Sewage treatment data management method and system based on artificial intelligence

Abstract

The present invention discloses an artificial intelligence-based sewage treatment data management method and system, which relates to the technical field. The method and system obtain real-time operation data of a sewage treatment system, extract key features of the sewage treatment system from the real-time operation data through a pre-trained multi-source data fusion network, input the key features into a pre-trained state evaluation model to generate a sewage treatment state; based on the sewage treatment state, determine whether the current operation state is in an abnormal working condition; if it is in an abnormal working condition, identify the cause of the abnormality based on the real-time operation data and the sewage treatment state through a diagnosis model based on a Bayesian network; based on the abnormal cause, input the key features into a strategy generation model based on a random forest to generate a targeted control strategy; and solve the data processing technical problem.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a sewage treatment data management method and system based on artificial intelligence. Background Art

[0002] Wastewater treatment is a crucial component of ensuring water resource security and protecting the ecological environment. In modern wastewater treatment systems, various process parameters such as pH, dissolved oxygen, chemical oxygen demand (COD), biochemical oxygen demand (BOD), and ammonia nitrogen content require real-time monitoring and control. The wastewater treatment process is complex and variable, affected by factors such as fluctuations in influent quality, ambient temperature, and microbial activity, making abnormal operating conditions prone to occur. If these anomalies are not promptly identified and appropriate measures implemented, they will lead to decreased treatment efficiency, wasted energy, and may even cause treated water to substandard and discharge, causing secondary pollution to the environment.

[0003] Traditional sewage treatment systems rely primarily on manual monitoring and regulation, resulting in delayed responses, high labor costs, and subjective judgment. With the development of IoT and sensor technologies, sewage treatment systems generate massive amounts of real-time data, which contains rich information about the system's operating status. There is an urgent need for intelligent data management methods that can automatically monitor anomalies, analyze their causes, and provide timely control strategies to ensure stable and efficient operation of sewage treatment systems.

[0004] To this end, the present invention proposes a sewage treatment data management method and system based on artificial intelligence. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a sewage treatment data management method and system based on artificial intelligence, which realizes the full-process intelligent management of sewage treatment system anomalies.

[0006] To achieve the above objectives, a sewage treatment data management method based on artificial intelligence is proposed, which includes the following steps:

[0007] Step 1: Obtain real-time operating data of the sewage treatment system;

[0008] Step 2: extracting key features of the sewage treatment system from the real-time operation data through a pre-trained multi-source data fusion network;

[0009] Step 3: Input the key features into a pre-trained state assessment model to generate a sewage treatment state; and determine whether the current operating state is in an abnormal condition based on the sewage treatment state;

[0010] Step 4: If the system is in an abnormal operating condition, the cause of the abnormality is identified based on the real-time operating data and sewage treatment status through a Bayesian network-based diagnostic model;

[0011] Step 5: Based on the cause of the abnormality, the key features are input into a random forest-based strategy generation model. The model classifies different abnormality types by integrating multiple decision trees and generates targeted control strategies based on historical processing experience and current operating parameters.

[0012] The acquisition of real-time operating data of the sewage treatment system comprises the following steps:

[0013] Step 11: Collect real-time parameter data of each process unit of the sewage treatment system through a distributed sensor network;

[0014] Step 12: The field control unit using the edge computing architecture structures the collected real-time parameter data to form structured real-time operation data;

[0015] The method of extracting key features of the sewage treatment system from the real-time operation data using a pre-trained multi-source data fusion network comprises the following steps:

[0016] Step 21: Collect historical operation data of the sewage treatment system and build a multi-source heterogeneous data training sample library;

[0017] Step 22: Construct a multi-source data fusion network model, wherein the multi-source data fusion network model adopts a multi-stream architecture design, including a feature extraction module, a time series modeling module, a multi-source fusion module, and a feature output module;

[0018] The structure of the multi-source data fusion network model includes:

[0019] In the feature extraction module, a one-dimensional convolutional neural network structure is used for the water quality parameter data stream, which contains three consecutive one-dimensional convolution blocks. Each convolution block consists of a one-dimensional convolution layer, a batch normalization layer, a ReLU activation function, and a maximum pooling layer.

