Aeration control system and method based on vision algorithm
Patent Information
- Application Number
- HK42026125515
- Authority / Receiving Office
- HK · HK
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2026-06-30
- Publication Date
- 2026-09-11
Smart Images

Figure 00000000_0000_ABST
Abstract
Description
(19) State Intellectual Property Office (12) Invention Patent Application (10) Application Publication Number (43) Application Publication Date (21) Application Number 202511089891.7 (22) Application Date 2025.08.05 (71) Applicant Jiangsu Lanchuang Intelligent Technology Co., Ltd. Address 214000 No. 28-52, Nanhu Middle Road, Wuxi City, Jiangsu Province (72) Inventors Huang Qinglan, Huang Hongjuan, Chen Shanshan (74) Patent Agency Harbin Sunshine Huiyuan Intellectual Property Agency Co., Ltd. 23211 Patent Attorney Hou Min (51) Int.Cl. G05B 19 / 042 (2006.01) (54) Invention Title An Aeration Control System and Method Based on Visual Algorithm (57) Abstract This invention discloses an aeration control system and method based on visual algorithm, which belongs to the field of intelligent control technology for sewage treatment. The method described introduces foam morphology, dynamic characteristics, and statistical features, and combines these with aeration equipment parameters and water quality indicators to predict aeration volume. Random forests are then used to correct the predicted values. Compared to traditional single-parameter DO control, this invention can reflect the aerobic state of microorganisms in real time and provide precise aeration volume control. Precise aeration avoids "ineffective aeration," significantly reducing energy consumption. Furthermore, this invention can suppress sludge bulking through real-time feedback adjustment, improving system stability. Actual measurements at wastewater treatment plants have shown that the removal rate of organic matter in wastewater is effectively improved, and the sludge index and sludge bulking rate are significantly reduced after adopting this control method. Therefore, this invention can be widely applied to intelligent control in various wastewater treatment processes. Claims (2 pages), Description (7 pages), Drawings (2 pages), CN 121254671 A 2026.01.02 CN 1 21 25 46 71 A 1. An aeration control method based on a visual algorithm, characterized in that the method includes: Step 1: Acquiring an image of the water surface in an aeration tank and preprocessing it; Step 2: Segmenting the preprocessed image of the water surface in the aeration tank to obtain a foam bounding box; Step 3: Performing pixel-level segmentation on the region within the foam bounding box to generate a foam mask; Step 4: Calculating foam morphology features, dynamic features, and statistical features based on the foam mask, wherein the foam morphology features include foam area, equivalent diameter, and aspect ratio; the dynamic features include foam generation rate and foam dissipation time; and the statistical features include the standard deviation of foam area over a continuous time period; Step 5: Based on the foam morphology features, dynamic features, statistical features, aeration equipment parameters, and water quality indicators, using an LSTM model to predict the future optimal aeration rate, and using a random forest model to correct the value to obtain the final target aeration rate; Step 6: Convert the target aeration volume into blower speed and send it to the blower control system.2. The aeration control method based on a vision algorithm according to claim 1, characterized in that step 2 uses a YOLOv8 model to obtain the foam bounding box. 3. The aeration control method based on a vision algorithm according to claim 1, characterized in that step 3 uses a U-Net++ model to perform pixel-level segmentation of the region within the bounding box. 4. The aeration control method based on a vision algorithm according to claim 1, characterized in that the method further includes: collecting foam morphology features at preset time intervals, calculating the foam dissipation time, and adjusting the fan speed according to the dissipation time. 5. An aeration control system based on a visual algorithm, characterized in that the system comprises: a video acquisition module for acquiring images of the water surface in an aeration tank; a data transmission module for transmitting the acquired image information to an image processing module; a data processing module configured to preprocess the images, segment the preprocessed aeration tank water surface image to obtain a foam bounding box; perform pixel-level segmentation on the region within the foam bounding box to generate a foam mask; calculate foam