A method for evaluating flocculation effect for sewage treatment

Abstract

The application is a flocculation effect evaluation method for sewage treatment, and the steps are as follows: collecting turbidity, floc particle size and stirring torque in the flocculation process through a multi-modal sensor array and performing pretreatment; inputting the pretreated turbidity and floc particle size into a pre-trained time sequence feature capturing model to obtain preliminary time sequence fluctuation features, and inputting the pretreated stirring torque into a pre-trained correlation analysis model to obtain torque correlation features; coupling the preliminary time sequence fluctuation features and the torque correlation features to obtain complete fluctuation features, and calculating real-time fluctuation coefficients and fluctuation durations of the turbidity and the floc particle size; evaluating the flocculation effect based on the real-time fluctuation coefficients and the fluctuation durations of the turbidity and the floc particle size, and outputting a flocculation stirring equipment rotating speed adjustment instruction when the flocculation effect is substandard, so as to realize accurate, real-time and intelligent evaluation and dynamic adjustment of the flocculation effect, guarantee the stability of effluent water quality and improve the sewage treatment efficiency.

Description

Technical Field

[0001] This invention relates to the field of wastewater treatment technology, and in particular to a method for evaluating the flocculation effect in wastewater treatment. Background Technology

[0002] Wastewater treatment is a crucial aspect of environmental protection. Its core processes include pretreatment, biological treatment, advanced treatment, and sludge disposal. Among these, the flocculation process in the pretreatment stage is key to improving the efficiency of subsequent treatment. Flocculation involves adding flocculants to cause suspended fine particles and colloidal substances in wastewater to aggregate into larger flocs, which facilitate subsequent sedimentation or filtration separation. This removes pollutants such as turbidity and organic matter, reduces the wastewater treatment load, and ensures that the effluent meets the standards. Traditional flocculation effect assessments often employ static endpoint evaluation methods, primarily relying on manual detection or single sensor data collection of indicators such as turbidity and floc particle size at the flocculation endpoint. These indicators are then compared to preset standards to determine the effectiveness. However, this assessment method fails to consider the dynamic fluctuations of parameters during the flocculation process, cannot reflect the flocculation system's resilience, and its simplistic approach lacks consideration of key influencing factors such as stirring conditions, resulting in low accuracy. Furthermore, the absence of a real-time feedback and adjustment mechanism makes it difficult to respond promptly to abnormal flocculation effects, easily leading to fluctuations in effluent quality and failing to meet the demands of intelligent and precise wastewater treatment. Summary of the Invention

[0003] In view of this, the present invention proposes a method for evaluating the flocculation effect in wastewater treatment, which can improve the accuracy of flocculation evaluation by incorporating dynamic fluctuations during the flocculation process.

[0004] The technical solution of this invention is implemented as follows: A method for evaluating the flocculation effect in wastewater treatment includes the following steps: Step S1: Collect turbidity, floc particle size and stirring torque during the flocculation process using a multimodal sensor array, and perform preprocessing. Step S2: Input the pre-processed turbidity and floc particle size into the pre-trained time series feature capture model to obtain preliminary time series fluctuation features; Step S3: Input the pre-processed stirring torque into the pre-trained correlation analysis model to obtain torque correlation features; Step S4: The preliminary time-series fluctuation characteristics and torque correlation characteristics are fused and coupled to obtain complete fluctuation characteristics, and the real-time fluctuation coefficients and fluctuation durations of turbidity and floc particle size are calculated. Step S5: Evaluate the flocculation effect based on the real-time fluctuation coefficient and fluctuation duration of turbidity, floc particle size, and when the flocculation effect is not up to standard, output a command to adjust the speed of the flocculation stirring equipment.

[0005] Preferably, the specific steps of step S1 are as follows: Step S11: Arrange turbidity sensors and particle size sensors at different monitoring points in the flocculation reaction tank, and install torque sensors on the stirring shaft of the flocculation stirring equipment. Step S12: Collect turbidity, floc particle size, and stirring torque using a turbidity sensor, a particle size sensor, and a torque sensor, respectively. Step S13: Noise reduction and standardization processing are performed on turbidity, floc particle size and stirring torque.

