Sewage treatment parameter prediction method based on plant group collaborative knowledge migration
By using a plant cluster collaborative knowledge transfer method, matching source domain plant data is selected to train a pre-trained model, which solves the problem of insufficient historical data for newly built or expanded sewage treatment plants and achieves high-precision prediction of sewage treatment parameters.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-16
- Publication Date
- 2026-03-17
AI Technical Summary
In existing technologies for newly built or expanded wastewater treatment plants, the accuracy of neural network predictions is low due to insufficient historical operating data, and the lack of effective data transfer methods leads to inaccurate prediction results.
The method based on factory cluster collaborative knowledge transfer is to select source domain factories that match the target factory by establishing a similarity model, use their data to train a pre-trained model, and then train it on the target factory through transfer learning to ensure model adaptability. CNN-LSTM, GRU, Transformer or TCN network is used for parameter prediction.
This improved the accuracy and adaptability of the wastewater treatment parameter prediction model, ensuring the accuracy of the prediction results.
Smart Images
Figure CN121687302A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method for predicting wastewater treatment parameters, and more particularly to a method for predicting wastewater treatment parameters based on collaborative knowledge transfer among plant clusters. Background Technology
[0002] The development of smart water management has promoted precise and intelligent control of the wastewater treatment process. In wastewater treatment, it is necessary to predict water quality parameters, provide early warnings of operating conditions, and optimize control of wastewater treatment plants. Existing technologies generally use Long Short-Term Memory (LSTM) networks and Convolutional Neural Networks (CNN) networks to achieve water quality parameter prediction and early warning processing.
[0003] However, to obtain accurate results, the various types of neural networks mentioned above require a large amount of sample data (or historical operating data) to train the network and update its parameters. However, for newly built, expanded, or poorly monitored wastewater treatment plants, historical operating data is scarce and cannot provide sufficient data support for the neural network. Therefore, the following approach is mainly adopted: wastewater treatment plants with abundant historical operating data are used as candidate plants (also known as source plants). Then, based on whether the processes are the same and the geographical differences are less than a set value, the data of the candidate plants are transferred to the target plant and used as the target plant's data to train the neural network. However, this approach results in extremely low accuracy of the final prediction results for the target plant. This is because it ignores the influence of different wastewater treatment plants' water quality characteristics, climate factors, and actual operating conditions. Moreover, how to select source plants based on these influencing factors and apply the value of source plant data to the target plant has become a technical challenge.
[0004] Therefore, in order to solve the above-mentioned technical problems, it is urgent to propose a new technical approach. Summary of the Invention
[0005] In view of this, the purpose of this invention is to provide a wastewater treatment parameter prediction method based on collaborative knowledge transfer among plant clusters. This method considers parameters related to the water quality, environmental conditions, and operational status of both the target plant and source plants, and establishes corresponding similarity models. Then, based on the similarity values, the final source plants are selected, and the parameter prediction model is trained using data from these selected source plants as a pre-trained model. Subsequently, the pre-trained model is transferred to the target plant, where it is trained using a small amount of historical data from the target plant. This effectively ensures the adaptability of the parameter prediction model to the actual state of the target plant, thereby ensuring the final prediction accuracy of the model.
[0006] This invention provides a wastewater treatment parameter prediction method based on plant cluster collaborative knowledge transfer, comprising the following steps:
[0007] S1. Identify the source and target plants;
[0008] S2. Select source plants that match the target plant based on the state parameters of the source plant and the target plant; the state parameters of the source plant and the target plant include process type, water quality similarity state parameters, environmental similarity state parameters, and operating condition similarity state parameters;
[0009] S3. Obtain the operating parameters of the source domain plants that match the target plant, construct a parameter prediction model, and input the operating parameters of the source domain plants that match the target plant into the parameter prediction model for training to obtain a pre-trained model.
[0010] S4. Obtain the operating parameters of the target plant, input the historical operating parameters of the target plant into the pre-trained model trained in S3 for transfer learning training, and input the real-time operating parameters of the target plant into the trained prediction model to obtain the prediction results of the target plant.
[0011] Furthermore, step S2 specifically includes:
[0012] S21. Determine whether the processing technology of the target plant and the source plant is the same. If yes, proceed to the next step. If no, the current source plant and the target plant do not match.