[0020] For the data stream of operating parameters, a multi-scale residual network structure is adopted, which consists of 4 residual blocks. Each residual block contains two one-dimensional convolutional layers and a skip connection to capture the characteristics of device operation at different time scales.

[0021] For the data stream of process control parameters, a gated recurrent unit (GRU) network is used. It consists of a two-layer bidirectional GRU structure with 64 hidden units in each layer. The GRU network can effectively model the long-term dependencies and temporal trends of process control parameters. The bidirectional structure considers both historical and future information, improving the comprehensiveness of feature extraction.

[0022] In the time series modeling module, a temporal attention mechanism is applied to the features extracted from the three data streams to learn the importance weights of features at different time points. Specifically, the temporal attention mechanism calculates the similarity between the query vector and the features at each time point, generates attention weights, and then performs a weighted summation of the features at each time point to obtain a feature representation with temporal importance.

[0023] The multi-source fusion module employs an adaptive feature fusion strategy, consisting of three steps: feature alignment, feature interaction, and feature fusion. The feature alignment step uses a fully connected layer to map the features output by the three data streams through the time series modeling module to the same feature space. The feature interaction step uses a cross-stream attention mechanism to calculate the mutual influence between the feature spaces corresponding to different data streams. Finally, the feature fusion step uses a gated fusion unit to adaptively determine fusion weights based on the correlation between the feature spaces output by the feature interaction step for each data stream, generating a comprehensive feature representation.

[0024] In the feature output module, the comprehensive feature representation is subjected to dimensionality reduction and feature selection through a two-layer fully connected network to extract the most representative key features.

[0025] Step 23: Based on the constructed multi-source heterogeneous data training sample library, an end-to-end supervised learning method is used to train the multi-source data fusion network model and optimize the network parameters;

[0026] Step 24: Use the trained multi-source data fusion network to process the real-time operation data and extract the key features of the sewage treatment system;

[0027] Inputting the key features into a pre-trained state assessment model to generate the sewage treatment state comprises the following steps:

[0028] Step 31: Synchronously construct a training sample library for the state assessment model with a multi-source heterogeneous data training sample library; the training sample library contains two types of data: normal operating condition data and abnormal operating condition data;

[0029] Step 32: Design a state assessment model with a network structure including a feature extraction layer, a time series modeling layer, and a state classification layer;

[0030] The structure of the state assessment model includes:

[0031] In the feature extraction layer, a multilayer perceptron network is used to perform nonlinear transformation and dimensionality reduction on key input features. The feature extraction layer consists of three fully connected layers, using the ReLU activation function to enhance the state assessment model's ability to express nonlinear relationships. Dropout layers are added between the fully connected layers to prevent overfitting.

[0032] In the time series modeling layer, a bidirectional long short-term memory network is used to capture the temporal dependencies of sewage treatment system parameters. The time series modeling layer consists of two bidirectional long short-term memory networks;

[0033] In the state classification layer, the output features of the time series modeling layer are passed through two fully connected layers, and then the state classification results of the sewage treatment system are output through the Softmax function.

[0034] Step 33: Based on the constructed training sample library, a supervised learning method is used to train the state assessment model with the optimization of state classification as the goal;

[0035] Step 34: The state assessment model receives the key features from step 2 and generates a state classification result of the sewage treatment state in real time.