[0006] Preferably, the step of locating the monitoring point in step S11 is as follows: The flow field distribution in the flocculation reaction tank was simulated by CFD, and the velocity gradient distribution cloud map of the flow field was obtained. Based on the flow field velocity gradient cloud map, the flocculation reaction tank is divided into a high gradient zone, a medium gradient zone, and a low gradient zone. The vertical planes of the central axis of the high gradient zone, medium gradient zone and low gradient zone are obtained respectively. A virtual line is constructed along the water flow direction, starting from the inlet of the flocculation reaction tank on the side of the high gradient zone. The intersection of the virtual line and the vertical plane of the central axis is used as the monitoring point.

[0007] Preferably, step S2 includes the following specific steps: Step S21: Input the pre-processed turbidity and floc particle size into the pre-trained BiLSTM-GRU hybrid model, respectively. Step S22: Extract bidirectional time-series features from turbidity and floc size time-series data using the BiLSTM module, and perform efficient feature compression and redundancy removal on the bidirectional time-series features using the GRU module. Step S23: Integrate the optimized bidirectional timing features output by the GRU module and output them as preliminary timing fluctuation features.

[0008] Preferably, the specific steps of step S22 are as follows: The forward LSTM unit and the reverse LSTM unit of the BiLSTM module capture historical fluctuation patterns and future fluctuation trends, respectively. The historical fluctuation patterns and future fluctuation trends of turbidity and floc size are spliced ​​together to obtain the bidirectional time series features corresponding to turbidity and floc size, and then integrated to obtain a preliminary set of bidirectional time series features. The initial set of bidirectional time series features is input into the reset gate of the GRU module to filter and retain the core fluctuation features; The core fluctuation features are weighted by the update gate of the GRU module, and the weighted core fluctuation features are nonlinearly transformed by the candidate activation unit. The features after nonlinear transformation are subjected to dimensional compression and redundant interference information is removed, and the optimized bidirectional time series features are output.

[0009] Preferably, step S3 includes the following specific steps: Step S31: Construct an association analysis model based on the CatBoost model, which is trained using historical association data; Step S32: Input the pre-processed stirring torque into the correlation analysis model, classify the stirring torque using the CatBoost model, and explore the nonlinear correlation between stirring torque and turbidity and floc particle size fluctuation. Step S33: Based on the correlation analysis results, extract the characteristic parameters that reflect the influence of torque on fluctuations, and output them as torque correlation features.

[0010] Preferably, the specific steps of step S32 are as follows: Based on the extreme and average values ​​of the pre-treated stirring torque, an adaptive threshold division method is used to divide the stirring torque into three torque ranges: low torque, medium torque, and high torque. By combining historical flocculation data, the stirring intensity level corresponding to each torque range is determined and correlated with the torque range to clarify the flocculation condition type corresponding to different torque ranges. The categorized stirring torque is labeled and matched with the preliminary time-series fluctuation features before being input into the feature association layer of the CatBoost model to initialize the association weights. Through iterative training of the CatBoost model, the correlation between different torque ranges and turbidity fluctuation amplitudes and floc particle size fluctuation amplitudes was explored. Calculate the correlation coefficients between stirring torque and turbidity fluctuation and floc particle size fluctuation, quantify the nonlinear correlation strength as the nonlinear correlation result and output it.

[0011] Preferably, step S4 includes the following specific steps: Step S41: Assign weights to the preliminary temporal fluctuation features and torque correlation features through the attention mechanism fusion model. The preliminary temporal fluctuation features and torque correlation features are then fused based on the weights to obtain the preliminary fused features. Step S42: Standardize and normalize the preliminary fusion features to obtain complete fluctuation features; Step S43: Based on the complete fluctuation characteristics, calculate the real-time fluctuation coefficients of turbidity and floc particle size using the variance calculation method; Step S44: Based on the complete fluctuation characteristics, the fluctuation duration of turbidity and floc particle size is calculated through continuous anomaly monitoring.

[0012] Preferably, in step S41, after obtaining the preliminary fusion features, a feature filter is introduced to remove redundancy and strengthen the core features of the preliminary fusion features.

[0013] Preferably, step S5 includes the following specific steps: Step S51: When the real-time fluctuation coefficient and fluctuation duration of turbidity, floc particle size are all less than the preset fluctuation coefficient threshold and fluctuation duration threshold, it is determined that the flocculation effect meets the standard. Step S52: When the flocculation effect is not up to standard, a speed adjustment command is triggered. Based on the degree of fluctuation exceeding the standard, an appropriate stirring speed adjustment value is calculated and output to the flocculation stirring equipment.

