An intelligent dosing control method for flocculation process based on image information

Abstract

The present invention relates to the field of sewage treatment technology, and specifically to an intelligent dosing control method for a flocculation process based on image information. An inlet water quality monitoring data set, a process water quality monitoring data set, an outlet water quality monitoring data set, an equipment operating condition monitoring data set, a dosing amount monitoring data set, and a flocculation area surface state monitoring data set are obtained; the inlet water quality, process water quality, outlet water quality, equipment operating condition, and dosing amount monitoring data sets are fused into a water quality operating condition dosing fusion data set, and the flocculation area surface state and dosing amount monitoring data sets are fused into an image dosing fusion data set; the water quality operating condition dosing fusion data set is input into a K-nearest neighbor regression sub-model with adjustable parameters, and the image dosing fusion data set is input into a target detection sub-model with adjustable parameters. The results of the two models are fused into a graph information fusion model, and the fusion is performed according to the weight coefficient normalized by the percentage error.

Description

Technical Field

[0001] The present invention relates to the technical field of sewage treatment, and in particular to an intelligent dosing control method for a flocculation process based on image information. Background Art

[0002] Supplying untreated raw water directly can lead to serious health and environmental problems. Raw water may contain harmful microorganisms, heavy metals, pesticide residues, and other chemical contaminants, which can cause illness and long-term health risks if consumed directly. Furthermore, untreated raw water can be unstable and susceptible to environmental changes and pollution. It can also contaminate the interior of water supply pipes, increasing maintenance costs and damaging the ecological environment. Therefore, rigorous water treatment processes are essential to ensure water quality meets standards and ensure water safety.

[0003] Flocculant addition plays a key role in water treatment. It neutralizes the charge of suspended particles in the water, causing them to aggregate into larger flocs, facilitating subsequent sedimentation and filtration. This not only effectively removes turbidity and organic matter from the water, but also significantly reduces turbidity and contaminant levels, improving water clarity and purity. Flocculant addition optimizes the entire water treatment process, improves treatment efficiency, ensures water safety and quality, and provides important safeguards for public health and environmental protection.

[0004] Existing water treatment dosing control relies on manual dosing rather than utilizing image information, which presents numerous drawbacks. Manual flocculant dosing is not only inefficient but also prone to inaccurate dosage due to human error, impacting treatment effectiveness and water quality stability. Furthermore, manual operation is difficult to monitor and adjust in real time, hindering rapid response to water quality changes. Image information, however, provides real-time visual data on water quality, enabling automated and intelligent control, improving flocculant dosing accuracy and response speed, thereby optimizing treatment results and reducing operating costs.

[0005] In summary, the existing technology has a problem in which the dosing control process relies on manual control without paying attention to image information, resulting in an inability to ensure that the dosage can accurately meet the needs of the water treatment process. The effluent from the water treatment does not meet the drinking water standards, which limits the efficiency and reliability of the water treatment. Summary of the Invention

[0006] The purpose of the present invention is to provide an intelligent dosing control method for flocculation process based on image information, so as to solve the problem that the dosing control in the existing technology relies solely on manual control and ignores the guiding role of image information on the change of dosing state, resulting in the inability to ensure that the dosing amount can accurately meet the needs of the water treatment process, and the water effluent from the water treatment does not meet the drinking water standards, which limits the efficiency and reliability of the water treatment.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent dosing control method for a flocculation process based on image information, comprising the following steps:

[0008] S1. Sampling influent water quality indicators to obtain an influent water quality monitoring dataset, sampling process water quality indicators to obtain a process water quality monitoring dataset, sampling effluent water quality indicators to obtain an effluent water quality monitoring dataset, collecting equipment data on water supply and water treatment equipment to obtain an equipment operating condition monitoring dataset; monitoring the dosage of the dosing device in the flocculation zone of the sedimentation tank to obtain a dosage monitoring dataset; and monitoring the video file of the surface of the flocculation zone of the sedimentation tank to obtain a flocculation zone surface state image dataset;