[0013] S22. Determine the water quality similarity weight, environmental similarity weight, and operating condition similarity weight between the target plant and the source area plant;
[0014] S23. Construct a comprehensive similarity model:
[0015] Score com =S wat ×W wat +S env ×W env +S roc ×W roc ;
[0016] Among them: Score com S represents the overall similarity value. wat Indicates the water quality similarity setpoint; S env Indicates the environmental similarity setting value; S roc W represents the similarity setting value for operating conditions. wat W represents the weight of water quality similarity. env W represents the environmental similarity weight. roc Indicates the weight of similarity in operating conditions;
[0017] Select the overall similarity score. com Source domain factories with values greater than the set value are used as matching source domain factories for the target factory.
[0018] Furthermore, the water quality similarity weight W is determined using the following method.wat :
[0019] The water quality similarity parameters include chemical oxygen demand, carbon-nitrogen ratio, carbon-phosphorus ratio, total phosphorus content, and ammonia nitrogen concentration.
[0020] Calculate the similarity value of a single water quality similarity state parameter:
[0021] ;
[0022] in: This represents the similarity value of the i-th water quality similarity state parameter. Let represent the mean value of the i-th water quality similarity state parameter of the target plant. Let represent the mean value of the i-th water quality similarity state parameter of the source region plant. This represents the standard deviation of the i-th water quality similarity state parameter of the target plant. The standard deviation of the i-th water quality similarity state parameter in the source region is represented.
[0023] Determine the water quality similarity weight W wat :
[0024] Where n=5.
[0025] Furthermore, the environmental similarity state parameter W is determined using the following method. env :
[0026] The environmental similarity parameters include the longitude of the target plant and the source plant, the latitude of the target plant and the source plant, the annual average temperature of the target plant and the source plant, and the proportion of industrial wastewater in the target plant and the source plant.
[0027] Calculate the geographical similarity value between the target plant and the source plant:
[0028] ;
[0029] ;
[0030] ;
[0031] ;
[0032] This represents the longitude similarity value between the target plant and the source plant. This represents the dimensional similarity value between the target factory and the source factory. This represents the annual average temperature similarity value between the target plant and the source plant. This indicates the similarity value of the proportion of industrial wastewater between the target plant and the source plant; These represent the longitude values of the target plant and the source plant, respectively. These represent the dimension values of the target factory and the source factory, respectively; This represents the target plant's average annual temperature. This indicates the average annual temperature of the source area plant; , These represent the annual temperature standard deviations of the target plant and the source plant, respectively. This represents the average percentage of industrial wastewater discharged by the target plant. This represents the average proportion of industrial wastewater from the source area plant; , These represent the standard deviations of the proportion of industrial wastewater from the target plant and the source plant, respectively.
[0033] Determine the environmental similarity weight W env :
[0034] Where: N=4.
[0035] Furthermore, the similarity weight W of the operating conditions is determined by the following method. roc :
[0036] The similarity parameters of the operating conditions include operating load rate, sludge age, dissolved oxygen value, and sludge return ratio;
[0037] Calculate the similarity value of individual runtime state parameters:
[0038] ;
[0039] in: This represents the similarity value of the r-th running similarity state parameter. Let represent the mean value of the r-th operational similarity state parameter of the target plant. Let represent the mean of the r-th operational similarity state parameter of the source domain plant. This represents the standard deviation of the r-th operational similarity state parameter of the target plant. The standard deviation of the r-th operational similarity state parameter of the source domain plant is represented;
[0040] Determine the similarity weight W of the operating conditions roc :
[0041] Where R=4.
[0042] Furthermore, the prediction model is any one of CNN-LSTM network, LSTM network, GRU network, Transformer network, and TCN network.