[0036] The method of identifying the cause of an abnormality based on the Bayesian network-based diagnostic model and real-time operation data and sewage treatment status includes the following steps:

[0037] Step 41: Based on the historical operation data of the sewage treatment system and the abnormal event records annotated by experts, a second training sample library of the Bayesian network diagnosis model including the abnormal type, cause and symptom causal relationship is constructed;

[0038] Step 42: Based on the second training sample library, a method combining structural learning and expert knowledge is used to construct a three-layer hierarchical Bayesian network structure including a root cause node layer, an intermediate state node layer, and an observation feature node layer;

[0039] The Bayesian network structure adopts a hierarchical design and includes three layers of nodes:

[0040] The first layer is the root cause node layer, which represents the root cause of abnormalities with a probability greater than the preset probability threshold, including the inlet water quality abnormality node group, equipment failure node group, and process parameter abnormality node group;

[0041] The second layer is the intermediate state node layer, which represents the system state changes caused by the root cause, including the biochemical reaction state node group, the hydraulic state node group and the solid-liquid separation state node group;

[0042] The third layer is the observation feature node layer, which represents the abnormal features that can be observed through the sensor network, including the water quality feature node group, the equipment operation feature node group and the process feature node group.

[0043] Step 43: Based on the labeled data in the second training sample library, the maximum likelihood estimation and expectation maximization algorithms are used to learn the conditional probability table of the causal relationship between each node in the Bayesian network, and the Bayesian estimation method is used to deal with the data sparsity problem;

[0044] Step 44: Based on the real-time operation data as observation evidence, the joint tree algorithm is used for probabilistic reasoning to calculate the posterior probability distribution of each root cause node and output the anomaly cause with the highest probability;

[0045] The integration of multiple decision trees to classify different abnormality types and generate targeted control strategies based on historical processing experience and current operating parameters includes the following steps:

[0046] Step 51: constructing a third training sample library containing historical abnormal events and successful handling experiences for the strategy generation model;

[0047] Step 52: Construct a strategy generation model based on random forest, including a feature preprocessing module, an anomaly classification module, and a strategy generation module;

[0048] In the feature preprocessing module, the input key feature vector and the abnormal cause are combined to form the input feature for strategy generation;

[0049] In the anomaly classification module, a random forest classifier is used to perform fine-grained anomaly type identification on the features output by the feature preprocessing module;

[0050] In the strategy generation module, a random forest regressor is used to generate specific control parameter values ​​for the fine-grained anomaly types identified by the anomaly classification module;

[0051] Step 53: Based on the third training sample library, an ensemble learning method is used to train the random forest strategy generation model;

[0052] Step 54: Input the abnormal causes and key features into the trained random forest strategy generation model to generate targeted control strategies;

[0053] This paper proposes an artificial intelligence-based sewage treatment data management system, which includes a real-time data collection module, a key feature extraction module, an abnormal condition judgment module, an abnormal cause identification module, and a strategy generation module. The modules are electrically connected to each other.

[0054] Real-time data collection module, which obtains the real-time operation data of the sewage treatment system and sends the real-time operation data to the key feature extraction module;

[0055] A key feature extraction module extracts key features of the sewage treatment system from the real-time operation data through a pre-trained multi-source data fusion network, and sends the key features to the abnormal operating condition judgment module;

[0056] The abnormal operating condition judgment module inputs the key features into a pre-trained state assessment model to generate a sewage treatment state; based on the sewage treatment state, it determines whether the current operating state is in an abnormal operating condition. If it is determined to be an abnormal operating condition, the real-time operating data and sewage treatment state are sent to the abnormal cause identification module;

[0057] The abnormality cause identification module uses a Bayesian network-based diagnostic model to identify the abnormality cause based on real-time operation data and sewage treatment status, and sends the abnormality cause and real-time operation data to the strategy generation module;