[0009] S2. Fusing the inlet water quality monitoring dataset, the process water quality monitoring dataset, the outlet water quality monitoring dataset, the equipment operating condition monitoring dataset, and the dosage monitoring dataset into a water quality operating condition dosage fusion dataset; fusing the flocculation area surface state image dataset and the dosage monitoring dataset into an image dosage fusion dataset;

[0010] S3. Input the water quality condition dosing fusion dataset into the K-nearest neighbor regression sub-model with adjustable parameters; input the image dosing fusion dataset into the target detection sub-model with adjustable parameters;

[0011] S4. Fusing the two model results into an image information fusion model with adjustable parameters, fusing them according to the weight coefficient after the percentage error is normalized, and embedding the obtained adjustable parameter fusion model into the control system terminal of the water treatment dosing equipment;

[0012] S5. Based on the obtained adjustable parameter image information fusion model, the dosing amount of the dosing equipment in the flocculation zone of the sedimentation tank in the water treatment dosing process is controlled according to the output result.

[0013] Furthermore, in the above S1, the monitoring time of the inlet water quality monitoring dataset, the process water quality monitoring dataset and the equipment operating condition monitoring dataset is time T, and the monitoring time of the outlet water quality monitoring dataset is time T-1.

[0014] Furthermore, in the aforementioned S1, the method for acquiring the flocculation zone surface state image dataset is as follows: the characteristics of the flocs on the surface of the flocculation zone of the sedimentation tank are captured from multiple angles, and the number, morphology, relative size, and existence time of the flocs on the surface of the flocculation zone of the sedimentation tank are extracted from the aerobic tank video state monitoring dataset through a general target monitoring algorithm to construct the flocculation zone surface state image dataset.

[0015] Furthermore, in the above-mentioned S1, the data processing method is: filling missing values ​​and deleting outliers.

[0016] Furthermore, in S2, the data fusion method is: aligning the data rows and data columns of the data set, and setting the data interval to the minimum sampling time.

[0017] Furthermore, in the S3 described above, the method for constructing the K nearest neighbor regression sub-model is as follows: first, prepare and preprocess 25% of the data set and divide it into a training set, a validation set, and a test set; store the training data, and when new data needs to be predicted, the algorithm calculates the distance between the new data point and all stored training data points; then, select the K nearest neighbors and perform regression fitting based on the target values ​​of these neighbors; repeat the above process until the loss function percentage error converges to less than 10% or reaches a predetermined number of training rounds of 1000, and outputs the adjustable parameters at the same time.

[0018] Furthermore, in the S3, the method for constructing the target detection sub-model is as follows: first, prepare and preprocess 25% of the data set and divide it into a training set, a validation set and a test set; then initialize the network weights, select the loss function and the optimizer; during the training process, calculate the output and loss value through forward propagation, and then update the weights through back propagation; use multiple batches of data for training in multiple training cycles, and use the validation set to evaluate the model performance after each cycle and adjust the hyperparameters; repeat the above process until the loss function percentage error converges to less than 10% or reaches the predetermined number of training rounds of 1000, and output the adjustable parameters at the same time.

[0019] Furthermore, the step S5 includes the following steps:

[0020] S51, setting the parameters of the image information fusion model as temporary adjustable parameters, inputting 100% of the fused data into the image information fusion model, and the model outputting the recommended dosage of the dosing device;

[0021] S52, the control module controls the water treatment dosing equipment in the flocculation zone of the sedimentation tank during the water treatment dosing process according to the dosing amount recommended by the model;

[0022] S53, using the data acquisition device to monitor the water quality of the water sample collected through the real-time effluent sampling tube, obtain the effluent water quality, and obtain a effluent water quality monitoring data set at time T;

[0023] S54. Determine whether the water quality index meets the standard. If the water quality index meets the standard, obtain a water supply and discharge instruction to allow it to be used as water supply.