[0043] The beneficial effects of this invention are as follows: By incorporating parameters related to the water quality, environmental conditions, and operational status of both the target plant and the source plant, a corresponding similarity model is established. Then, based on the similarity values, the final source plant is selected. Data from the selected source plant is used to train the parameter prediction model as a pre-trained model. Subsequently, the pre-trained model is transferred to the target plant, where it is trained using a small amount of historical data. This effectively ensures the adaptability of the parameter prediction model to the actual state of the target plant, thereby ensuring the final prediction accuracy of the model. Attached Figure Description
[0044] The present invention will be further described below with reference to the accompanying drawings and embodiments:
[0045] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation
[0046] The present invention will be further described in detail below:
[0047] This invention provides a wastewater treatment parameter prediction method based on plant cluster collaborative knowledge transfer, comprising the following steps:
[0048] S1. Identify the source and target plants;
[0049] S2. Filter source plants that match the target plant based on the state parameters of the source plant and the target plant; the state parameters of the source plant and the target plant include process type, water quality similarity state parameters, environmental similarity state parameters, and operating condition similarity state parameters;
[0050] S3. Obtain the operating parameters of the source domain factory that matches the target factory, construct a parameter prediction model, and input the operating parameters of the source domain factory that matches the target factory into the parameter prediction model for training to obtain a pre-trained model; wherein, the parameter prediction model is any one of CNN-LSTM network, LSTM network, GRU network, Transformer network, and TCN network.
[0051] S4. Obtain the operating parameters of the target factory. Input the historical operating parameters of the target factory into the pre-trained model trained in S3 for transfer learning training. Input the real-time operating parameters of the target factory into the trained prediction model to obtain the prediction results of the target factory. Here, the data from the source domain factory is used to train the parameter prediction model, and then the target factory trains the prediction model again. However, when the target factory trains the prediction model, it only trains the network parameters of the output layer of the prediction model, while the network parameters before the output layer are frozen. For example, a convolutional neural network consists of an input layer, convolutional layer, pooling layer, flattening layer, fully connected layer and softmax function. Then, the convolutional network is trained with the data from the source domain factory until the convergence condition is reached. At this time, the parameters of all networks before the fully connected layer are frozen. That is, when the training set of the target factory is input, the parameters of these networks remain unchanged. The parameters of the fully connected layer are trained and updated with the training parameters of the target factory.
[0052] The above method takes into account the water quality, environmental conditions, and operational status of both the target plant and the source plant, and establishes a corresponding similarity model. Then, the final source plant is selected based on the similarity value. The parameter prediction model is trained using the data from the selected source plant as a pre-trained model. Subsequently, the pre-trained model is transferred to the target plant and trained using a small amount of historical data from the target plant. This effectively ensures the adaptability of the parameter prediction model to the actual state of the target plant, thereby ensuring the final prediction accuracy of the prediction model.
[0053] In this embodiment, step S2 specifically includes:
[0054] S21. Determine whether the treatment processes of the target plant and the source plant are the same. If yes, proceed to the next step; otherwise, the current source plant and target plant are not a match. For treatment processes, such as the A² / O process (anaerobic-anoxic-aerobic biological nitrogen and phosphorus removal process) or the SBR process (sequencing batch activated sludge process), there are other existing process types. You can directly determine whether the target plant and the source plant are a match by checking the process type. If the treatment processes are different, then there is no need for further screening.
[0055] S22. Determine the water quality similarity weight, environmental similarity weight, and operating condition similarity weight between the target plant and the source area plant;
[0056] S23. Construct a comprehensive similarity model:
[0057] Score com =S wat ×W wat +S env ×W env +Sroc ×W roc ;
[0058] Among them: Score com S represents the overall similarity value. wat Indicates the water quality similarity setpoint; S env Indicates the environmental similarity setting value; S roc W represents the similarity setting value for operating conditions. wat W represents the weight of water quality similarity. env W represents the environmental similarity weight. roc Indicates the weight of similarity in operating conditions;
[0059] Select the overall similarity score. com Source domain factories with values greater than a set value are used as matching source domain factories for the target factory. In the above, all three factors are considered, thus effectively ensuring the accuracy of the final prediction result. For S... wat S env and S roc The value can be set according to the actual operating conditions of the target factory.