[0058] The strategy generation module inputs the key features into a random forest-based strategy generation model based on the abnormal cause. The model classifies different abnormal types by integrating multiple decision trees, and generates targeted control strategies based on historical processing experience and current operating parameters.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] The present invention uses a sensor network distributed throughout each process step of the sewage treatment process to collect multi-source data on water quality parameters in real time, generating real-time operational data for the sewage treatment system. A pre-trained multi-source data fusion network is then used to process the real-time operational data, extracting and fusing features from data of different sources and dimensions to identify key features that reflect the system's operational status. This addresses the high-dimensionality and information redundancy issues inherent in traditional methods, improving the efficiency and accuracy of subsequent analysis. The extracted key features are then fed into a pre-trained state assessment model to assess the current operational status of the sewage treatment system and determine whether it is in an abnormal operating condition. If an abnormal operating condition is determined, a Bayesian network-based diagnostic model is activated. Based on the identified abnormal cause, the system feeds the key features into a random forest-based strategy generation model. By integrating multiple decision trees, different types of abnormalities are refined and classified. Furthermore, a targeted control strategy is generated based on historical processing experience and current operating parameters. This forms a closed-loop intelligent management system, encompassing data collection, feature extraction, state assessment, abnormality diagnosis, and strategy generation, enabling full-process intelligent management of sewage treatment system abnormalities. This significantly improves the operational stability and efficiency of the sewage treatment system, while reducing energy consumption and operation and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 This is a flow chart of the sewage treatment data management method based on artificial intelligence in Example 1 of the present invention;

[0062] Figure 2 This is a module connection diagram of the artificial intelligence-based sewage treatment data management system in Example 2 of the present invention. DETAILED DESCRIPTION

[0063] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention. Example 1

[0064] like Figure 1 As shown, the sewage treatment data management method based on artificial intelligence includes the following steps:

[0065] Step 1: Obtain real-time operating data of the sewage treatment system;

[0066] Step 2: extracting key features of the sewage treatment system from the real-time operation data through a pre-trained multi-source data fusion network;

[0067] Step 3: Input the key features into a pre-trained state assessment model to generate a sewage treatment state; and determine whether the current operating state is in an abnormal condition based on the sewage treatment state;

[0068] Step 4: If the system is in an abnormal operating condition, the cause of the abnormality is identified based on the real-time operating data and sewage treatment status through a Bayesian network-based diagnostic model;

[0069] Step 5: Based on the cause of the abnormality, the key features are input into a random forest-based strategy generation model. The model classifies different abnormality types by integrating multiple decision trees and generates targeted control strategies based on historical processing experience and current operating parameters.

[0070] In an embodiment of the present invention, obtaining real-time operating data of the sewage treatment system includes the following steps:

[0071] Step 11: Collect real-time parameter data of each process unit of the sewage treatment system through a distributed sensor network;

[0072] Specifically, the distributed sensor network is composed of multiple types of sensors, including water quality parameter sensors, equipment operation status sensors, and process control parameter sensors.

[0073] Among them, water quality parameter sensors are arranged at the water inlet, each sewage treatment unit and the water outlet of the sewage treatment system to monitor key water quality indicators such as pH value, dissolved oxygen, chemical oxygen demand, ammonia nitrogen, total phosphorus, total nitrogen, and suspended solids concentration in real time; the sewage treatment unit refers to a group of sewage treatment equipment and water quality parameter monitoring equipment between the water inlet and the water outlet for real-time sewage treatment;

[0074] The equipment operation status sensor is installed on core equipment such as aeration equipment, mixing equipment, and pumping stations to monitor the equipment's operating parameters such as operating current, speed, vibration frequency, and temperature.

[0075] The process control parameter sensors monitor the hydraulic load, sludge concentration, reflow ratio, aeration volume and other process control parameters of each sewage treatment unit.

[0076] In the specific implementation of the present invention, during the data collection process, each sensor continuously collects parameter data at a preset sampling frequency. The water quality parameter sensor is sampled every 5 minutes to ensure timely capture of water quality trends; the equipment operation status sensor is sampled every 30 seconds to monitor equipment operation in real time; and the process control parameter sensor is sampled every minute to ensure real-time adjustment of process parameters.

[0077] The key water quality indicators, the operating parameters and the process control parameters collected each time constitute real-time parameter data;

[0078] Step 12: The field control unit using the edge computing architecture structures the collected real-time parameter data to form structured real-time operation data;

[0079] Specifically, a field control unit is configured for each sewage treatment unit, which is responsible for receiving data from all water quality parameter sensors, equipment operation status sensors and process control parameter sensors in the sewage treatment unit, and performing signal conversion, data verification and outlier detection.