[0024] Beneficial effects of the present invention:

[0025] 1. It solves the problem in the existing technology that dosing control relies solely on manual control and ignores the guiding role of image information on dosing status changes. The dosing amount can accurately meet the needs of the water treatment process, improving the efficiency and reliability of water treatment;

[0026] 2. The fusion model can use image information to accurately adjust the dosage of flocculation reagents based on real-time and historical data, and conduct real-time monitoring and prediction of water quality through big data analysis, thereby improving the accuracy and efficiency of water treatment and reducing operating costs and energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a schematic flow diagram of the method of the present invention;

[0028] Figure 2 Schematic diagram of monitoring sites of the method of the present invention;

[0029] Figure 3 It is a structural diagram of the control module of the method of the present invention. DETAILED DESCRIPTION

[0030] The technical solutions in the embodiments of the present invention will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present invention.

[0031] Example 1

[0032] An intelligent dosing control method for a flocculation process based on image information comprises the following steps:

[0033] S100, monitor water quality, equipment working condition, dosage, and flocculation zone video information using data acquisition devices

[0034] SA100, monitor water quality indicators and equipment operating conditions based on data acquisition devices

[0035] SA1001. Monitor the water quality of water samples collected through the inlet sampling pipe using a data acquisition device to obtain an inlet water quality monitoring data set.

[0036] SA10011. Obtain the historical influent sampling dataset of the target water treatment plant;

[0037] SA10012. Perform data preprocessing, fill missing values, and delete outliers based on the historical water sampling dataset.

[0038] SA10013. Configure parameters of water quality monitoring instrument according to the fluctuation characteristics of water quality indicators.

[0039] SA1002. Monitor the water quality of water samples collected through the sampling tubes in the mixing zone, flocculation zone, sedimentation zone, filter tank, disinfection tank, deep treatment zone, and post-treatment zones of the sedimentation tank using the data acquisition device to obtain a process water quality monitoring data set.

[0040] SA10021. Obtain the historical sampling data sets of the target water treatment plant's process sedimentation tank mixing zone, sedimentation tank flocculation zone, sedimentation tank settling zone, filter tank, disinfection tank, deep treatment, and post-treatment;

[0041] SA10022. Sample the data set according to the past process segment, perform data preprocessing, fill in missing values, and delete outliers;

[0042] SA10023. Configure parameters of water quality monitoring instrument according to the fluctuation characteristics of water quality indicators.

[0043] SA1003. Monitor the water quality of water samples collected through the effluent sampling pipe using a data acquisition device to obtain an effluent water quality monitoring data set; wherein the water quality data acquisition device includes at least three water quality monitors at each location to obtain an average value to reduce errors;

[0044] SA10031. Obtain the historical effluent sampling dataset of the target water treatment plant;

[0045] SA10032. Perform data preprocessing, fill missing values, and delete outliers based on the past water discharge sampling data set;

[0046] SA10033. Configure parameters of water quality monitoring instrument according to the fluctuation characteristics of water quality indicators.

[0047] SA1004. Acquire equipment condition monitoring data sets by collecting equipment data from the water treatment equipment. The monitoring time for the inlet water quality monitoring data set, the process water quality monitoring data set, and the equipment condition monitoring data set is time T, and the monitoring time for the outlet water quality monitoring data set is time T-1.

[0048] SA10041. Obtain the equipment operating condition dataset of the target water treatment plant;

[0049] SA10042. Preprocess the equipment operating condition data set, fill in missing values, and delete outliers.