morphology features, dynamic features, and statistical features based on the foam mask, wherein the foam morphology features include foam area, equivalent diameter, and aspect ratio; the dynamic features include foam generation rate and foam dissipation time; and the statistical features include the standard deviation of foam area over a continuous time period; a prediction module configured to predict the future optimal aeration rate using an LSTM model based on the foam morphology features, dynamic features, statistical features, aeration equipment parameters, and water quality indicators, and to correct the value using a random forest model to obtain the final target aeration rate; and an output module configured to convert the target aeration rate into blower speed and send it to the blower control system. 6. The aeration control system based on a visual algorithm according to claim 5, characterized in that the video acquisition module adopts an industrial camera, installed directly above the aeration tank, with the lens vertically shooting down at the water surface. 7. The aeration control system based on a visual algorithm according to claim 5, characterized in that the data transmission module transmits the video stream to the data processing module via gigabit Ethernet. 8. The aeration control system based on a visual algorithm according to claim 5, characterized in that the preprocessing includes: using median filtering to remove salt and pepper noise, and combining adaptive histogram equalization to enhance the contrast between foam and water. 9. An aeration control device based on a visual algorithm, characterized in that it includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement the aeration control method based on the visual algorithm according to any one of claims 1 to 4 when the computer program is executed. 10. A computer-readable storage medium, characterized in that the storage medium stores a computer program, which, when executed...When the computer program is executed by the processor, it implements the aeration control method based on the visual algorithm as described in any one of claims 1 to 4. Claims 2 / 2 Page 3 CN 121254671 A Aeration Control System and Method Based on Visual Algorithm Technical Field
[0001] This invention relates to an aeration control system and method based on a visual algorithm, belonging to the field of intelligent control technology for wastewater treatment. Background Art
[0002] As the core unit of wastewater treatment process, the aeration efficiency of the aeration tank directly affects the degradation effect of organic matter (COD / BOD5) and the effluent quality of the activated sludge process. Traditional aeration control mainly has the following pain points:
[0003] First, the experience-based control mode has strong lag. Most wastewater treatment plants rely on operators to set aeration parameters (such as aeration duration and blower speed) based on historical experience. However, wastewater quality (intermittent discharge of industrial wastewater and diurnal fluctuations in domestic sewage), water volume (dry season / rainy season differences can reach 3-5 times), and microbial activity (temperature and pH fluctuations) are dynamic. Parameters controlled based on experience are difficult to match the actual oxygen demand in real time, leading to insufficient aeration (inhibition of microbial activity, decreased COD / BOD5 removal rate) or excessive aeration (energy waste, increased risk of sludge bulking).
[0004] Second, the one-sidedness of single-parameter control is prominent. Although some systems introduce dissolved oxygen (DO) sensors as feedback, DO only reflects the instantaneous concentration of dissolved oxygen in the water body and cannot characterize the "spatiotemporal heterogeneity" of microbial oxygen demand. The DO distribution difference in different areas of the aeration tank (such as the front / back end of the plug flow process) can reach 2-3 mg / L. Data from a single measuring point cannot represent the overall demand, thus making it impossible to achieve accurate aeration control.
[0005] Third, the contradiction between energy consumption and efficiency is prominent. Statistics show that aeration energy consumption in sewage treatment plants accounts for as much as 40%-60%, while under the traditional control mode, about 35% of the aeration volume is not effectively utilized (i.e., "ineffective aeration"), which directly leads to annual electricity costs of over 10 billion yuan. At the same time, excessive aeration will damage the sludge floc structure (sludge index SVI increases by 20%-30%), causing sludge bulking and other operational abnormalities.