[0080] During the signal conversion process, the analog or digital signals output by water quality parameter sensors, equipment operating status sensors, and process control parameter sensors are uniformly converted to a standard signal format. During the data verification phase, redundant sensor cross-validation is used to compare multiple sensor readings of the same parameter to identify and flag possible sensor failures or abnormal readings. Outlier detection uses a statistically based anomaly detection algorithm to identify data points that fall outside the normal range and flag them based on the degree of abnormality.

[0081] After structuring processing, the field control unit organizes the data into a structured real-time operation data packet, which contains at least the following information: timestamp, accurate to the second level; parameter identifier, which uniquely identifies each monitoring parameter; parameter value, which is the actual measured value after verification; quality label, which indicates the reliability level of the data; device identifier, which identifies the specific device or location where the data comes from.

[0082] Furthermore, the method of extracting key features of the sewage treatment system from the real-time operation data through the pre-trained multi-source data fusion network includes the following steps:

[0083] Step 21: Collect historical operation data of the sewage treatment system and build a multi-source heterogeneous data training sample library;

[0084] Specifically, in this embodiment, during the historical operation data collection process, the operation data for at least 12 consecutive months are extracted from the background data management system of the sewage treatment plant in the same manner as step one to ensure that the samples cover the treatment conditions in different seasons, different weather conditions and different influent water quality conditions, that is, the collected data include three types of heterogeneous data: water quality parameter data, equipment operation status data and process control parameter data in the historical sewage treatment process.

[0085] In an embodiment of the present invention, the multi-source heterogeneous data training sample library is organized according to time series. Each sample contains operational data within a time window. Typically, the time window length is set to 2 hours, with a sliding step size of 10 minutes. For each time window, water quality parameters, equipment operating status, and process control parameters within that time window are extracted to form a multi-source heterogeneous data sample. Simultaneously, treatment effect indicators corresponding to that time window, including effluent water quality compliance, energy consumption level, and treatment efficiency, are manually recorded as sample labels.

[0086] Step 22: Construct a multi-source data fusion network model, wherein the multi-source data fusion network model adopts a multi-stream architecture design, including a feature extraction module, a time series modeling module, a multi-source fusion module, and a feature output module;

[0087] Specifically, the multi-source data fusion network model adopts the overall architecture of "multi-stream-fusion-output". According to the characteristics of the three types of heterogeneous data in the sewage treatment system, namely water quality parameters, operation parameters and process control parameters, three parallel feature extraction streams are constructed to process different types of data respectively. The features extracted from each stream are then integrated through the fusion module to finally output the key features of the sewage treatment system.

[0088] In the feature extraction module, a one-dimensional convolutional neural network architecture consisting of three consecutive one-dimensional convolutional blocks is used for the water quality parameter data stream. Each convolutional block consists of a one-dimensional convolutional layer with a kernel size of 5, a batch normalization layer, a ReLU activation function, and a max pooling layer. The first convolutional block has 32 output channels, and this number doubles with each subsequent block. This architecture effectively captures spatial correlations and short-term variation patterns among water quality parameters.

[0089] For the operational parameter data stream, a multi-scale residual network architecture consisting of four residual blocks is employed. Each residual block consists of two one-dimensional convolutional layers and a skip connection, with kernel sizes of 3 and 5, respectively, to capture device operational characteristics at different time scales. Residual connections help alleviate the vanishing gradient problem in deep networks, improving model training stability. This architecture is well-suited for handling high-frequency fluctuations and sudden changes in device operational parameter data.

[0090] For the data stream of process control parameters, a gated recurrent unit (GRU) network is used. It consists of a two-layer bidirectional GRU structure with 64 hidden units in each layer. The GRU network can effectively model the long-term dependencies and temporal trends of process control parameters. The bidirectional structure considers both historical and future information, improving the comprehensiveness of feature extraction.