[0050] Specifically, in the present embodiments SA1001, SA1002, and SA1003, the target water treatment plant inlet water quality indicators include turbidity, chromaticity, temperature, pH value, conductivity, total dissolved solids, total hardness, chemical oxygen demand, biochemical oxygen demand, chloride, sulfate, nitrate, metal ions (such as iron, manganese, copper, lead, etc.), total coliform bacteria, and total bacteria count at the time of inlet water; the target water treatment plant process water quality indicators include turbidity, pH value, and chloride in the sedimentation tank mixing area, sedimentation tank flocculation area, sedimentation tank sedimentation area, filter tank, disinfection tank, deep treatment, and post-treatment; the target water treatment plant effluent water quality indicators include turbidity, chromaticity, pH value, residual chlorine, total dissolved solids, total hardness, fluoride, chloride, nitrate, metal ions, total coliform bacteria, total bacteria count, and disinfection by-products at the time of effluent water.

[0051] The target sewage treatment plant is a professional sewage treatment organization that adopts biochemical treatment to treat sewage. The sewage undergoes inlet regulation, outlet regulation, mixing process, flocculation process, sedimentation process, filtration process, disinfection reaction, deep treatment and post-treatment and other sewage biochemical or physicochemical treatment processes in the sewage treatment equipment of the target sewage treatment plant.

[0052] Specifically, in this embodiment SA1004, the raw water undergoes sewage biochemical or physicochemical treatment processes such as inlet regulation, outlet regulation, mixing process, flocculation process, sedimentation process, filtration process, disinfection reaction, deep treatment and post-treatment in the water treatment equipment of the target sewage treatment plant. The dosing time, mixing reaction time, flocculation time, sedimentation time, filtration time and disinfection contact time are adjusted based on the control system to achieve the reasonable arrangement of the mixing conditions, flocculation conditions, sedimentation conditions, filtration conditions and disinfection conditions of the water treatment equipment and the normal progress of the deep treatment, so that the water supply can effectively undergo treatment processes such as inlet regulation, outlet regulation, mixing process, flocculation process, sedimentation process, filtration process, disinfection reaction, deep treatment and post-treatment.

[0053] Data collection is performed on the control system used to control water treatment equipment for water treatment, and data collection is performed on the actual operation data of water treatment equipment to obtain equipment status monitoring data sets. The data types in the equipment condition monitoring data sets include dosing time, mixing reaction time, flocculation time, sedimentation time, filtration time and disinfection contact time. Data collection is performed on the dosing control system used to control water treatment equipment for water treatment,

[0054] SB100, monitor the dosage of the flocculation area of ​​the sedimentation tank according to the data acquisition device

[0055] Data collection is performed on the dosage data of the water treatment equipment to obtain a dosage monitoring data set, wherein the data types in the dosage monitoring data set include the dosage of the reagents of the water treatment equipment in the flocculation zone of the sedimentation tank.

[0056] SC100, based on the data acquisition device, monitors the surface image information of the flocculation area of ​​the sedimentation tank.

[0057] At least three high-resolution cameras were used to capture the characteristics of flocs on the surface of the flocculation zone of the sedimentation tank from multiple angles. The number, morphology, relative size, and existence time of flocs on the surface of the flocculation zone of the sedimentation tank were extracted from the aerobic tank video state monitoring dataset using a general target monitoring algorithm to construct an image dataset of the flocculation zone surface state.

[0058] The required image resolution and coverage were determined based on the surface conditions of the flocculation area in the sedimentation tanks of the water treatment plant. Potential influencing factors were also examined to ensure that the selected cameras were waterproof, dustproof, and corrosion-resistant to withstand the complex environmental conditions of the sewage treatment plant. Camera installation locations were determined, located around the edges of the aerobic tanks and on elevated platforms, ensuring full coverage of the aerobic tank surface and avoiding blind spots. The cameras were positioned 20 cm below the edge of the aerobic tanks. The camera angle was adjusted for optimal monitoring to avoid image quality degradation due to factors such as sunlight reflection. The camera was angled at a 35-degree angle to the gravity line. A mounting bracket and fixture were installed at the camera location. The bracket was corrosion-resistant and designed to withstand long-term use in the sewage treatment plant environment. The fixtures were secure to prevent the camera from shifting due to external forces. A waterproof network cable was used for the camera power cable to ensure stable power supply, and a power protection device was installed to prevent power outages or fluctuations. A protective cover was installed around the camera to protect it from environmental influences such as rain, dust, and chemical corrosion.