[0006] In order to improve the efficiency and accuracy of aeration control and reduce energy consumption, the present invention provides an aeration control system and method based on a visual algorithm. The technical solution is as follows:
[0007] The first objective of the present invention is to provide an aeration control method based on a visual algorithm, the method comprising:
[0008] Step 1: Acquiring an image of the water surface of the aeration tank and preprocessing it;
[0009] Step 2: Segmenting the preprocessed image of the water surface of the aeration tank to obtain a foam bounding box;
[0010] Step 3: Performing pixel-level segmentation on the region within the foam bounding box to generate a foam mask;
[0011] Step 4: Calculating the foam morphological features, dynamic features, and statistical features based on the foam mask, the foam morphological features, dynamic features, and statistical features ...The morphological features include foam area, equivalent diameter, and aspect ratio; the dynamic features include foam generation rate and foam dissipation time; and the statistical features include the standard deviation of foam area over a continuous time period.
[0012] Step 5: Based on the foam morphological features, dynamic features, statistical features, aeration equipment parameters, and water quality indicators, the LSTM model is used to predict the future optimal aeration volume, and the value is corrected using a random forest model to obtain the final target aeration volume.
[0013] Step 6: The target aeration volume is converted into the blower speed and sent to the blower control system.
[0014] Optionally, step 2 uses the YOLOv8 model to obtain the foam bounding box.
[0015] Optionally, step 3 uses the U-Net++ model to perform pixel-level segmentation of the region within the bounding box.
[0016] Optionally, the method further includes: collecting foam morphological features every preset time interval, calculating the foam dissipation time, and adjusting the blower speed according to the dissipation time.
[0017] A second objective of the present invention is to provide an aeration control system based on a visual algorithm, the system comprising:
[0018] a video acquisition module for acquiring images of the water surface in an aeration tank;
[0019] a data transmission module for transmitting the acquired image information to an image processing module;
[0020] a data processing module configured to preprocess the image, segment the preprocessed aeration tank water surface image to obtain a foam bounding box; perform pixel-level segmentation of the region within the foam bounding box to generate a foam mask; calculate foam morphological features, dynamic features, and statistical features based on the foam mask, wherein the foam morphological features include foam area, equivalent diameter, and aspect ratio; the dynamic features include foam generation rate and foam dissipation time; and the statistical features include the standard deviation of foam area over a continuous time period;
[0021] a prediction module configured to predict the future optimal aeration volume using an LSTM model based on the foam morphological features, dynamic features, statistical features, aeration equipment parameters, and water quality indicators, and to correct the value using a random forest model to obtain the final target aeration volume;
[0022] The output module is configured to convert the target aeration volume into blower speed and send it to the blower control system.
[0023] Optionally, the video acquisition module uses an industrial camera, installed directly above the aeration tank, with the lens vertically shooting down at the water surface.
[0024] Optionally, the data transmission module transmits the video stream to the data processing module via gigabit Ethernet.
[0025] Optionally, the preprocessing includes: using median filtering to remove salt-and-pepper noise, and combining adaptive histogram equalization to enhance the contrast between foam and water.
[0026] A third objective of the present invention is to provide an aeration control device based on a visual algorithm, including a memory and a processor;
[0027] The memory is used to store a computer program;
[0028] The processor is used to implement the aeration control method based on the vision algorithm as described in any of the preceding claims when the computer program is executed.
[0029] A fourth object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the aeration control method based on the vision algorithm as described in any of the preceding claims.
[0030] The beneficial effects of the present invention are as follows:
[0031] The present invention introduces foam morphology characteristics, dynamic characteristics and statistical characteristics, and then combines aeration equipment parameters and water quality index characteristics to predict the aeration volume, and uses random forest to correct the predicted value. Compared with the traditional single parameter control of DO, the present invention can reflect the aerobic state of microorganisms in real time and provide precise control of aeration volume. According to the actual test of the sewage treatment plant, after adopting the control method of the present invention, the COD removal rate increased from 75% to 88% (an increase of 17.3%), the BOD5 removal rate increased from 68% to 82% (an increase of 17.6%), and the effluent COD was stable at ≤40mg / L and BOD5 at ≤10mg / L (Class A standard).