[0091] In the time series modeling module, a temporal attention mechanism is applied to the features extracted from the three data streams to learn the importance weights of features at different time points. Specifically, the temporal attention mechanism calculates the similarity between the query vector and the features at each time point to generate attention weights. It then performs a weighted summation of the features at each time point to obtain a feature representation that reflects temporal importance. This mechanism enables the model to adaptively focus on critical time points in the wastewater treatment process, such as sudden changes in water quality, equipment startup and shutdown, or process parameter adjustments. A time point feature represents the data state at a specific moment in the wastewater treatment system's operation. Specifically, for the water quality parameter data stream, the time point feature represents the numerical state of water quality indicators such as pH, dissolved oxygen, and COD at that moment; for the equipment operating status data stream, the time point feature represents the state of operating parameters such as equipment current, speed, and vibration frequency at that moment; and for the process control parameter data stream, the time point feature represents the state of process parameters such as hydraulic load, sludge concentration, and recirculation ratio at that moment.

[0092] In the multi-source fusion module, an adaptive feature fusion strategy is employed, comprising three steps: feature alignment, feature interaction, and feature fusion. The feature alignment step uses a fully connected layer to map the features output by the three data streams through the time series modeling module to the same feature space. The feature interaction step uses a cross-stream attention mechanism to calculate the mutual influence between the feature spaces corresponding to different data streams, such as the impact of equipment operating status on water quality parameters or the impact of process control parameters on equipment operation. The feature fusion step uses a gated fusion unit to adaptively determine fusion weights based on the correlation between the feature spaces output by the feature interaction step for each data stream, generating a comprehensive feature representation.

[0093] In the feature output module, a two-layer fully connected network performs dimensionality reduction and feature selection on the comprehensive feature representation to extract the most representative key features. The output dimension of the first fully connected layer is 128, using the ReLU activation function; the output dimension of the second fully connected layer is 64, corresponding to the final key feature dimension.

[0094] Step 23: Based on the constructed multi-source heterogeneous data training sample library, an end-to-end supervised learning method is used to train the multi-source data fusion network model and optimize the network parameters;

[0095] Specifically, the training of the multi-source data fusion network model takes sewage treatment effect as the optimization goal. The training process is divided into the following steps:

[0096] First, the multi-source heterogeneous data training sample library is divided into training, validation, and test sets in an 8:1:1 ratio. The training set is used to learn model parameters, the validation set is used for model selection and hyperparameter adjustment, and the test set is used to evaluate the performance of the final model.

[0097] In terms of loss function design, a multi-task learning framework is employed to simultaneously optimize multiple objectives. The main loss functions include: effluent quality prediction loss, which uses mean squared error (MSE) to calculate the difference between the predicted and actual values; treatment efficiency prediction loss, also calculated using MSE; and abnormal condition detection loss, which uses binary cross entropy to calculate the difference between the predicted abnormal probability and the actual label. These three loss functions are combined using a weighted summation, with weights set based on the importance of each task.

[0098] The multi-source data fusion network model was optimized using the Adam optimizer. The initial learning rate was set to 0.001, and a learning rate decay strategy was used, multiplying the learning rate by 0.9 every 10 training epochs. The batch size was set to 64, and the number of training epochs was 100. During training, an early stopping strategy was used. Training was stopped when the loss on the validation set showed no improvement for five consecutive epochs to avoid overfitting. The model parameters with the best performance on the validation set were saved as the final model.

[0099] Step 24: Use the trained multi-source data fusion network to process the real-time operation data and extract the key features of the sewage treatment system;

[0100] Specifically, in the actual application stage, the structured real-time operation data obtained in step 1 is input into the trained multi-source data fusion network, and a 64-dimensional key feature vector is generated through the feature extraction module, time series modeling module, multi-source fusion module and feature output module in step 22.

[0101] The key feature vector contains the core information of the current operating status of the sewage treatment system. In the embodiment of the present invention, it specifically includes: a water quality feature sub-vector (20 dimensions), which characterizes the key characteristics and changing trends of the inlet and outlet water quality; an equipment feature sub-vector (20 dimensions), which characterizes the operating status and health of the core equipment; a process feature sub-vector (20 dimensions), which characterizes the configuration of process parameters and the control effect; and a system overall feature sub-vector (4 dimensions), which characterizes the overall operating efficiency and stability of the sewage treatment system.