[0059] The camera system was debugged and optimized through preliminary experiments, adjusting the camera's angle and focal length to ensure clear surveillance footage and comprehensive coverage. Next, the camera was connected to the surveillance system to test image transmission and storage to ensure proper system operation. The stability and reliability of the system were verified through features such as real-time monitoring and historical playback. Finally, a regular maintenance cycle was established, set to 24 hours. The camera lens was regularly cleaned, the equipment's operating status was checked, and any issues that arose were promptly addressed to ensure the long-term stability of the surveillance system.

[0060] The schematic diagram of monitoring sites at the water supply plant is as follows Figure 2 shown.

[0061] S200, fusion of multi-source monitoring data sets into water quality condition dosing fusion data sets and image dosing fusion data sets

[0062] The data fusion method aligns the data rows and columns of the data set, and the data interval is set to the minimum sampling time.

[0063] SA200, merges the inlet water quality monitoring dataset, process water quality monitoring dataset, outlet water quality monitoring dataset, equipment operating condition monitoring dataset and dosage monitoring dataset into a water quality, operating condition and dosage fusion dataset;

[0064] SB200, fused the flocculation area surface state image dataset and the dosage monitoring dataset into an image dosage fusion dataset;

[0065] S300, build an adaptive model, perform model training and loss optimization

[0066] This includes the SA300 K-nearest neighbor regression sub-model and the SB300 target detection sub-model.

[0067] SA3001. Build the adjustable parameter K nearest neighbor regression sub-model, wherein the adjustable parameter adaptive model parameters include the number of nearest neighbors and the distance metric power;

[0068] SA3002: Using 25% of the water quality and drug administration fusion dataset as input variables, train a K-nearest neighbor regression sub-model and obtain temporary adjustable parameters. This includes inputting the water quality and drug administration fusion dataset into the constructed K-nearest neighbor regression sub-model.

[0069] The K-nearest neighbor regression model calculates the distance between the sample to be predicted and all training samples, selects the K nearest neighbors, and then makes a prediction based on the target values ​​of these neighbors. For each point to be predicted, the K-nearest neighbor regression model calculates the distance to each training sample and selects the K nearest neighbors. The target values ​​of these neighbors are then averaged to obtain the final predicted value.

[0070] The method for constructing the K-nearest neighbor regression submodel includes the following steps: First, prepare and preprocess 25% of the dataset, dividing it into a training set, a validation set, and a test set. The training data is stored. When new data needs to be predicted, the algorithm calculates the distance between the new data point and all stored training data points. Then, the K nearest neighbors are selected and regression fitting is performed based on the target values ​​of these neighbors. This process is repeated until the loss function converges to a percentage error of less than 10% or the predetermined number of training rounds of 1000 is reached, and the adjustable parameters are output.

[0071] The mathematical model of the constructed K-nearest neighbor regression sub-model is:

[0072]

[0073]

[0074] x * is the new input feature vector;

[0075] x i is the feature vector of the i-th training sample;

[0076] x *，j is x * The jth feature of

[0077] x i，j is x i The jth feature of

[0078] p is the distance metric power

[0079] is the predicted output value for the new input;

[0080] k is the number of nearest neighbors, which is also d(x *， x i )The smallest k values, corresponding to the index {i1, i2, i3, ~, i k}.

[0081] SB3001. Build an adjustable parameter target detection sub-model, where the adjustable parameter adaptive model consists of (1) convolutional layer; (2) activation function; (3) pooling layer; (4) fully connected layer.