[0060] Generally, a source region typically includes multiple wastewater treatment plants with abundant data. When the similarity among source region plants is greater than or equal to two (above a set value), the prediction model can be trained on all of these selected source region plants. For example, if three source region plants meet the requirements, training the prediction model on these three plants is equivalent to having three prediction models (e.g., all using convolutional neural network structures, but trained on data from three different source region plants, resulting in different model parameters and thus three independent prediction models). Then, the three prediction models are retrained using historical data from the target plant. Real-time parameters from the target plant are then input into the three prediction models, which will output three predicted values. The average of the three predicted values is then used as the final prediction result, thus achieving plant group collaboration. The prediction result can be the water quality parameters after wastewater treatment, thereby determining the quality level of wastewater treatment. Alternatively, it can be used to predict wastewater treatment process parameters and control the wastewater treatment process accordingly.
[0061] Specifically, the water quality similarity weight W is determined using the following method. wat :
[0062] The water quality similarity parameters include chemical oxygen demand, carbon-nitrogen ratio, carbon-phosphorus ratio, total phosphorus content, and ammonia nitrogen concentration.
[0063] Calculate the similarity value of a single water quality similarity state parameter:
[0064] ;
[0065] in: This represents the similarity value of the i-th water quality similarity state parameter. Let represent the mean value of the i-th water quality similarity state parameter of the target plant. Let represent the mean value of the i-th water quality similarity state parameter of the source region plant. This represents the standard deviation of the i-th water quality similarity state parameter of the target plant. The standard deviation of the i-th water quality similarity state parameter in the source region is represented.
[0066] Determine the water quality similarity weight W wat :
[0067] Where n=5, for example, when i takes values from 1 to 5, they represent the similarity values of chemical oxygen demand, carbon-nitrogen ratio, carbon-phosphorus ratio, total phosphorus content, and ammonia nitrogen concentration, respectively. Of course, they can also be sorted in other ways, such as setting the sequence number of the similarity value of ammonia nitrogen concentration to 1 and the sequence number of the similarity value of chemical oxygen demand to 5.
[0068] The environmental similarity state parameter W is determined using the following method. env :
[0069] The environmental similarity parameters include the longitude of the target plant and the source plant, the latitude of the target plant and the source plant, the annual average temperature of the target plant and the source plant, and the proportion of industrial wastewater in the target plant and the source plant.
[0070] Calculate the geographical similarity value between the target plant and the source plant:
[0071] ;
[0072] ;
[0073] ;
[0074] ;
[0075] This represents the longitude similarity value between the target plant and the source plant. This represents the dimensional similarity value between the target factory and the source factory. This represents the annual average temperature similarity value between the target plant and the source plant. This indicates the similarity value of the proportion of industrial wastewater between the target plant and the source plant; These represent the longitude values of the target plant and the source plant, respectively. These represent the dimension values of the target factory and the source factory, respectively; This represents the target plant's average annual temperature. This indicates the average annual temperature of the source area plant; , These represent the annual temperature standard deviations of the target plant and the source plant, respectively. This represents the average percentage of industrial wastewater discharged by the target plant. This represents the average proportion of industrial wastewater from the source area plant; , These represent the standard deviations of the proportion of industrial wastewater from the target plant and the source plant, respectively.
[0076] Determine the environmental similarity weight W env :
[0077] Where: N=4.
[0078] The similarity weight W of the operating conditions is determined by the following method. roc :
[0079] The similarity parameters of the operating conditions include operating load rate, sludge age, dissolved oxygen value, and sludge return ratio;
[0080] Calculate the similarity value of individual runtime state parameters:
[0081] ;
[0082] in: This represents the similarity value of the r-th running similarity state parameter. Let represent the mean value of the r-th operational similarity state parameter of the target plant. Let represent the mean of the r-th operational similarity state parameter of the source domain plant. This represents the standard deviation of the r-th operational similarity state parameter of the target plant. The standard deviation of the r-th operational similarity state parameter of the source domain plant is represented;
[0083] Determine the similarity weight W of the operating conditions roc :
[0084] Where R=4.