[0032] The present invention avoids "ineffective aeration" based on the precise aeration of foam characteristics, and significantly reduces energy consumption. The average operating power of the blower is reduced from 120kW to 90kW (a reduction of 25%), and the annual power consumption is reduced by about 100,000 kWh. Calculated at an industrial electricity price of 0.8 yuan / kWh, the annual electricity cost is saved by 80,000 yuan (based on 2 aeration systems). Specification 2 / 7 pages 5 CN 121254671 A
[0033] In addition, the present invention can suppress sludge bulking through real-time feedback adjustment, improve the stability of system operation, reduce the sludge index (SVI) from 150mL / g to 100mL / g (a reduction of 33%), and reduce the sludge bulking rate from 12% to below 2%. Brief Description of the Drawings
[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of 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.
[0035] Figure 1 is a structural diagram of the aeration control system based on the visual algorithm of the present invention.
[0036] Figure 2 is a flowchart of the aeration control method based on the visual algorithm of the present invention. Detailed Embodiments
[0037] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.
[0038] Embodiment 1:
[0039] This embodiment provides an aeration control method based on a visual algorithm. Referring to Figure 2, the method includes:
[0040] Step 1: Acquire images of the water surface in the aeration tank.
[0041] This embodiment uses an industrial camera to vertically capture images of the aeration tank surface with preset parameters (25 frames / second, 1920×1080 resolution). The video stream is compressed using H.265 encoding and then transmitted to the data processing module via Gigabit Ethernet (IEEE 802.3ab), with a transmission delay ≤200ms and a packet loss rate <0.1%.
[0042] Step 2: Image preprocessing and foam feature extraction.
[0043] Median filtering (3×3 kernels) is applied to each frame of the image for noise reduction, and CLAHE is applied to enhance contrast, outputting a preprocessed image.
[0044] The YOLOv8 model is used to infer the preprocessed image and output the foam bounding box.
[0045] The region within the bounding box is segmented at the pixel level based on the U-Net++ model to generate a foam mask.
[0046] The quantitative features of the foam (time resolution 1 frame / second) are calculated based on the segmentation results, including: morphological features, dynamic features, and statistical features. The morphological features include area (pixel → actual area, converted by calibration coefficient), equivalent diameter (based on circle fitting), and aspect ratio (longest axis / shortest axis of the foam region). The dynamic features include foam generation rate (new foam area per unit time) and average dissipation time (time from appearance to complete fusion). The statistical feature is the standard deviation of foam area within 1 minute (reflecting foam stability). The calculation methods and physical meanings of some features are shown in Table 1 below:
[0047] Table 1 Calculation methods and physical meanings of some features
[0048] Feature type Calculation formula Physical meaning Foam area Area = number of mask pixels × 0.01m2 / pixel Reflects the scale of foam coverage Equivalent diameter √(4×area / π) Characterizes the average size of foam Aspect ratio Long side / short side of the maximum inscribed rectangle Indicates the irregularity of foam shape Foam dissipation time Time (seconds) for the peak area to decay to 10% Directly related to microbial oxygen demand intensity
[0049] Step 3: Multi-source data fusion and pattern mining.
[0050] Data Synchronization: Real-time foam characteristics (1 second / time) are aligned with aeration equipment parameters (1 second / time) and water quality indicators (1 minute / time) through timestamps to generate time series data. (Instruction manual, page 3 / 7, CN 121254671 A)
[0051] Model Input: The most recent 30 minutes of historical data (i.e., 30 time steps, each time step containing foam characteristics, aeration parameters, and water quality indicators) are used as model input.
[0052] Parameter Prediction: The LSTM model outputs the optimal aeration volume prediction value for the next 10 minutes, and the random forest model corrects this value (error correction range ±3%) to obtain the final target aeration volume.
[0053] Step 4: Dynamic adjustment and feedback optimization of aeration equipment.