[0014] Compared with the prior art, the beneficial effects of the present invention are: This invention provides a method for evaluating the flocculation effect in wastewater treatment. It collects turbidity and floc size not only after flocculation is complete, but throughout the entire flocculation process. The method then captures the characteristics of turbidity and floc size to obtain preliminary temporal fluctuation characteristics, clarifying the changes in turbidity and floc size throughout the flocculation process. In addition to collecting turbidity and floc size, it also collects the stirring torque of the agitator and obtains torque correlation characteristics through a correlation analysis model. The preliminary temporal fluctuation characteristics are then coupled and fused with the torque correlation characteristics to obtain complete fluctuation characteristics, improving the accuracy and completeness of fluctuation feature extraction. Finally, the method calculates the real-time fluctuation coefficient and fluctuation duration of turbidity and floc size under the action of stirring torque using the complete fluctuation characteristics. Based on this, it determines whether the standard is met. If the standard is not met, the speed of the agitator is adjusted in a timely manner to achieve closed-loop regulation and improve the wastewater flocculation effect. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only preferred embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of a method for evaluating the flocculation effect in wastewater treatment according to the present invention. Detailed Implementation

[0017] To better understand the technical content of this invention, a specific embodiment is provided below, and the invention will be further described in conjunction with the accompanying drawings.

[0018] See Figure 1 The present invention provides a method for evaluating the flocculation effect in wastewater treatment, comprising the following steps: Step S1: Collect turbidity, floc particle size and stirring torque during the flocculation process using a multimodal sensor array, and perform preprocessing. Step S2: Input the pre-processed turbidity and floc particle size into the pre-trained time series feature capture model to obtain preliminary time series fluctuation features; Step S3: Input the pre-processed stirring torque into the pre-trained correlation analysis model to obtain torque correlation features; Step S4: The preliminary time-series fluctuation characteristics and torque correlation characteristics are fused and coupled to obtain complete fluctuation characteristics, and the real-time fluctuation coefficients and fluctuation durations of turbidity and floc particle size are calculated. Step S5: Evaluate the flocculation effect based on the real-time fluctuation coefficient and fluctuation duration of turbidity, floc particle size, and when the flocculation effect is not up to standard, output a command to adjust the speed of the flocculation stirring equipment.

[0019] During the flocculation treatment of wastewater, the turbidity of the wastewater continuously decreases, while the floc particle size continuously increases. Traditional flocculation effect assessments primarily focus on the turbidity and floc particle size after flocculation. However, this invention provides a method for evaluating the flocculation effect in wastewater treatment that monitors turbidity changes and floc particle size throughout the entire flocculation process. A multimodal sensor array is deployed in the flocculation reaction tank to collect time-series data on turbidity and floc particle size during flocculation. This data is then processed by a pre-trained time-series feature capture model to obtain preliminary time-series fluctuation features. These features reflect the bidirectional time-series fluctuation patterns and core fluctuation characteristics of turbidity and floc particle size during flocculation, providing fundamental time-series data support for subsequent feature fusion. In addition to collecting turbidity and floc particle size, other data are also collected... The stirring torque of the agitator during the flocculation process is collected, and the torque correlation characteristics are obtained through correlation analysis model. This is used to reflect the classification of the stirring torque during flocculation and its nonlinear correlation with turbidity and floc particle size fluctuations. Finally, after coupling and fusing the preliminary temporal fluctuation characteristics with the torque correlation characteristics, complete fluctuation characteristics can be obtained to reflect the correlation between the temporal fluctuations of turbidity and floc particle size and the stirring torque. From the complete fluctuation characteristics, the real-time fluctuation coefficients and fluctuation durations of turbidity and floc particle size can be calculated. Finally, based on the real-time fluctuation coefficients and fluctuation durations, the flocculation effect can be evaluated, which can accurately reflect the dynamic anti-disturbance capability of the flocculation system, avoid the bias of traditional static endpoint assessment, provide a scientific basis for judging the flocculation effect, and thus ensure the stability of effluent water quality and improve the reliability of the flocculation process.