[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A method for predicting parameters of sewage treatment based on collaborative knowledge transfer of plants, characterized in that: The method comprises the following steps: S1. determining a source domain plant and a target plant; S2. screening out a source domain plant matched with the target plant based on source domain plant and target plant state parameters; the source domain plant and target plant state parameters comprise process type, water quality similarity state parameters, environmental similarity state parameters, and operating condition similarity state parameters; S3. obtaining operating parameters of the source domain plant matched with the target plant, constructing a parameter prediction model, inputting the operating parameters of the source domain plant matched with the target plant into the parameter prediction model for training to obtain a pre-trained model; S4. obtaining operating parameters of the target plant, inputting historical operating parameters of the target plant into the pre-trained model trained in S3 for transfer learning training, and inputting real-time operating parameters of the target plant into the trained prediction model to obtain a prediction result of the target plant.
2. The method of claim 1, wherein the method comprises: Step S2 specifically comprises: S21. judging whether the treatment process of the target plant and the source domain plant is the same, if yes, proceeding to the next step, and if no, the current source domain plant and the target plant are not matched; S22. determining water quality similarity weight, environmental similarity weight, and operating condition similarity weight of the target plant and the source domain plant; S23. constructing a comprehensive similarity model: Score com = S wat x W wat + S env x W env + S roc x W roc ; wherein: Score com represents the overall similarity value; S wat represents the water quality similarity setting value; S env represents the environmental similarity setting value; S roc represents the operating condition similarity setting value, W wat represents the water quality similarity weight; W env represents the environmental similarity weight; W roc represents the operating condition similarity weight; Selecting a comprehensive similarity value Score com Source domain plants with a value greater than the set value are matched source domain plants for the target plant.
3. The method of claim 2, wherein the method further comprises: The water quality similarity weight W is determined by the following method wat : The water quality similarity state parameters comprise chemical oxygen demand, carbon-nitrogen content ratio, carbon-phosphorus content ratio, total phosphorus content value, and ammonia nitrogen concentration; The single water quality similarity state parameter similarity value is calculated as: ； wherein: represents a similarity value of the i-th water quality similarity state parameter, represents a mean value of the i-th water quality similarity state parameter of the target plant, represents a mean value of the i-th water quality similarity state parameter of the source domain plant, represents a standard deviation of the i-th water quality similarity state parameter of the target plant, represents a standard deviation of the i-th water quality similarity state parameter of the source domain plant, Determining water quality similarity weight W wat : ; wherein n = 5.
4. The method of claim 2, wherein the method further comprises: The environmental similarity state parameter W is determined by the following method env : The environmental similarity state parameters comprise longitude of the target plant and the source domain plant, dimension of the target plant and the source domain plant, annual average temperature of the target plant and the source domain plant, and industrial wastewater proportion of the target plant and the source domain plant; The geographical similarity value of the target plant and the source domain plant is calculated as: ； ； ； ； This represents the longitude similarity value between the target plant and the source plant. This represents the dimensional similarity value between the target factory and the source factory. This represents the annual average temperature similarity value between the target plant and the source plant. This indicates the similarity value of the proportion of industrial wastewater between the target plant and the source plant; These represent the longitude values of the target plant and the source plant, respectively. These represent the dimension values of the target factory and the source factory, respectively; This represents the target plant's average annual temperature. This indicates the average annual temperature of the source area plant; , These represent the annual temperature standard deviations of the target plant and the source plant, respectively. This represents the average percentage of industrial wastewater discharged by the target plant. This represents the average proportion of industrial wastewater from the source area plant; , These represent the standard deviations of the proportion of industrial wastewater from the target plant and the source plant, respectively. Determining an environmental similarity weight W env : ; where: N = 4.
5. The method of claim 2, wherein the method further comprises: The running condition similarity weight W is determined by the following method roc : The operating condition similarity state parameters comprise operating load rate, sludge age, dissolved oxygen value, and sludge return ratio; The single operating similarity state parameter similarity value is calculated as: ； wherein: denotes a similarity value of the rth operating similarity state parameter, denotes a mean value of the rth operating similarity state parameter of the target plant, denotes a mean value of the rth operating similarity state parameter of the source domain plant, denotes a standard deviation of the rth operating similarity state parameter of the target plant, denotes a standard deviation of the rth operating similarity state parameter of the source domain plant, determining a running condition similarity weight W roc : ; where R = 4.
6. The method of claim 1, wherein the method further comprises: The prediction model is any one of CNN-LSTM network, LSTM network, GRU network, Transformer network, and TCN network.