[0054] Baseline Control: The target aeration volume is converted into fan speed (speed = target aeration volume × K, K = 0.12 r / min).The data (min·m3 / h) is sent to the blower PLC via Modbus TCP.
[0055] Feedback adjustment: The latest foam characteristics are collected every 5 minutes, and the foam dissipation time is calculated (the time from the current moment back to when the foam area decreases by 90%). If the dissipation time is greater than the predicted value × 1.2 (insufficient oxygen supply), the speed is increased by 2%; if the dissipation time is less than the predicted value × 0.8 (excessive oxygen supply), the speed is decreased by 1%, where the predicted value refers to the predicted future aeration volume.
[0056] Abnormal protection: When the foam area suddenly increases by 50% (possibly due to a sudden change in the influent shock load), the system automatically switches to "safe mode", maintains the current aeration volume, and triggers an alarm.
[0057] Example 2
[0058] This example provides an aeration control system based on a visual algorithm. See Figure 1. The system includes the following modules.
[0059] (1) Video Acquisition Module
[0060] Hardware: An industrial-grade anti-shake network camera (such as Hikvision DS-2CD6326FWDV2-I) is used, installed 6-8 meters directly above the aeration tank (preferably 6.5 meters, balancing field of view coverage and foam detail capture), with the lens shooting vertically downwards at the water surface (angle of 90°±2°) to avoid interference from water surface reflections.
[0061] Parameters: Supports real-time acquisition at 25 frames / second (PAL format), resolution 1920×1080 (Full HD), built-in automatic exposure (AE) and wide dynamic range (WDR) adjustment functions to adapt to different lighting conditions (such as cloudy / sunny days).
[0062] Transmission: The video stream is transmitted to the data processing module via Gigabit Ethernet (IEEE 802.3ab), with a latency ≤ 200ms.
[0063] (2) Data processing module
[0064] Image preprocessing: Median filtering (kernel size 3×3) is used to remove salt and pepper noise, and adaptive histogram equalization (CLAHE) is combined to enhance the contrast between foam and water.
[0065] Foam detection and segmentation: First, the foam region is located using the YOLOv8 object detection algorithm (a custom foam dataset is used during training, containing foam samples under different lighting and densities, with mAP@0.5≥92%); then, the U-Net++ semantic segmentation algorithm is used to perform pixel-level segmentation of the foam region (loss function is Dice Loss + cross-entropy, segmentation accuracy ≥ 95%), and a foam mask is output;
[0066] Feature extraction: Based on the segmentation results, the quantitative features of the foam are calculated (temporal resolution 1 frame / second), including:
[0067] Morphological features: area (pixels → actual area, converted by calibration coefficients), equivalent diameter (based on circle fitting), aspect ratio (longest axis / shortest axis of the foam region);
[0068] Dynamic features: foam generation rate (new foam area per unit time), average dissipation time (time from appearance to complete fusion);
[0069] Statistical characteristics: Standard deviation of foam area within 1 minute (reflecting foam stability).
[0070] (3) Prediction module
[0071] Input characteristics: Real-time foam characteristics (the above morphological / dynamic / statistical characteristics) + synchronous aeration equipment parameters (aeration volume, aeration duration, blower speed) + water quality indicators (COD, BOD5, DO, water temperature, pH).
[0072] Model architecture: LSTM (Long Short-Term Memory Network) is used as the main model, combined with Random Forest for feature importance screening and ensemble learning to improve the model's generalization ability.
[0073] LSTM layer: Input dimension is 15 (5 types of features × 3 time steps), hidden layer has 128 neurons, output dimension is 1 (predicted aeration volume).
[0074] Random Forest: Used to correct the output of LSTM, integrating 100 decision trees, with the feature importance threshold set to 0.01 (removing low-contribution features), and the correction formula is:
[0075]
[0076] Wherein, wi represents the feature weight, and ΔFi represents the feature bias.
