundant parameters are eliminated to ensure that the correlation dimension settings fit the coupling rules and control requirements.

[0080] In determining the matrix dimensions, the row dimension is based solely on the future control step size. Combining the control cycle parameters and prediction duration requirements, the total number of rows and the time scale of the matrix are clearly defined. Each row corresponds to an independent control step size, ensuring that the row dimensions progress continuously in time sequence without overlap or interruption. The column dimension is arranged in an orderly manner according to four types of time series sets. Each column corresponds to a core feature parameter, which includes dissolved oxygen control core features, microbial activity replenishment features, influent operating condition features, and historical aeration operation features in sequence. The total number of column dimensions is consistent with the number of core feature parameters after screening. Subsequently, the row and column dimensions are checked for consistency to ensure that the row dimension time series is standardized, the column dimension parameters are complete and clearly classified, and highly matched with the multivariate time series coupling association rules. Finally, a multivariate coupling prediction matrix framework with a standard structure and reasonable dimensions is formed.

[0081] S52. Integrate the feature parameters of the four types of time series sets according to the coupling association rules, preset the corresponding dimension empty matrix and fill the feature parameters to generate a multivariate coupled prediction matrix. In this embodiment of the invention, the step of integrating the feature parameters of four types of time series sets according to coupling association rules, pre-setting an empty matrix of the corresponding dimension and filling it with feature parameters to generate a multivariate coupled prediction matrix includes the following steps: S521. Clarify the correlation relationship of the four types of time series feature parameters in the coupling association rule, and determine the feature parameter integration rule by combining the row and column dimensions of the prediction matrix; Specifically, based on the coupling strength, influence transmission direction, time-dependent period and linkage response characteristics defined in the coupling association rules, the evolution of internal features of the four types of time series sets and the synergistic constraints between cross sets are sorted out in layers. Strong coupling dominant features, weak coupling auxiliary features and association constraint boundaries are distinguished to ensure that the association accurately reflects the dissolved oxygen regulation mechanism and process operation logic.

[0082] Based on a clear understanding of the relationships, and taking the control step size as the row dimension and the core feature parameters of the four time series sets as the column dimension of the multivariate coupled prediction matrix as the benchmark, a unified integration rule is formulated: taking the control step size as the time series guideline, the feature parameters of the four time series sets within the same period are aligned according to time series to ensure a one-to-one correspondence between parameters and control step size. The core control features, microbial activity features, influent operating condition features, and historical aeration features are classified and assigned to the corresponding column dimensions of the matrix in sequence, thereby achieving parameter classification and dimensional matching. The principle of uniqueness and completeness is followed to ensure that each feature parameter uniquely corresponds to the row and column position of the matrix, with no duplication, omission, or misalignment. At the same time, the parameters are linked and verified in conjunction with the relationships to ensure that the coupling relationship between the integrated parameters is consistent with the rules.

[0083] S522. Extract all feature parameters of the four types of time series sets and classify and organize them according to the integration rules; Specifically, the four types of time series sets are traversed along the time axis in a controlled cycle to fully extract all effective feature parameters after coupling analysis and feature screening, covering key information such as time series changes, correlation strength, and dependency characteristics, ensuring that the extraction process is complete without omissions, repetitions, or distortions, and fully preserving the core and auxiliary features related to dissolved oxygen regulation.

[0084] After extraction, based on the multivariate temporal coupling association rules and matrix integration requirements, standardized classification was carried out with the control step size as the main line and the time series set type as the classification basis: parameters directly related to dissolved oxygen regulation were classified into the core control feature category, parameters reflecting microbial metabolic needs were classified into the microbial activity feature category, parameters characterizing influent load fluctuations were classified into the influent operating condition feature category, and parameters reflecting historical operating patterns were classified into the historical aeration feature category. This ensured that each parameter uniquely corresponded to its category and matched one-to-one with the matrix column dimensions. During the classification process, temporal alignment verification and integrity checks were carried out simultaneously to ensure that the time axis of various parameters under the same control step size was unified, the data was continuous and effective, and the coupling relationship met the rule requirements. After extraction, classification, and verification, a time series set feature parameter with a clear structure, accurate classification, and regular temporal sequence was formed.

[0085] S523. Set a coupling empty matrix based on the row and column dimensions of the prediction matrix. The row dimension corresponds to the control step size, and the column dimension corresponds to the core feature parameters of the four types of time series sets. Specifically, the control step size is used as the sole basis for setting the row dimension. Based on the prediction duration, control cycle parameters and process control requirements, the total number of rows in the matrix is ​​determined. Each row strictly corresponds to an independent control step size and is numbered sequentially according to time sequence to ensure that the row dimension is continuous in time, uniform in interval, without overlap or missing, and highly matched with the control rhythm.