[0082] SB3002 uses 25% of the image drug fusion dataset as input variables to train a K-nearest neighbor regression sub-model and obtain temporary adjustable parameters. The main structure of the object detection sub-model includes 13 convolutional layers, 5 pooling layers, 3 fully connected layers, and activation functions. Each convolutional layer extracts features from the input image by applying a convolution kernel. An activation function is applied after each convolution operation to introduce nonlinearity and enhance the model's expressive power. The number of convolution kernels increases with each layer, capturing more complex features. Each pooling layer reduces the size of the feature map, reducing computational complexity and preventing overfitting. The final fully connected layer combines and classifies the extracted features, outputting the predicted image label. This hierarchical structure and activation function enable efficient image feature extraction and classification. The layer structure of the object detection sub-model is as follows: first, two convolutional layers, each followed by an activation function, followed by a max pooling layer. Next, three convolutional layers, each followed by an activation function, followed by a max pooling layer. Next come three convolutional layers, each followed by an activation function, followed by a max pooling layer. Next come three more convolutional layers, each followed by an activation function, followed by a max pooling layer. Finally, three more convolutional layers, each followed by an activation function, followed by a max pooling layer. Following the convolutional and pooling layers are three fully connected layers, each followed by an activation function, and the final fully connected layer followed by an activation function.

[0083] The object detection submodel is constructed using the following methods: First, 25% of the dataset is prepared and preprocessed, splitting it into training, validation, and test sets. The network weights are then initialized, and the loss function and optimizer are selected. During training, the output and loss values ​​are calculated using forward propagation, and the weights are updated using backpropagation. Training is performed over multiple training cycles using multiple batches of data. After each cycle, model performance is evaluated using the validation set and hyperparameters are adjusted. This process is repeated until the loss function converges to a percentage error of less than 10% or until the predetermined number of training cycles, 1000, is reached. The tunable parameters are then output.

[0084] The mathematical model of the target detection sub-model constructed is:

[0085] (1) Convolutional layer The formula for convolution operation is as follows:

[0086]

[0087] in:

[0088] x, input feature map, each element represents the pixel value at a specific location;

[0089] y, output feature map, each element represents the pixel value after convolution;

[0090] w, convolution kernel weight, represents the weight of the filter;

[0091] b. Bias, a constant added to the convolution result;

[0092] i, j, position index in the feature map;

[0093] k, channel index of the output feature map;

[0094] m, n, the position index in the convolution kernel;

[0095] c. Channel index of the input feature map.

[0096] (2) The activation function formula is as follows:

[0097] f(x)=max(0,x)

[0098] in:

[0099] x, the output value of the convolutional layer.

[0100] (3) The pooling layer formula is as follows:

[0101] y i，j，k =max(x i+m，j+n，k ), where (m, n)∈P

[0102] in:

[0103] x, input feature map;

[0104] y, output feature map;

[0105] P, the size of the pooling window (e.g., 2x2).

[0106] (4) The formula of the fully connected layer is as follows:

[0107]

[0108] in:

[0109] x, input feature vector;

[0110] y, output vector;

[0111] w, weight matrix;

[0112] b. Bias;

[0113] i, j, vector index.

[0114] S400, the K nearest neighbor regression sub-model and the target detection sub-model are integrated into a graph information fusion model, and the weight coefficients after the percentage error normalization are integrated. The adjustable parameter adaptive model is embedded in the control system terminal of the water treatment dosing equipment.

[0115] The module of intelligent dosing control system for flocculation process based on graph information is as follows: Figure 3 shown.

[0116] The mathematical model of the image information fusion model is:

[0117] 1. Calculate RMSE separately

[0118]

[0119]

[0120] 2. Calculate weights

[0121]

[0122]

[0123] 3. Calculate normalized weights

[0124]

[0125]

[0126] 4. Calculate the weighted fusion value

[0127] y i =w1yi1 +w2y i2

[0128] in:

[0129] x i1 is the input value of the K nearest neighbor regression sub-model;

[0130] y i1 is the output value of the K nearest neighbor regression sub-model;

[0131] x i1 Is the input value of the target detection sub-model;

[0132] y i1 is the output value of the target detection sub-model;

[0133] p i is the true value corresponding to the input value;

[0134] RMSEANN(p i ,y i1 ) is the root mean square error of the K nearest neighbor regression submodel;

[0135] RMSETD(p i ,y i2 ) is the root mean square error of the target detection sub-model;

[0136] is the weight of the K nearest neighbor regression sub-model;

[0137] is the weight of the target detection sub-model;

[0138] w1 is the normalized weight of the K nearest neighbor regression sub-model;

[0139] w2 is the normalized weight of the target detection sub-model;

[0140] y i Weighted fusion value.