[0077] Wherein, the feature weight wi represents the feature weight output by the random forest model, reflecting the importance of each input feature to the correction result.
[0078] How to obtain the feature weight wi: First, the random forest model is trained on historical data (training set). During the training process, the model will automatically calculate the importance of each feature; according to the setting of "the feature importance threshold is set to 0.01", features with an importance score ≥ 0.01 are selected, low-contribution features (importance < 0.01) are removed, and only the weights of these high-importance features are retained to participate in the correction calculation; the final weight is the feature weight vector output by the random forest model after threshold filtering (each feature corresponds to a weight value).
[0079] Feature bias ΔFi represents the deviation between the actual value and the benchmark value (or expected value) of the i-th feature, used to quantify the degree of deviation of the feature from the ideal state.
[0080] How to obtain feature bias ΔFi: For real-time input features (such as real-time foam features, aeration equipment parameters, etc.), calculate the difference between its current value Fi and the benchmark value Fi, i.e., ΔFi; the benchmark value can be the historical statistical mean, the model prediction value, or the standard value set by the process (which needs to be determined according to the business scenario, for example, based on the feature value corresponding to the historical optimal aeration volume).
[0081] Random forest works in the following way:
[0082] Each decision tree splits the input feature, and finally obtains a continuous output by averaging the prediction values of all trees.
[0083] Correction logic: Random forest does not directly classify, but learns the "residual relationship between LSTM prediction value and true value", that is, the input is a feature vector, and the output is the correction amount ∑wiΔFi of the LSTM prediction value.
[0084] FormulaThis can be understood as follows: if ∑wiΔFi>0, it means that the current feature is biased towards "the need to increase the aeration rate", and the final result is amplified based on the LSTM prediction value; if ∑wiΔFi<0, it is biased towards "the need to reduce the aeration rate", and the result is reduced. The essence is to dynamically adjust the direction and magnitude of the LSTM prediction by weighted sum of feature biases to make up for the model's insufficient sensitivity to real-time feature changes.
[0085] Training data: Collected continuous operation data of a municipal sewage treatment plant for six months (180 days) (sampling frequency 1 time / minute), containing 5400 sets of samples, of which the training set accounts for 70% (3780 sets), the validation set accounts for 20% (1080 sets), and the test set accounts for 10% (540 sets).
[0086] Model performance: After cross-validation, the root mean square error (RMSE) of the model's predicted aeration rate is ≤5m3 / h (accounting for 2%-3% of the aeration rate range), and the coefficient of determination (R2) is ≥0.92.
[0087] (4) Aeration Control Module (Execution and Feedback Mechanism) Manual 5 / 7 Page 8 CN 121254671 A
[0088] Communication: Connected to the PLC controller of the Roots blower (such as Jin Gong JGR600) via Modbus TCP protocol, supporting stepless speed adjustment from 0 to 100%.
[0089] Control Strategy:
[0090] Baseline Adjustment: Calculate the target speed (blower speed is linearly related to aeration volume, calibration coefficient K = 0.12 r / min·m3 / h) based on the optimal aeration volume (range 80-250 m3 / h) output by the prediction module.
[0091] Feedback Optimization: Re-collect foam characteristics every 5 minutes. If the actual foam dissipation time is 20% longer than the predicted value (indicating insufficient oxygen supply), increase the aeration volume by 5%; if the foam area is 15% smaller than the predicted value (indicating excessive oxygen supply), decrease the aeration volume by 3%.
[0092] Actuator: The speed of the blower motor is adjusted by a frequency converter (such as ABB ACS880), with a response time ≤ 2 seconds.
[0093] Example 3:
[0094] This example implements the present invention using a municipal sewage treatment plant with a daily processing capacity of 50,000 tons as an example. The specific scheme is as follows:
[0095] (1) Hardware deployment
[0096] Video acquisition module: A network camera (4 million pixels, F1.6 large aperture lens, minimum illumination 0.005 Lux) is selected and installed 6.5 meters above the aeration tank (covering a tank surface width of 12 meters). It is fixed by a stainless steel bracket and has an anti-shake rating of IP66.