[0086] The column dimensions are set based on the core feature parameters of the core control feature time series set, microbial activity time series set, influent operating condition time series set, and historical aeration time series set. They are arranged sequentially according to the preset classification order, with each column independently corresponding to one core feature parameter. The total number of columns is completely consistent with the number of core feature parameters after screening, ensuring that the column dimensions fully cover the key parameters of dissolved oxygen regulation. After the row and column dimensions are accurately determined, an initial coupled empty matrix is ​​created according to the standard matrix data structure. The dimension definition, index number, data format, and storage type are uniformly initialized to make the matrix format adaptable to subsequent feature parameter filling and time series dynamic inference calculation. Then, the dimension consistency of the coupled empty matrix is ​​checked to verify the matching of the number of rows with the control step size and the correspondence of the number of columns with the core feature parameters. It is confirmed that the row and column division is clear, the dimension is standardized and reasonable, and it is highly consistent with the multivariate time series coupling association rules and parameter integration requirements, generating a coupled empty matrix with standard structure, accurate dimensions, and standardized format.

[0087] S524. Align and fill the rearranged feature parameters one by one into the coupling empty matrix, so that each element in the matrix corresponds to a unique feature parameter and time sequence node, and generate a multivariate coupling prediction matrix.

[0088] Specifically, based on the time-ordered feature parameters of the time series set and the precisely coupled empty matrix, the feature parameters are precisely filled into the corresponding positions of the matrix according to the principles of time alignment, classification mapping, and unique matching, thereby generating a standardized multivariate coupled prediction matrix.

[0089] The filling process uses the control step size as the time series main line, traversing row by row along the matrix dimension, with each row corresponding to an independent time series node. At the same time, it uses the core feature parameters of four types of time series sets as the classification basis, and positions them in an orderly manner along the column dimension, with each column corresponding to a unique feature parameter. This ensures that the data filling strictly follows the row and column dimension constraints. Then, the time series set feature parameters are aligned and mapped to the corresponding cells of the coupled empty matrix according to the control period number and feature type, so that each matrix element is uniquely bound to a time series node and a feature parameter, with no repetition, no omission, and no misalignment.

[0090] After the data is filled, a full matrix verification is performed to check the continuity of time series, the completeness of parameters, the uniqueness of positions, and the consistency of coupling relationships. It is confirmed that all feature parameters accurately fall into the corresponding spatiotemporal coordinates and are highly matched with the multivariate temporal coupling association rules. After verification, the data filling and matrix formatting are completed to generate a multivariate coupling prediction matrix with standardized dimensions, complete data, time series alignment, and clear coupling relationships.

[0091] S53. Combining the control cycle parameters and the coupling analysis results, determine the future control step size and time-series dependency, perform time-series dynamic deduction on the prediction matrix, and predict the changing trends of various characteristic parameters within the future control step size; Specifically, based on the system control cycle parameters, and combined with the response speed of dissolved oxygen regulation and the action lag characteristics of aeration equipment, the number of control steps for future predictions is reasonably determined. The single step length is consistent with the system control cycle to ensure that the prediction rhythm and the on-site process control rhythm are fully matched.

[0092] Simultaneously, based on the time lag period, cross-cycle transmission law, and variable linkage response characteristics extracted from the coupling analysis results, the mutual influence relationship and dynamic evolution logic among the four types of time series sets are clarified. On this basis, the mature LSTM rolling time-domain recursive method in the field of intelligent wastewater treatment control is adopted. Using a structured multivariate coupling prediction matrix as the data carrier, dynamic time-series extrapolation is carried out: starting with the field-measured data of the current control cycle as the prediction starting point, the initial position of the multivariate coupling prediction matrix is ​​filled in, and recursive calculations are performed cycle by cycle along the time sequence of future control steps. At each step, the prediction results of the previous cycle, the coupling rules between parameters, and the time lag characteristics are simultaneously substituted to obtain the predicted values ​​of various characteristic parameters for the current step, which are then filled into the corresponding positions of the prediction matrix, until the extrapolation of all future steps is completed, filling the entire multivariate coupling prediction matrix.

[0093] During the simulation, the rising and falling trends, abrupt change nodes, stable intervals, and periodic fluctuation patterns of various parameters are captured simultaneously. After the simulation is completed, the rationality of the prediction results is verified by combining the biochemical reaction mechanism of aerobic wastewater treatment and process constraints such as equipment operating boundaries. Abnormal results that do not conform to the actual operating logic are eliminated, and finally the changing trends of various characteristic parameters within each control step are obtained.

[0094] S54. Based on the simulation results and the process's required dissolved oxygen threshold, the optimal dissolved oxygen setpoint within the future control step is determined.