[0102] It should be noted that the multi-source data fusion network utilizes a multi-stream parallel feature extraction architecture to address the multi-source and heterogeneous nature of data in sewage treatment systems. Each data stream utilizes a neural network structure tailored to the characteristics of that data type, such as a one-dimensional convolutional network for water quality parameters, a residual network for equipment status, and a GRU network for process parameters. This enables efficient feature extraction for diverse data types. Secondly, the multi-source data fusion network incorporates temporal and cross-stream attention mechanisms to adaptively focus on key moments in the time series and the interactions between different data sources, thereby capturing the dynamic changes in sewage treatment processes. Thirdly, an adaptive feature fusion strategy is employed, dynamically adjusting the fusion weights of different data sources through a gated fusion unit. This enables the multi-source data fusion network to flexibly integrate multi-source information based on the current operating conditions, improving the relevance and accuracy of feature extraction. The multi-source data fusion network is trained using a multi-task learning framework, simultaneously optimizing multiple objectives such as effluent quality prediction, treatment efficiency prediction, and abnormal condition detection. This results in extracted key features with enhanced comprehensive representation capabilities, enabling subsequent tasks such as status assessment, anomaly diagnosis, and strategy generation.

[0103] Furthermore, the step of inputting the key features into a pre-trained state assessment model to generate the sewage treatment state includes the following steps:

[0104] Step 31: Synchronously construct a training sample library for the state assessment model with a multi-source heterogeneous data training sample library; the training sample library contains two types of data: normal operating condition data and abnormal operating condition data;

[0105] Specifically, in a preferred embodiment, the process of constructing the training sample library of the state assessment model is performed simultaneously with the collection of the multi-source heterogeneous data training sample library of the multi-source data fusion network in step 2.

[0106] During the normal operating data collection phase, a basic sample set is selected from several consecutive days of operational data collected during the stable operation of the sewage treatment system. Each basic sample record includes a timestamp, water quality parameters for each sewage treatment unit, equipment operating status parameters, process control parameters, and corresponding treatment effect indicators (such as whether the effluent water quality meets the standard).

[0107] During the abnormal operating condition data collection phase, marked abnormal event data is extracted from historical operating data as abnormal operating condition samples. The abnormal operating condition samples are marked with different abnormality levels, such as "normal operation" and "abnormal", to record all parameter changes during the system response process.

[0108] In a preferred embodiment, each abnormal operating condition sample can also be annotated by domain experts to determine the degree of abnormality and the scope of impact; abnormal conditions are divided into sudden changes in influent water quality, equipment failure, process parameter deviation, etc.; the scope of impact marks the affected process units and possible chain reactions.

[0109] Step 32: Design a state assessment model with a network structure including a feature extraction layer, a time series modeling layer, and a state classification layer;

[0110] Specifically, in the feature extraction layer, a multilayer perceptron network is used to perform nonlinear transformation and dimensionality reduction on key input features. The feature extraction layer consists of three fully connected layers, each containing 128, 64, and 32 neurons, respectively. The ReLU activation function enhances the state assessment model's ability to express nonlinear relationships. Dropout layers (with a dropout rate of 0.3) are added between the fully connected layers to prevent overfitting.

[0111] In the time series modeling layer, a bidirectional long short-term memory (Bi-LSTM) network is used to capture the temporal dependencies of wastewater treatment system parameters. This layer comprises two bidirectional LSTM networks, each containing 64 LSTM units. These networks simultaneously consider information from past and future time points to provide a more comprehensive understanding of parameter trends. The output of the Bi-LSTM layer is processed through an attention mechanism, adaptively assigning weights to features at different time points and highlighting state changes at key moments.

[0112] In the state classification layer, the output features of the time series modeling layer are passed through two fully connected layers (containing 32 and 16 neurons, respectively), and then a softmax function is used to output the sewage treatment system state classification result. In this embodiment, the state classification results include two categories: "normal operation" and "abnormal".