[0097] Data processing module: An industrial control computer (Intel i7-10700 CPU, 32GB memory, supports 64-channel video decoding) is used and deployed in the plant control room.
[0098] Prediction module: A server (2×Intel Xeon Gold 6330 CPU, 1TB memory) is used to run Python.3.8 TensorFlow 2.6 framework in environment.
[0099] Aeration control module: PLC (CPU 1516-3PN / DP) is connected to blower (rated air volume 300m3 / min, rated power 132kW) through Profibus DP bus, supporting 0-100% frequency conversion adjustment.
[0100] (2) Software configuration and model training
[0101] Image annotation: 1000 water surface images of aeration tank (including scenes with different lighting and foam density) were manually annotated to build a foam detection dataset (YOLOv8 training set). The annotation content includes the foam bounding box coordinates (x,y,w,h).
[0102] Model training:
[0103] YOLOv8: The CSPDarknet backbone network was used and trained for 100 rounds (batch size = 16) on a custom dataset. The initial learning rate was 0.01 (cosine annealing decay), and mAP@0.5 reached 93.2%.
[0104] U-Net++: VGG16 was used as the encoder, and the loss function was Dice Loss (weight 0.7) + cross-entropy loss (weight 0.3). After 50 training rounds (batch size = 8), the Dice coefficient for segmentation reached 0.96.
[0105] LSTM + Random Forest: The features (area, diameter, aspect ratio, dissipation time, etc., 15 dimensions) output by YOLOv8 and U-Net++ were input into the LSTM model. The training parameters were: learning rate 0.001, batch size = 32, and training was carried out for 100 rounds (stopping when the validation set loss no longer decreased). Then, 100 random forests were integrated (each tree had a maximum depth of 10). The final predicted RMSE was 4.2 m3 / h (accounting for 1.7% of the aeration range).
[0106] (3) System integration and operation test
[0107] Cold start phase (week 1): The system is operated in "safe mode" (aeration volume is maintained at 80% of the traditional control), foam characteristics and water quality data are collected, and the communication stability of each module (packet loss rate < 0.1%) and the accuracy of model prediction (the deviation between predicted aeration volume and actual demand < 5%) are verified.
[0108] Normal operation phase (weeks 2-12): Switch to "intelligent control mode" and dynamically adjust the aeration volume according to the model prediction results. Instruction manual, pages 6 / 7, CN 121254671 A
[0109] Key indicator tests:
[0110] Aeration volume fluctuation range: 80-250 m³ / h (covering the traditional control range of 50-300 m³ / h);
[0111] Feedback adjustment response time: from the change in foam characteristics to the completion of aeration volume adjustment ≤ 10 seconds;
[0112] Effluent water quality: continuous monitoring for 90 days, COD ≤ 40 mg / L (compliance rate 100%), BOD5 ≤ 10 mg / L (compliance rate 100%);
[0113] Energy consumption: The average monthly electricity cost has decreased by 25% (from RMB 32,000 / month to RMB 24,000 / month).
[0114] Some steps in the embodiments of the present invention can be implemented using software, and the corresponding software program can be stored in a readable storage medium, such as an optical disc or a hard disk.