[0095] Specifically, the core basis for this approach is the future parameter change trend obtained through time-series dynamic extrapolation. A comprehensive analysis is conducted on influent load fluctuations, changes in microbial metabolic activity, and aeration system operating characteristics within each control step. This allows for accurate prediction of dissolved oxygen supply and demand trends, fluctuation ranges, and potential disturbance risks. Simultaneously, the process dissolved oxygen demand threshold is clearly defined, and hard constraints are established for the dissolved oxygen setpoint. These constraints include the minimum dissolved oxygen limit to ensure microbial activity, the maximum dissolved oxygen limit to avoid energy waste and sludge bulking, the mandatory requirements for effluent quality compliance, and the adjustment capacity boundaries of the aeration equipment. This ensures that the final setpoint fully aligns with the on-site operating conditions and process control objectives.

[0096] Based on this, a mature multi-constraint priority optimization method in the field of industrial process control is adopted. Prioritizing achieving effluent quality standards as the first core objective, optimal aeration energy consumption as the second objective, precise matching of dissolved oxygen supply to the actual oxygen demand of microorganisms as the third objective, and stable system operation as the fallback objective, the optimal dissolved oxygen setpoint is determined through cycle-by-cycle screening within each future control step. The first step is to use the hard constraints of the process as the screening threshold, and generate candidate dissolved oxygen setpoints of different gradients within the defined upper and lower limits of dissolved oxygen. Invalid candidate values ​​that cannot guarantee the effluent water quality to meet the standards or exceed the equipment's adjustment capacity are eliminated. The second step involves sorting the remaining candidate values ​​according to a preset priority order and considering the projected trends of future operating conditions. This process involves: prioritizing the matching of future influent load fluctuations to ensure stable effluent quality; matching changes in microbial metabolic activity to ensure dissolved oxygen supply matches the actual oxygen demand of microorganisms; simultaneously considering aeration energy consumption to avoid energy waste caused by excessive aeration; and avoiding large and frequent fluctuations in the set values ​​to ensure stable system operation. The third step is to perform a final process adaptability check on the selected candidate values, verify the setpoint's ability to resist disturbances and its control response speed under future changes in operating conditions, confirm that it fully meets the process operation requirements, and finally determine the optimal dissolved oxygen setpoint that accurately adapts to changes in operating conditions within each control step in the future.

[0097] S6. A closed-loop control model for an aeration system with feedforward compensation, feedback correction and disturbance suppression capabilities is constructed using a model predictive control algorithm. The optimal dissolved oxygen setpoint is converted into a real-time control target parameter input model. The optimal operating control parameters for the variable frequency fan, air diffusion device and liquid level regulating valve are calculated to generate the optimal aeration control sequence. In this embodiment of the invention, the step of constructing a closed-loop control model for the aeration system with feedforward compensation, feedback correction, and disturbance suppression capabilities using a model predictive control algorithm, converting the optimal dissolved oxygen setpoint into a real-time control target parameter input model, calculating the optimal operating control parameters for the variable frequency fan, air diffuser, and liquid level regulating valve, and generating the optimal aeration control sequence includes the following steps: S61. Based on the operating rules of the aeration system reflected by the control cycle parameters and historical aeration operation data, a closed-loop control model of the aeration system is constructed by combining the model predictive control algorithm. Specifically, firstly, using preset control cycle parameters as a unified time series benchmark, and combining historical aeration operation data, we deeply explore the dynamic response characteristics, action delay characteristics, energy consumption characteristics, effective adjustment range, and stable operation constraints of the aeration equipment, and establish the a priori laws of the input, output, and state evolution of the aeration system, providing a real field operation basis for model construction.

[0098] Subsequently, a closed-loop control framework was built around the model predictive control algorithm, comprising three main modules: predictive model, rolling optimization, and feedback correction. The core of the predictive model is an aerobic tank dissolved oxygen kinetic mechanism model simplified from the widely used and mature activated sludge mechanism model in the wastewater treatment industry. This model clearly defines the dynamic quantitative mapping relationship between dissolved oxygen concentration in the aerobic tank and aeration oxygen supply, influent load, and microbial oxygen consumption. Those skilled in the art can directly construct operable control strategies based on this model. The core logic and structure of the model are as follows: This model uses the measured dissolved oxygen concentration in the aerobic tank during the current control cycle as the core controlled state, the actual oxygen supply of the aeration system as the controllable input (the oxygen supply is linearly positively correlated with the operating frequency of the variable frequency fan and the opening degree of the air diffusion device), and the total oxygen consumption of the system due to the influent as the measurable disturbance (total oxygen consumption consists of three parts: oxygen consumption from the degradation of organic matter in the influent, oxygen consumption from ammonia nitrification, and oxygen consumption from endogenous respiration of microorganisms, which can be directly calculated from real-time influent operating data and microbial activity data). The final output is a predicted value of the dissolved oxygen concentration for the future control cycle. The core operating logic of the model is that the dissolved oxygen concentration in the future control cycle is determined by three core factors: first, the natural maintenance characteristics of the dissolved oxygen concentration in the current cycle; second, the effect of aeration oxygen supply on increasing the dissolved oxygen concentration; and third, the effect of influent load fluctuations and oxygen consumption from microbial metabolism on decreasing the dissolved oxygen concentration.