[0020] Preferably, the specific steps of step S1 are as follows: Step S11: Arrange turbidity sensors and particle size sensors at different monitoring points in the flocculation reaction tank, and install torque sensors on the stirring shaft of the flocculation stirring equipment. Step S12: Collect turbidity, floc particle size, and stirring torque using a turbidity sensor, a particle size sensor, and a torque sensor, respectively. Step S13: Noise reduction and standardization processing are performed on turbidity, floc particle size and stirring torque.

[0021] Turbidity sensors and particle size sensors can collect turbidity data and floc particle size during the flocculation process, respectively, while torque sensors can collect the stirring torque of the stirring shaft of the stirring equipment. After collecting data in three dimensions, noise reduction processing is performed to remove interfering data, and after standardization processing, the data dimensions are unified to facilitate subsequent model input.

[0022] Preferably, the step of locating the monitoring point in step S11 is as follows: The flow field distribution in the flocculation reaction tank was simulated using CFD (Computational Fluid Dynamics), and the velocity gradient distribution cloud map of the flow field was obtained. Based on the flow field velocity gradient cloud map, the flocculation reaction tank is divided into a high gradient zone, a medium gradient zone, and a low gradient zone. The vertical planes of the central axis of the high gradient zone, medium gradient zone and low gradient zone are obtained respectively. A virtual line is constructed along the water flow direction, starting from the inlet of the flocculation reaction tank on the side of the high gradient zone. The intersection of the virtual line and the vertical plane of the central axis is used as the monitoring point.

[0023] To ensure the accuracy of turbidity and floc size data acquisition, turbidity and floc size sensors need to be placed at corresponding locations within the flocculation reactor to cover the entire flow field of the flocculation process. Therefore, the flow field distribution within the flocculation reactor is first simulated using CFD. Based on the flow field velocity gradient distribution cloud map, the reactor is divided into high-gradient, medium-gradient, and low-gradient zones, corresponding to the inlet, middle, and outlet sections of the flocculation reactor, respectively, and the initial, middle, and final stages of flocculation. The line connecting the three selected points needs to be parallel to the water flow direction to fully cover the flow field, avoid monitoring blind spots, and ensure that the collected turbidity and floc size data can completely reflect the dynamic changes of the flocculation process.

[0024] Preferably, step S2 includes the following specific steps: Step S21: Input the pre-processed turbidity and floc particle size into the pre-trained BiLSTM-GRU hybrid model, respectively. Step S22: Extract bidirectional time-series features from turbidity and floc size time-series data using the BiLSTM module, and perform efficient feature compression and redundancy removal on the bidirectional time-series features using the GRU module. Step S23: Integrate the optimized bidirectional timing features output by the GRU module and output them as preliminary timing fluctuation features.

[0025] The time series feature capture model adopts a BiLSTM-GRU hybrid model, which is trained on historical time series data. The BiLSTM module of the BiLSTM-GRU hybrid model is used to extract bidirectional time series features, while the GRU module is used to optimize the features, thus finally outputting preliminary time series fluctuation features.

[0026] Preferably, the specific steps of step S22 are as follows: The forward LSTM unit and the reverse LSTM unit of the BiLSTM module capture historical fluctuation patterns and future fluctuation trends, respectively. The historical fluctuation patterns and future fluctuation trends of turbidity and floc size are spliced ​​together to obtain the bidirectional time series features corresponding to turbidity and floc size, and then integrated to obtain a preliminary set of bidirectional time series features. The initial set of bidirectional time series features is input into the reset gate of the GRU module to filter and retain the core fluctuation features; The core fluctuation features are weighted by the update gate of the GRU module, and the weighted core fluctuation features are nonlinearly transformed by the candidate activation unit. The features after nonlinear transformation are subjected to dimensional compression and redundant interference information is removed, and the optimized bidirectional time series features are output.

[0027] In the initial extraction of temporal fluctuation features, the BiLSTM module and the GRU module process the data sequentially. The BiLSTM module includes a forward LSTM unit and a reverse LSTM unit. The forward LSTM unit is used to capture the historical fluctuation patterns of turbidity and floc size, while the reverse LSTM unit can capture the future fluctuation trends of turbidity and floc size respectively. After concatenating the historical fluctuation patterns and future fluctuation trends, bidirectional temporal features corresponding to turbidity and floc size can be obtained. Then, the GRU module's reset gate and update gate are used for filtering and weight allocation to select core fluctuation features and weaken the influence of redundant features. Finally, after nonlinear transformation through candidate activation units, dimensionality compression is performed to reduce the amount of feature data, improve the efficiency of subsequent processing, and remove redundant interference information after compression, strengthening the core features related to fluctuation. Finally, the compressed and optimized bidirectional temporal features are output for subsequent integration of the initial temporal fluctuation features.