[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. Instruction sheet 7 / 7 page 10 CN 121254671 A Figure 1 Instruction drawing 1 / 2 page 11 CN 121254671 A Figure 2 Instruction sheet drawing 2 / 2 page 12 CN 121254671 A Abstract The invention discloses an aeration control system and method based on visual algorithms, which belongs to the technical field of intelligent control for sewage treatment. The proposed method introduces foam morphological features, dynamic features and statistical features, and combines aeration equipment parameters and water quality index features to predict aeration volume. A random forest algorithm is adopted to correct the predicted values. Compared with the traditional single-parameter DO control, the invention can indicate the oxygen demand state of microorganisms in real time and achieve precise control of aeration volume. Precise aeration eliminates "ineffective aeration" and greatly reducesenergy consumption. In addition, the invention can suppress sludge bulking through real-time feedback regulation and improve the operational stability of the system. Field tests conducted in sewage treatment plants verify that the adoption of the control method of the invention effectively improves the removal rate of organic matter in sewage, and significantly reduces the sludge volume index and the incidence of sludge bulking. Therefore, the invention can be widely applied to intelligent control of various sewage treatment processes.
Claims
1. A method for aeration control based on vision algorithm, characterized in that, The method comprises: Step 1: Collecting an aeration tank water surface image and performing preprocessing; Step 2: Segmenting the preprocessed aeration tank water surface image to obtain a foam bounding box; Step 3: Performing pixel-level segmentation on the region within the foam bounding box to generate a foam mask; Step 4: Calculating foam morphological features, dynamic features, and statistical features according to the foam mask, wherein the foam morphological features include foam area, equivalent diameter, and aspect ratio, the dynamic features include foam generation rate and foam dissipation time, and the statistical features include standard deviation of foam area within consecutive time; Step 5: Based on the foam morphological features, dynamic features, statistical features, aeration equipment parameters, and water quality indicators, using an LSTM model to predict a future optimal aeration amount prediction value, and using a random forest model to correct the value to obtain a final target aeration amount; Step 6: Converting the target aeration amount into fan speed and sending it to the blower control system.
2. The visual algorithm based aeration control method of claim 1, wherein, The step 2 uses a YOLOv8 model to obtain the foam bounding box.
3. The visual algorithm based aeration control method of claim 1, wherein, The step 3 uses a U-Net++ model to perform pixel-level segmentation on the region within the bounding box.
4. The visual algorithm based aeration control method of claim 1, wherein, The method further comprises: collecting foam morphological features every preset time, calculating foam dissipation time, and adjusting the fan speed according to the dissipation time.
5. A vision algorithm based aeration control system, characterized in that, The system comprises: A video acquisition module for acquiring an aeration tank water surface image; A data transmission module for transmitting the collected image information to an image processing module; A data processing module configured to preprocess the image, segment the preprocessed aeration tank water surface image to obtain a foam bounding box, perform pixel-level segmentation on the region within the foam bounding box to generate a foam mask, and calculate foam morphological features, dynamic features, and statistical features according to the foam mask, wherein the foam morphological features include foam area, equivalent diameter, and aspect ratio, the dynamic features include foam generation rate and foam dissipation time, and the statistical features include standard deviation of foam area within consecutive time; A prediction module configured to predict a future optimal aeration amount prediction value based on the foam morphological features, dynamic features, statistical features, aeration equipment parameters, and water quality indicators, and correct the value using a random forest model to obtain a final target aeration amount; An output module configured to convert the target aeration amount into fan speed and send it to the blower control system.
6. The visual algorithm based aeration control system of claim 5, wherein, The video acquisition module uses an industrial camera installed directly above the aeration tank with the lens vertically shooting down at the water surface.
7. The vision algorithm based aeration control system of claim 5, wherein, The data transmission module transmits the video stream to the data processing module through a gigabit Ethernet.
8. The vision algorithm based aeration control system of claim 5, wherein, The preprocessing includes removing salt and pepper noise using median filtering and enhancing the contrast between foam and water body using adaptive histogram equalization.
9. A vision algorithm based aeration control device, characterized by, It comprises a memory and a processor; The memory is used to store a computer program; The processor is used to implement the visual algorithm-based aeration control method of any one of claims 1 to 4 when executing the computer program.
10. A computer-readable storage medium, characterized in that, The storage medium has a computer program stored thereon, and when the computer program is executed by the processor, the visual algorithm-based aeration control method of any one of claims 1 to 4 is implemented.