[0099] The core characteristic coefficients of the model, including the dissolved oxygen concentration self-maintenance coefficient, the efficiency coefficient of dissolved oxygen improvement per unit aeration rate, and the influence coefficient of dissolved oxygen decay per unit oxygen consumption, were all identified by fitting historical operating data from the field using the mature least squares method in this field. No additional complex experiments are required, and those skilled in the art can directly complete the parameter calibration. After multi-condition simulation and historical data verification, and the final parameter calibration, a dissolved oxygen kinetic mechanism model adapted to the field conditions can be obtained.

[0100] Based on the dissolved oxygen kinetic mechanism model, and combined with the previously generated optimal dissolved oxygen setpoint for the future control step, the predicted results of time-series characteristic parameters, and process constraint thresholds, a dynamic prediction model for model predictive control is constructed. This model can accurately predict the dissolved oxygen change trend corresponding to aeration adjustment within the future control step. Through a rolling optimization process, the aeration control quantity that satisfies optimal energy consumption, rapid response, and minimal fluctuation is solved, outputting control commands such as fan frequency and aeration volume. Simultaneously, a real-time operation feedback loop is introduced, feeding back the actual dissolved oxygen monitoring values ​​from the field with the model prediction values ​​and the output control quantities. After comparing the deviations, the model parameters are dynamically corrected, forming a complete closed loop of prediction-optimization-execution-feedback-correction. During model construction, the control cycle timing is strictly matched to ensure that the model calculation cycle is completely synchronized with the field control rhythm. Through multi-condition simulation and historical data verification, the robustness and control accuracy of the model under disturbances such as influent fluctuations and load changes are verified. Finally, a closed-loop control model for the aeration system that adapts to the aerobic wastewater treatment process and integrates field data patterns and model predictions is formed.

[0101] S62. The optimal dissolved oxygen setpoint is taken as the core controlled target. The feedforward compensation term and feedback correction term are constructed by combining the influent operating data and historical aeration operation data, and the setpoint is converted into real-time control target parameters. Specifically, the optimal dissolved oxygen setpoint is first used as the benchmark control target. Real-time influent flow rate, pollutant load, water quality fluctuations and other influent operating data are simultaneously input. Combined with the influent disturbance response law, aeration delay characteristics and regulation efficiency extracted from historical aeration operation data, and based on the total system oxygen consumption in the aforementioned dissolved oxygen kinetic mechanism model, a feedforward compensation term is constructed: based on real-time influent operating data, the system oxygen consumption change of future control steps is predicted, and the corresponding aeration compensation amount is calculated in advance to pre-compensate for the dissolved oxygen supply and demand deviation caused by influent fluctuations, thereby weakening the control deviation caused by lag disturbances from the source.

[0102] Simultaneously, real-time dissolved oxygen monitoring values ​​are collected on-site, and the real-time deviation between these values ​​and the baseline setpoint is used as a feedback correction term. Based on a closed-loop correction mechanism, a dynamic correction quantity is formed to offset control deviations caused by model fitting errors, random on-site disturbances, and drift of aeration equipment characteristics. Subsequently, the baseline optimal setpoint, feedforward compensation, and feedback correction quantity are weighted and fused according to the control cycle sequence. Smoothing is performed in conjunction with the aeration system adjustment constraints and process safety range, and abnormal jump values ​​are eliminated to ensure the continuous and stable target curve. The static optimal dissolved oxygen setpoint is transformed into a real-time control target parameter that conforms to the dynamic changes on-site.

[0103] S63. Substitute the feedforward compensation term, feedback correction term, and real-time control target parameters into the model, calculate the optimal operating control parameters for the three types of actuators, and integrate them to generate the optimal aeration control sequence.

[0104] In this embodiment of the invention, the step of substituting the feedforward compensation term, the feedback correction term, and the real-time control target parameters into the model, calculating the optimal operating control parameters for the three types of actuators, and integrating them to generate the optimal aeration control sequence includes the following steps: S631. Take the real-time control target parameters as the core controlled reference of the model, substitute them with feedforward compensation terms and feedback correction terms, and clarify the priority and correlation of the coupling operation of the three types of parameters in the model. Specifically, the real-time control target parameter serves as the primary core benchmark, carrying the optimal dissolved oxygen setpoint after dynamic smoothing and adaptation to process constraints. This benchmark adjustment target acts as the top-level guide for the entire closed-loop model, determining the overall control direction and target range of the aeration system. The feedforward compensation term, generated based on real-time influent operating condition fluctuations and historical aeration operation patterns, serves as the secondary advanced compensation parameter, second only to the core benchmark in priority. It is used to predict and offset predictable disturbances such as influent load and water quality changes, achieving source control. This corresponds to the compensation amount for the total oxygen consumption of the system in the aforementioned model.