[0028] Preferably, step S3 includes the following specific steps: Step S31: Construct an association analysis model based on the CatBoost model, which is trained using historical association data; Step S32: Input the pre-processed stirring torque into the correlation analysis model, classify the stirring torque using the CatBoost model, and explore the nonlinear correlation between stirring torque and turbidity and floc particle size fluctuation. Step S33: Based on the correlation analysis results, extract the characteristic parameters that reflect the influence of torque on fluctuations, and output them as torque correlation features.

[0029] The correlation analysis model is implemented using the CatBoost model. After training with historical correlation data to reach a certain accuracy, the stirring torque can be processed. The CatBoost model can classify the stirring torque, determine the flocculation condition type in different torque ranges, and then explore the nonlinear correlation between stirring torque and turbidity and floc particle size fluctuation. Finally, the characteristic parameters are extracted from the correlation analysis results and output as torque correlation features.

[0030] Preferably, the specific steps of step S32 are as follows: Based on the extreme and average values ​​of the pre-treated stirring torque, an adaptive threshold division method is used to divide the stirring torque into three torque ranges: low torque, medium torque, and high torque. By combining historical flocculation data, the stirring intensity level corresponding to each torque range is determined and correlated with the torque range to clarify the flocculation condition type corresponding to different torque ranges. The categorized stirring torque is labeled and matched with the preliminary time-series fluctuation features before being input into the feature association layer of the CatBoost model to initialize the association weights. Through iterative training of the CatBoost model, the correlation between different torque ranges and turbidity fluctuation amplitudes and floc particle size fluctuation amplitudes was explored. Calculate the correlation coefficients between stirring torque and turbidity fluctuation and floc particle size fluctuation, quantify the nonlinear correlation strength as the nonlinear correlation result and output it.

[0031] The extreme values, average values, and fluctuation ranges of the stirring torque can be extracted from the pre-processed stirring torque. Then, based on the extreme values ​​and average values, an adaptive threshold classification method is used to classify the stirring torque into low torque, medium torque, and high torque. Then, combined with historical flocculation condition data, the flocculation condition types of different torque ranges are classified, and the classified torque data is labeled for easy retrieval later. At the same time, the accuracy of the classification results is verified, and stirring torques with large classification deviations are removed.

[0032] The categorized stirring torque is paired with the preliminary time-series fluctuation features. The paired torque-fluctuation features are then input into the feature association layer of the CatBoost model to initialize the association weights. Subsequently, the CatBoost model's decision tree is iteratively trained to uncover the association patterns between different torque ranges and turbidity fluctuation amplitudes. Simultaneously, through model iterative training, the association patterns between different torque ranges and floc particle size fluctuation amplitudes are uncovered. The correlation coefficients between stirring torque and turbidity fluctuations and floc particle size fluctuations are calculated to quantify the nonlinear association strength. Then, core associations with correlation coefficients greater than a set threshold are selected, and weak association interferences are eliminated. The quantified nonlinear association results are output, providing a basis for subsequent torque association feature extraction.

[0033] Preferably, step S4 includes the following specific steps: Step S41: Assign weights to the preliminary temporal fluctuation features and torque correlation features through the attention mechanism fusion model. The preliminary temporal fluctuation features and torque correlation features are then fused based on the weights to obtain the preliminary fused features. Step S42: Introduce a feature filter to remove redundancy and strengthen core features of the preliminary fusion features. Standardize and normalize the filtered preliminary fusion features to obtain complete fluctuation features. Step S43: Based on the complete fluctuation characteristics, calculate the real-time fluctuation coefficients of turbidity and floc particle size using the variance calculation method; Step S44: Based on the complete fluctuation characteristics, the fluctuation duration of turbidity and floc particle size is calculated through continuous anomaly monitoring.