[0141] S500, input the fusion data into the image information fusion model, and control the water treatment dosing equipment in the flocculation zone of the sedimentation tank in the water treatment dosing process according to the adaptive parameters based on the image information fusion model.

[0142] S510, setting the parameters of the image information fusion model as temporary adjustable parameters, inputting 100% of the fused data into the image information fusion model, and the model outputting the recommended dosage of the dosing device;

[0143] S520, the control module controls the water treatment dosing equipment in the flocculation zone of the sedimentation tank during the water treatment dosing process according to the dosing amount recommended by the model;

[0144] S530, using the data acquisition device to monitor the water quality of the water sample collected through the real-time effluent sampling pipe, obtain the effluent water quality, and obtain a effluent water quality monitoring data set at time T;

[0145] S540: Determine whether the water quality index meets the water quality index allowed by the national standard. If the water quality index meets the water quality index allowed by the national standard, obtain a water supply discharge instruction to allow it to be used as water supply.

[0146] Table 1 Comparison of operation data using manual dosing and model-controlled dosing

[0147]

[0148]

[0149] It can be seen that the use of model control, compared with manual dosing, can significantly reduce the dosage while ensuring the water output indicators, thereby reducing operating costs and energy consumption.

[0150] The present invention is not limited to the above-mentioned optimal implementation mode. Anyone can derive other forms of products under the inspiration of the present invention. However, no matter what changes are made in the shape or structure, any technical solution that is the same or similar to that of the present application falls within the scope of protection of the present invention.

Claims

1. An intelligent dosing control method for flocculation process based on image information, characterized in that: The following steps are involved: S1. Sampling influent water quality indicators to obtain an influent water quality monitoring data set, sampling process water quality indicators to obtain a process water quality monitoring data set, sampling effluent water quality indicators to obtain an effluent water quality monitoring data set, collecting equipment data on water treatment equipment to obtain an equipment operating condition monitoring data set; monitoring the dosage of the dosing device in the flocculation zone of the sedimentation tank to obtain a dosage monitoring data set; Perform video monitoring on the surface of the flocculation area of ​​the sedimentation tank to obtain an image dataset of the surface state of the flocculation area; S2. Fusion the inlet water quality monitoring dataset, process water quality monitoring dataset, outlet water quality monitoring dataset, equipment operating condition monitoring dataset, and dosage monitoring dataset into a water quality, operating condition, and dosage fusion dataset; The flocculation area surface state image dataset and the dosage monitoring dataset are fused into an image dosage fusion dataset; S3, inputting the water quality condition and dosing fusion data set into the K-nearest neighbor regression sub-model with adjustable parameters; Input the image drug delivery fusion dataset into the target detection sub-model with adjustable parameters; S4. The results of the two models are fused into an image information fusion model with adjustable parameters. The fusion is performed according to the weight coefficient after the percentage error is normalized. The obtained adjustable parameter fusion model is embedded in the control system terminal of the water treatment dosing equipment. The mathematical model of the image information fusion model is: a1. Calculate RMSE separately a2. Calculate weight a3. Calculate normalized weights a4. Calculate the weighted fusion value and i =w1y i1 +w2y i2 in: x i1 is the input value of the K nearest neighbor regression sub-model; y i1 is the output value of the K nearest neighbor regression sub-model; x i1 Is the input value of the target detection sub-model; y i1 is the output value of the target detection sub-model; p i is the true value corresponding to the input value; RMSEANN(p i ,y i1 ) is the root mean square error of the K nearest neighbor regression submodel; RMSETD(p i ,y i2 ) is the root mean square error of the target detection sub-model; is the weight of the K nearest neighbor regression sub-model; is the weight of the target detection sub-model; w1 is the normalized weight of the K nearest neighbor regression sub-model; w2 is the normalized weight of the target detection sub-model; y i Weighted fusion value; S5. Based on the obtained adjustable parameter image information fusion model, the dosing amount of the dosing equipment in the flocculation zone of the sedimentation tank in the water treatment dosing process is controlled according to the output result.