[0105] The feedback correction term is generated based on the deviation between the on-site measured dissolved oxygen and the baseline value. As a third-level real-time correction parameter, it has final correction authority and is used to correct dynamic deviations caused by model errors, random disturbances, and equipment drift. It corresponds to the correction amount for the deviation between the model prediction value and the on-site measured value. In the coupled calculation, a priority order is followed: core baseline priority, feedforward compensation follow-up, and feedback correction as a fallback. A progressive association is established: baseline sets the target, feedforward eliminates disturbances, and feedback improves accuracy. The three types of parameters are synchronously and weighted according to the control cycle time sequence, while simultaneously binding constraints such as the aeration system adjustment rate, safety threshold, and response characteristics. The corresponding mapping relationship between parameters and control outputs such as aeration volume and fan frequency is clearly defined.

[0106] S632. Based on the control cycle parameters and the operating law of the aeration system, the three types of coupled parameters are calculated step by step through the model to determine the optimal operating frequency of the variable frequency fan, the optimal opening degree of the air diffusion device and the optimal adjustment amount of the liquid level regulating valve. Specifically, the three types of coupling parameters are first input into the closed-loop control model of the aeration system according to a preset calculation priority. The optimization objectives are real-time tracking of the target dissolved oxygen concentration, optimal aeration energy consumption, and stable adjustment response. Process constraints such as the aeration system's adjustment rate, safety threshold, and load adaptation range are bound to these parameters. The solution is iteratively solved cycle by cycle along the time-series axis of the future control step size. Historical operating patterns are fully integrated into the calculation process. Based on the aforementioned dissolved oxygen kinetic mechanism model, precise mapping relationships are established between the frequency of the variable frequency fan and the oxygen supply, the opening degree of the air diffuser and the uniformity of air distribution, and the adjustment amount of the liquid level regulating valve and the hydraulic residence time of the aerobic tank. The linkage logic of these three factors is coordinated synchronously to avoid adjustment conflicts and overshoot oscillations.

[0107] For each control step, the model outputs candidate parameter combinations that satisfy all constraints through rolling optimization. After comprehensive verification in four dimensions—steady-state control accuracy, response speed, energy consumption level, and operational stability—the global optimal solution is selected. The optimal operating frequency of the variable frequency fan, the optimal opening degree of the air diffusion device, and the optimal adjustment amount of the liquid level regulating valve are determined for that control step. All optimal parameters are strictly synchronized with the control cycle and can be directly sent to the field actuators, providing reliable execution instructions for the aeration system to achieve closed-loop control of dissolved oxygen.

[0108] S633. Integrate the three types of optimal operating control parameters to generate the optimal aeration control sequence.

[0109] Specifically, the three types of optimal control parameters are first checked for timing synchronization to ensure that the fan frequency, diffuser opening, and regulating valve adjustment amount correspond one-to-one and are time-aligned under the same control step size, without misalignment, missing, or adjustment conflict. Combined with the linkage characteristics of the aeration system equipment, regulation rate constraints, safe operation thresholds, and process response delay characteristics, the linkage rationality of the parameters is checked, and parameter combinations that may cause overshoot, oscillation, or uncoordinated operation of the equipment are eliminated to ensure smooth coordination between parameters and stable control process.

[0110] Subsequently, using the control step size as the sequential index, the optimal operating control parameters that have passed the verification in each cycle are encapsulated and arranged chronologically to form a control sequence skeleton that covers the entire prediction duration, has complete parameters, and is time-sequential. The sequence is then smoothed by interpolation and boundary constraints to prevent parameter abrupt changes from impacting equipment and process operation, thereby improving the stability and reliability of sequence execution. After integration, the optimal aeration control sequence is verified through dual validation using historical operating data and simulated operating conditions. This verifies the adaptability, control accuracy, and energy-saving effect of the optimal aeration control sequence under fluctuating influent load and changing operating conditions, confirming that the sequence meets the dissolved oxygen closed-loop control target and equipment safety operation requirements. Finally, a standardized, time-sequential, and parameter-accurate optimal aeration control sequence is generated that can be directly distributed to field actuators.

[0111] S7. Perform real-time verification and dynamic optimization of the execution effect and dissolved oxygen control accuracy of the optimal aeration control sequence, verify the robustness of the system under complex disturbance conditions, and generate a dissolved oxygen control operation evaluation report based on the verified and optimized operating control parameters.

[0112] Specifically, firstly, real-time monitoring instruments are used to collect comprehensive operational data, including dissolved oxygen concentration, variable frequency fan operating frequency, air diffusion device opening, liquid level regulating valve adjustment amount, influent flow rate, load fluctuation, and water quality indicators. The actual output is then compared with the target value of the optimal aeration control sequence step by step to quantitatively verify the sequence execution rate, dissolved oxygen steady-state accuracy, dynamic response speed, overshoot, and regulation stability, and to assess whether the control effect meets the standards.