[0034] When fusing and coupling the initial time-series fluctuation features and torque correlation features, an attention mechanism fusion model is used to calculate the importance weights of the initial time-series fluctuation features and torque correlation features, realizing the adaptive allocation of feature weights. Then, based on the adaptively allocated weights, the initial time-series fluctuation features and torque correlation features are weighted and fused to obtain the initial fused features. At this time, a feature filter is introduced to remove redundancy and strengthen the core features of the initial fused features to improve feature purity. The filtered fused features are then subjected to dual processing of standardization and normalization to eliminate dimensional differences and feature interference, obtaining complete fluctuation features. Finally, based on the complete fluctuation features, variance analysis and continuous anomaly monitoring methods are used to calculate the real-time fluctuation coefficients and fluctuation durations of turbidity and floc particle size.

[0035] Preferably, step S5 includes the following specific steps: Step S51: When the real-time fluctuation coefficient and fluctuation duration of turbidity, floc particle size are all less than the preset fluctuation coefficient threshold and fluctuation duration threshold, it is determined that the flocculation effect meets the standard. Step S52: When the flocculation effect is not up to standard, a speed adjustment command is triggered. Based on the degree of fluctuation exceeding the standard, an appropriate stirring speed adjustment value is calculated and output to the flocculation stirring equipment.

[0036] To evaluate the flocculation effect, a fluctuation coefficient threshold and a fluctuation duration threshold are set. Flocculation is considered to be up to standard only when the real-time fluctuation coefficient and fluctuation duration of turbidity and floc size are both less than the fluctuation coefficient threshold and fluctuation duration threshold, respectively. If any one of them exceeds the threshold, flocculation is considered to be unsatisfactory. In this case, the stirring torque of the stirring equipment needs to be adjusted. After calculating the appropriate stirring speed adjustment value based on the degree of exceedance, it can be output to the flocculation stirring equipment to achieve closed-loop adjustment of the sewage flocculation process. It can respond in real time to abnormal flocculation effects, reduce fluctuations in effluent water quality, and the entire flocculation evaluation and adjustment process is intelligently processed without manual intervention, adapting to the needs of precise and intelligent sewage treatment.

[0037] 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 evaluating the flocculation effect in wastewater treatment, characterized in that, Includes the following steps: Step S1: Collect turbidity, floc particle size and stirring torque during the flocculation process using a multimodal sensor array, and perform preprocessing. Step S2: Input the pre-processed turbidity and floc particle size into the pre-trained time series feature capture model to obtain preliminary time series fluctuation features; Step S3: Input the pre-processed stirring torque into the pre-trained correlation analysis model to obtain torque correlation features; Step S4: The preliminary time-series fluctuation characteristics and torque correlation characteristics are fused and coupled to obtain complete fluctuation characteristics, and the real-time fluctuation coefficients and fluctuation durations of turbidity and floc particle size are calculated. Step S5: Evaluate the flocculation effect based on the real-time fluctuation coefficient and fluctuation duration of turbidity, floc particle size, and when the flocculation effect is not up to standard, output a command to adjust the speed of the flocculation stirring equipment.

2. The method for evaluating the flocculation effect in wastewater treatment according to claim 1, characterized in that, The specific steps of step S1 are as follows: Step S11: Arrange turbidity sensors and particle size sensors at different monitoring points in the flocculation reaction tank, and install torque sensors on the stirring shaft of the flocculation stirring equipment. Step S12: Collect turbidity, floc particle size, and stirring torque using a turbidity sensor, a particle size sensor, and a torque sensor, respectively. Step S13: Noise reduction and standardization processing are performed on turbidity, floc particle size and stirring torque.

3. The method for evaluating the flocculation effect in wastewater treatment according to claim 2, characterized in that, The step of locating the monitoring point in step S11 is as follows: The flow field distribution in the flocculation reaction tank was simulated by CFD, and the velocity gradient distribution cloud map of the flow field was obtained. Based on the flow field velocity gradient cloud map, the flocculation reaction tank is divided into a high gradient zone, a medium gradient zone, and a low gradient zone. The vertical planes of the central axis of the high gradient zone, medium gradient zone and low gradient zone are obtained respectively. A virtual line is constructed along the water flow direction, starting from the inlet of the flocculation reaction tank on the side of the high gradient zone. The intersection of the virtual line and the vertical plane of the central axis is used as the monitoring point.