2. The method for intelligent drug dosing control in a flocculation process based on image information according to claim 1, characterized in that: In the above S1, the monitoring time of the inlet water quality monitoring data set, the process water quality monitoring data set and the equipment operating condition monitoring data set is time T, and the monitoring time of the outlet water quality monitoring data set is time T-1.

3. The method for intelligent drug dosing control of flocculation process based on image information according to claim 1, characterized in that: In the aforementioned S1, the method for acquiring the flocculation zone surface state image dataset is as follows: the characteristics of the flocs on the surface of the flocculation zone of the sedimentation tank are captured from multiple angles, and the number, morphology, relative size, and existence time of the flocs on the surface of the flocculation zone of the sedimentation tank are extracted from the aerobic tank video state monitoring dataset through a general target monitoring algorithm to construct the flocculation zone surface state image dataset.

4. The method for intelligent drug dosing control of flocculation process based on image information according to claim 1, characterized in that: In the above-mentioned S1, the data processing method is: filling missing values ​​and deleting outliers.

5. The method for intelligent drug dosing control of flocculation process based on image information according to claim 1, characterized in that: In the aforementioned S2, the data fusion method is: aligning the data rows and data columns of the data set, and setting the data interval to the minimum sampling time.

6. The method for intelligent drug dosing control of flocculation process based on image information according to claim 1, characterized in that: In the above S3, the construction method of the K nearest neighbor regression sub-model is as follows: first, prepare and preprocess 25% of the data set and divide it into a training set, a validation set and a test set; store the training data, and when new data needs to be predicted, the algorithm calculates the distance between the new data point and all stored training data points; then, select the K nearest neighbors and perform regression fitting based on the target values ​​of these neighbors; repeat the above process until the loss function percentage error converges to less than 10% or reaches the predetermined number of training rounds of 1000 rounds, and output the adjustable parameters at the same time.

7. The method for intelligent drug dosing control in a flocculation process based on image information according to claim 1, characterized in that: In S3, the object detection sub-model is constructed as follows: first, prepare and preprocess 25% of the dataset and divide it into training set, validation set and test set; then initialize the network weights, select the loss function and optimizer; During the training process, the output and loss values ​​are calculated through forward propagation, and the weights are updated through backpropagation. During multiple training cycles, multiple batches of data are used for training, and after each cycle, the validation set is used to evaluate the model performance and adjust the hyperparameters. The above process is repeated until the loss function percentage error is less than 10% and converges or the predetermined number of training rounds, 1000, is reached, and the adjustable parameters are output at the same time.

8. The method for intelligent drug dosing control in flocculation process based on image information according to claim 1, characterized in that: The step S5 comprises the following steps: S51, setting the parameters of the image information fusion model as temporary adjustable parameters, inputting 100% of the fused data into the image information fusion model, and the model outputting the recommended dosage of the dosing device; S52, the control module controls the water treatment dosing equipment in the flocculation zone of the sedimentation tank during the water treatment dosing process according to the dosing amount recommended by the model; S53, using the data acquisition device to monitor the water quality of the water sample collected through the real-time effluent sampling tube, obtain the effluent water quality, and obtain a effluent water quality monitoring data set at time T; S54. Determine whether the water quality index meets the standard. If the water quality index meets the standard, obtain a water supply and discharge instruction to allow it to be used as water supply.

Images