[0113] To address deviations identified during verification, a rolling optimization module was activated to dynamically correct control sequence parameters, combining feedforward compensation and feedback correction mechanisms. This eliminated the impact of model errors, random disturbances, and equipment lag, continuously improving control accuracy and execution performance. Simultaneously, typical complex disturbances such as sudden changes in influent load, water quality fluctuations, and equipment drift were proactively introduced to conduct system robustness tests. These tests verified the stability, disturbance rejection capability, and rapid recovery performance of the control strategy under extreme conditions. Real-time verification data, dynamic optimization results, robustness test indicators, and energy consumption operation data were systematically collected and statistically analyzed to objectively evaluate dissolved oxygen control levels, equipment operating efficiency, energy-saving effects, and system reliability. The report also identified strengths, problems, and improvement directions, ultimately forming a standardized dissolved oxygen control operation evaluation report for the wastewater aerobic treatment system, including operational data, accuracy indicators, robustness evaluation, and optimization suggestions.

[0114] According to another aspect of the invention, such as Figure 2 As shown, a precise dissolved oxygen control system for aerobic wastewater treatment is provided. The system includes: a data acquisition module 1, a characteristic parameter module 2, a time sequence parameter module 3, an analysis and correlation module 4, a matrix setting module 5, a control sequence module 6, and an evaluation report module 7. Data acquisition module 1 is used to acquire multi-source sensing data, influent operating condition data, control cycle parameters and historical aeration operation data of the aerobic wastewater treatment system. The multi-source sensing data includes core data of the aerobic tank water environment and microbial metabolic state data, and preprocesses the core data of the water environment and microbial metabolic state data. Feature parameter module 2 is used to preset and extract core feature parameters for dissolved oxygen control and supplementary feature parameters for microbial activity from the preprocessed data based on a multi-source data fusion feature extraction strategy. The timing parameter module 3 is used to perform time-series division of core feature parameters, supplementary feature parameters, influent operating condition data and historical aeration operation data based on control cycle parameters, to obtain four types of time-series sets: core control features, microbial activity, influent operating conditions and historical aeration. The analysis module 4 is used to perform in-depth analysis of multivariate coupling relationships on four types of time series based on a long short-term memory network with attention mechanism, and to set multivariate temporal coupling association rules among the four types of time series. The matrix setting module 5 is used to generate a multivariate coupled prediction matrix from four types of time series sets according to the coupling association rules, perform time series dynamic reasoning on the matrix, and generate the optimal dissolved oxygen setpoint within the future control step. The control sequence module 6 is used to construct a closed-loop control model of the aeration system with feedforward compensation, feedback correction and disturbance suppression capabilities using model predictive control algorithm. It converts the optimal dissolved oxygen setpoint into the real-time control target parameter input model, calculates the optimal operating control parameters of the variable frequency fan, air diffusion device and liquid level regulating valve, and generates the optimal aeration control sequence. The evaluation report module 7 is used to verify and dynamically optimize the execution effect of the optimal aeration control sequence and the accuracy of dissolved oxygen control in real time, verify the robustness of the system under complex disturbance conditions, and generate a dissolved oxygen control operation evaluation report based on the verified and optimized operating control parameters.

[0115] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for precise control of dissolved oxygen in aerobic wastewater treatment, characterized in that, Includes the following steps: S1. Acquire multi-source sensing data, influent operating condition data, control cycle parameters and historical aeration operation data of the aerobic wastewater treatment system. The multi-source sensing data includes core data of the aerobic tank water environment and microbial metabolic state data, and preprocess the core data of the water environment and microbial metabolic state data. S2. Based on a multi-source data fusion feature extraction strategy, extract the core feature parameters for dissolved oxygen control and supplementary feature parameters for microbial activity from the preprocessed data. S3. Based on the control cycle parameters, the core characteristic parameters, supplementary characteristic parameters, influent operating condition data and historical aeration operation data are divided into time series to obtain four types of time series sets: core control characteristics, microbial activity, influent operating conditions and historical aeration. S4. Based on the Long Short-Term Memory network with attention mechanism, a deep analysis of the multivariate coupling relationship of four types of time series sets is carried out, and multivariate temporal coupling association rules are set among the four types of time series sets. S5. Based on the coupling association rule, generate a multivariate coupling prediction matrix from the four types of time series sets, perform time series dynamic reasoning on the matrix, and generate the optimal dissolved oxygen setpoint within the future control step. S6. A closed-loop control model for an aeration system with feedforward compensation, feedback correction and disturbance suppression capabilities is constructed using a model predictive control algorithm. The optimal dissolved oxygen setpoint is converted into a real-time control target parameter input model. The optimal operating control parameters for the variable frequency fan, air diffusion device and liquid level regulating valve are calculated to generate the optimal aeration control sequence. S7. Perform real-time verification and dynamic optimization of the execution effect and dissolved oxygen control accuracy of the optimal aeration control sequence, verify the robustness of the system under complex disturbance conditions, and generate a dissolved oxygen control operation evaluation report based on the verified and optimized operating control parameters.