4. The method for evaluating the flocculation effect in wastewater treatment according to claim 1, characterized in that, The specific steps of step S2 include: Step S21: Input the pre-processed turbidity and floc particle size into the pre-trained BiLSTM-GRU hybrid model, respectively. Step S22: Extract bidirectional time-series features from turbidity and floc size time-series data using the BiLSTM module, and perform efficient feature compression and redundancy removal on the bidirectional time-series features using the GRU module. Step S23: Integrate the optimized bidirectional timing features output by the GRU module and output them as preliminary timing fluctuation features.

5. The method for evaluating the flocculation effect in wastewater treatment according to claim 4, characterized in that, The specific steps of step S22 are as follows: The forward LSTM unit and the reverse LSTM unit of the BiLSTM module capture historical fluctuation patterns and future fluctuation trends, respectively. The historical fluctuation patterns and future fluctuation trends of turbidity and floc size are spliced ​​together to obtain the bidirectional time series features corresponding to turbidity and floc size, and then integrated to obtain a preliminary set of bidirectional time series features. The initial set of bidirectional time series features is input into the reset gate of the GRU module to filter and retain the core fluctuation features; The core fluctuation features are weighted by the update gate of the GRU module, and the weighted core fluctuation features are nonlinearly transformed by the candidate activation unit. The features after nonlinear transformation are subjected to dimensional compression and redundant interference information is removed, and the optimized bidirectional time series features are output.

6. The method for evaluating the flocculation effect in wastewater treatment according to claim 1, characterized in that, The specific steps of step S3 include: Step S31: Construct an association analysis model based on the CatBoost model, which is trained using historical association data; Step S32: Input the pre-processed stirring torque into the correlation analysis model, classify the stirring torque using the CatBoost model, and explore the nonlinear correlation between stirring torque and turbidity and floc particle size fluctuation. Step S33: Based on the correlation analysis results, extract the characteristic parameters that reflect the influence of torque on fluctuations, and output them as torque correlation features.

7. The method for evaluating the flocculation effect in wastewater treatment according to claim 6, characterized in that, The specific steps of step S32 are as follows: Based on the extreme and average values ​​of the pre-treated stirring torque, an adaptive threshold division method is used to divide the stirring torque into three torque ranges: low torque, medium torque, and high torque. By combining historical flocculation data, the stirring intensity level corresponding to each torque range is determined and correlated with the torque range to clarify the flocculation condition type corresponding to different torque ranges. The categorized stirring torque is labeled and matched with the preliminary time-series fluctuation features before being input into the feature association layer of the CatBoost model to initialize the association weights. Through iterative training of the CatBoost model, the correlation between different torque ranges and turbidity fluctuation amplitudes and floc particle size fluctuation amplitudes was explored. Calculate the correlation coefficients between stirring torque and turbidity fluctuation and floc particle size fluctuation, quantify the nonlinear correlation strength as the nonlinear correlation result and output it.

8. The method for evaluating the flocculation effect in wastewater treatment according to claim 1, characterized in that, The specific steps of step S4 include: Step S41: Assign weights to the preliminary temporal fluctuation features and torque correlation features through the attention mechanism fusion model. The preliminary temporal fluctuation features and torque correlation features are then fused based on the weights to obtain the preliminary fused features. Step S42: Standardize and normalize the preliminary fusion features to obtain complete fluctuation features; Step S43: Based on the complete fluctuation characteristics, calculate the real-time fluctuation coefficients of turbidity and floc particle size using the variance calculation method; Step S44: Based on the complete fluctuation characteristics, the fluctuation duration of turbidity and floc particle size is calculated through continuous anomaly monitoring.

9. The method for evaluating the flocculation effect in wastewater treatment according to claim 8, characterized in that, After obtaining the preliminary fusion features in step S41 A feature filter is introduced to remove redundancy and strengthen core features in the initial fusion.

10. The method for evaluating the flocculation effect in wastewater treatment according to claim 1, characterized in that, The specific steps of step S5 include: Step S51: When the real-time fluctuation coefficient and fluctuation duration of turbidity, floc particle size are all less than the preset fluctuation coefficient threshold and fluctuation duration threshold, it is determined that the flocculation effect meets the standard. Step S52: When the flocculation effect is not up to standard, a speed adjustment command is triggered. Based on the degree of fluctuation exceeding the standard, an appropriate stirring speed adjustment value is calculated and output to the flocculation stirring equipment.

Images