2. The method for precise control of dissolved oxygen in aerobic wastewater treatment according to claim 1, characterized in that, The step of dividing the core characteristic parameters, supplementary characteristic parameters, influent operating condition data, and historical aeration operation data into four time series sets based on the control cycle parameters to obtain four types of time series sets: core control characteristics, microbial activity, influent operating conditions, and historical aeration, includes the following steps: S31. Using the control cycle parameter as a unified time reference, perform time synchronization and alignment on various parameters, match the timing sampling frequency with the control cycle, and determine the time window for dividing the control cycle. S32. After synchronization and alignment, various parameters are segmented periodically according to the time window, and four types of time series sets are generated according to the data type, and the time dimension of each type of time series set remains consistent. S33. Verify and adjust the timing length and data integrity of the four types of timing sets to complete the timing partitioning.

3. The method for precise control of dissolved oxygen in aerobic wastewater treatment according to claim 1, characterized in that, The long short-term memory network with attention mechanism is used to perform deep analysis of multivariate coupling relationships on four types of time series sets, and the multivariate temporal coupling association rules between the four types of time series sets are set, including the following steps: S41. Combining the characteristics of multivariate time series data of four types of time series sets with the requirements of aerobic wastewater treatment process, configure the model input layer, attention weight allocation layer and long short-term memory network hidden layer to construct a long short-term memory network model with attention mechanism. S42. After normalizing and preprocessing the four types of time series sets, they are used as multi-channel input data and fed into the long short-term memory network model. S43. The attention weight allocation layer is used to assign feature importance weights to the four types of time series sets. The hidden layer is used to capture the temporal dependencies, and the deep analysis of multivariate coupling relationships is carried out to output the coupling analysis results. S44. Based on the coupling analysis results and process requirements, clarify the correlation threshold and timing response relationship of the characteristic parameters of each time series set, and set multivariate timing coupling correlation rules among the four types of time series sets.

4. The method for precise control of dissolved oxygen in aerobic wastewater treatment according to claim 1, characterized in that, The process of generating a multivariate coupled prediction matrix from four types of time series based on coupling association rules, and then performing time-series dynamic inference on this matrix to generate the optimal dissolved oxygen setpoint within the future control step includes the following steps: S51. Based on the coupling association rule, clarify the association dimension of the feature parameters of the four types of time series sets, determine the row and column dimensions of the multivariate coupling prediction matrix, the row dimension corresponds to the control step size, and the column dimension corresponds to the core feature parameters of the four types of time series sets. S52. Integrate the feature parameters of the four types of time series sets according to the coupling association rules, preset the corresponding dimension empty matrix and fill the feature parameters to generate a multivariate coupled prediction matrix. S53. Combining the control cycle parameters and the coupling analysis results, determine the future control step size and time-series dependency, perform time-series dynamic deduction on the prediction matrix, and predict the changing trends of various characteristic parameters within the future control step size; S54. Based on the simulation results and the process's required dissolved oxygen threshold, the optimal dissolved oxygen setpoint within the future control step is determined.

5. The method for precise control of dissolved oxygen in aerobic wastewater treatment according to claim 1, characterized in that, The process of constructing a closed-loop control model for the aeration system using a model predictive control algorithm, incorporating feedforward compensation, feedback correction, and disturbance suppression capabilities, converts the optimal dissolved oxygen setpoint into a real-time control target parameter input model, calculates the optimal operating control parameters for the variable frequency fan, air diffuser, and liquid level regulating valve, and generates the optimal aeration control sequence, includes the following steps: S61. Based on the operating rules of the aeration system reflected by the control cycle parameters and historical aeration operation data, a closed-loop control model of the aeration system is constructed by combining the model predictive control algorithm. S62. The optimal dissolved oxygen setpoint is taken as the core controlled target. The feedforward compensation term and feedback correction term are constructed by combining the influent operating data and historical aeration operation data, and the setpoint is converted into real-time control target parameters. S63. Substitute the feedforward compensation term, feedback correction term, and real-time control target parameters into the model, calculate the optimal operating control parameters for the three types of actuators, and integrate them to generate the optimal aeration control sequence.

6. The method for precise control of dissolved oxygen in aerobic wastewater treatment according to claim 2, characterized in that, The process of segmenting various parameters after synchronization and alignment according to time windows, generating four types of time series sets according to data type, and ensuring that the time dimension of each type of time series set remains consistent includes the following steps: S321. Clearly define the start and end timestamps of the control cycle time window and match them with the sampling frequency of various parameters after synchronization and alignment; S322. Based on the matching results, segment and extract various parameters periodically to ensure that each segment of data corresponds to a unique control period. S323. The extracted parameters are categorized into four time series sets, and the output is generated after verifying that the time dimension is consistent.

7. The method for precise control of dissolved oxygen in aerobic wastewater treatment according to claim 3, characterized in that, The process of assigning feature importance weights to four types of time series sets through an attention weight allocation layer, capturing temporal dependencies using hidden layers, conducting deep analysis of multivariate coupling relationships, and outputting the coupling analysis results includes the following steps: S431. Initialize the attention weight allocation layer, using the influence of the four types of time series on dissolved oxygen control as the core basis for weight allocation, so that the weight allocation fits the process control requirements. S432. Input the four types of time series sets into the attention weight allocation layer, perform feature differentiation weight allocation and complete the assignment; S433. Input the four types of time series sets after weight assignment into the hidden layer, capture the changing patterns of feature parameters periodically, explore the correlation and temporal dependency relationships among the four types of time series sets, and obtain preliminary analysis results of coupling relationships. S434. Integrate and organize the preliminary analysis results to form coupled analysis results.

8. The method for precise control of dissolved oxygen in aerobic wastewater treatment according to claim 4, characterized in that, The process of integrating the feature parameters of the four types of time series sets according to coupling association rules, pre-setting an empty matrix of the corresponding dimension and filling it with feature parameters to generate a multivariate coupled prediction matrix includes the following steps: S521. Clarify the correlation relationship of the four types of time series feature parameters in the coupling association rule, and determine the feature parameter integration rule by combining the row and column dimensions of the prediction matrix; S522. Extract all feature parameters of the four types of time series sets and classify and organize them according to the integration rules; S523. Set a coupling empty matrix based on the row and column dimensions of the prediction matrix. The row dimension corresponds to the control step size, and the column dimension corresponds to the core feature parameters of the four types of time series sets. S524. Align and fill the rearranged feature parameters one by one into the coupling empty matrix, so that each element in the matrix corresponds to a unique feature parameter and time sequence node, and generate a multivariate coupling prediction matrix.

9. A method for precise control of dissolved oxygen in aerobic wastewater treatment according to claim 5, characterized in that, The process of substituting the feedforward compensation term, feedback correction term, and real-time control target parameters into the model to calculate the optimal operating control parameters for the three types of actuators and integrate them to generate the optimal aeration control sequence includes the following steps: S631. Take the real-time control target parameters as the core controlled reference of the model, substitute them with feedforward compensation terms and feedback correction terms, and clarify the priority and correlation of the coupling operation of the three types of parameters in the model. S632. Based on the control cycle parameters and the operating law of the aeration system, the three types of coupled parameters are calculated step by step through the model to determine the optimal operating frequency of the variable frequency fan, the optimal opening degree of the air diffusion device and the optimal adjustment amount of the liquid level regulating valve. S633. Integrate the three types of optimal operating control parameters to generate the optimal aeration control sequence.

10. A precise dissolved oxygen control system for aerobic wastewater treatment, used to implement the precise dissolved oxygen control method for aerobic wastewater treatment as described in any one of claims 1-9, characterized in that, The system includes: a data acquisition module, a feature parameter module, a time series parameter module, an analysis and correlation module, a matrix setting module, a control sequence module, and an evaluation report module; The data acquisition module is used to acquire multi-source sensing data, influent operating condition data, control cycle parameters and historical aeration operation data of the aerobic wastewater treatment system. The multi-source sensing data includes core data of the aerobic tank water environment and microbial metabolic state data, and preprocesses the core data of the water environment and microbial metabolic state data. The feature parameter module is used to preset and extract core feature parameters for dissolved oxygen control and supplementary feature parameters for microbial activity from the preprocessed data based on a multi-source data fusion feature extraction strategy. The timing parameter module is used to perform time-series partitioning of core feature parameters, supplementary feature parameters, influent operating condition data and historical aeration operation data based on control cycle parameters, resulting in four types of time-series sets: core control features, microbial activity, influent operating conditions and historical aeration. The analysis and association module is used to perform in-depth analysis of multivariate coupling relationships on four types of time series based on a long short-term memory network with attention mechanism, and to set multivariate temporal coupling association rules among the four types of time series; The matrix setting module is used to generate a multivariate coupled prediction matrix from four types of time series sets according to the coupling association rules, perform time series dynamic inference on the matrix, and generate the optimal dissolved oxygen setpoint within the future control step. The control sequence module is used to construct a closed-loop control model of the aeration system with feedforward compensation, feedback correction and disturbance suppression capabilities using model predictive control algorithms. It converts the optimal dissolved oxygen setpoint into the real-time control target parameter input model, calculates the optimal operating control parameters of the variable frequency fan, air diffusion device and liquid level regulating valve, and generates the optimal aeration control sequence. The evaluation report module is used to verify and dynamically optimize the execution effect of the optimal aeration control sequence and the accuracy of dissolved oxygen control in real time, verify the robustness of the system under complex disturbance conditions, and generate a dissolved oxygen control operation evaluation report based on the verified and optimized operating control parameters.

Images