[0113] Step 33: Based on the constructed training sample library, a supervised learning method is used to train the state assessment model with the optimization of state classification as the goal;

[0114] Specifically, first, the training sample library is divided into a training set and a validation set in a ratio of 8:2.

[0115] During model training, we used a mini-batch gradient descent algorithm with a batch size of 64 and an initial learning rate of 0.001. We also used a learning rate decay strategy, reducing the learning rate by a factor of 0.9 every 50 epochs. We also used an early stopping strategy during training, stopping training if performance on the validation set did not improve for 10 consecutive epochs to prevent overfitting.

[0116] The loss function of the model is the state classification loss, which uses the cross entropy loss function to measure the difference between the predicted state and the actual state;

[0117] Step 34: The state assessment model receives the key features from step 2 and generates a state classification result of the sewage treatment state in real time.

[0118] Furthermore, the method of judging whether the current operating state is in an abnormal condition according to the sewage treatment state is:

[0119] If the sewage treatment state is indicated as "abnormal", it is determined that the current sewage treatment is in an abnormal condition;

[0120] Furthermore, the method of identifying the cause of the abnormality based on the real-time operation data and sewage treatment status by using a Bayesian network-based diagnostic model includes the following steps:

[0121] Step 41: Based on the historical operation data of the sewage treatment system and the abnormal event records annotated by experts, a second training sample library of the Bayesian network diagnosis model including the abnormal type, cause and symptom causal relationship is constructed;

[0122] Specifically, in an embodiment of the present invention, during the construction phase of the second training sample library, all confirmed abnormal event data is screened from the historical operating data of the sewage treatment system. For each abnormal operating condition sample, the following information is simultaneously supplemented: the time period in which the abnormality occurred, the real-time operating data before and after the abnormality occurred, the type and root cause of the abnormality, the propagation path of the abnormality, the scope and extent of the abnormality's impact, and the measures taken to address the abnormality and their effectiveness. It is understandable that, unlike the training sample library for the state assessment model, the second training sample library places greater emphasis on recording the causal chain of abnormal events, ensuring that the propagation patterns of abnormalities in the system can be accurately captured.

[0123] In a preferred embodiment of the present invention, abnormalities can be categorized into three main categories based on their source: abnormal influent water quality, abnormal equipment operation, and abnormal process parameters. Abnormal influent water quality includes, but is not limited to, sudden increases in influent COD, ammonia nitrogen load shock, abnormal pH values, and toxic substance shock; abnormal equipment operation includes, but is not limited to, aeration equipment failure, mixing equipment failure, return pump failure, and instrument failure; and abnormal process parameters include, but are not limited to, insufficient dissolved oxygen, abnormal sludge concentration, hydraulic load fluctuations, and improper return ratios.

[0124] A team of experts, comprised of wastewater treatment process and equipment specialists, annotated each abnormal event in the second training sample library, identifying the causal chain between the root cause, secondary causes, and symptoms. For example, aeration equipment failure (the root cause) could lead to insufficient dissolved oxygen (the secondary cause), which in turn could cause decreased nitrifying bacteria activity (the intermediate state), ultimately leading to excessive ammonia nitrogen in the effluent (the symptom). These detailed causal chain annotations provide precise prior knowledge for Bayesian network structure learning, complementing the state assessment model's training sample library, which focuses solely on abnormal states.

[0125] Step 42: Based on the second training sample library, a method combining structural learning and expert knowledge is used to construct a three-layer hierarchical Bayesian network structure including a root cause node layer, an intermediate state node layer, and an observation feature node layer;

[0126] Specifically, the Bayesian network structure adopts a hierarchical design and includes three layers of nodes:

[0127] The first layer is the root cause node layer, representing the fundamental causes of abnormalities with probabilities greater than the preset probability threshold. These include the influent water quality abnormality node group, the equipment failure node group, and the process parameter abnormality node group. The influent water quality abnormality node group includes nodes such as influent COD surge, ammonia nitrogen load shock, and pH abnormality. The equipment failure node group includes nodes such as aeration equipment failure, mixing equipment failure, and return pump failure. The process parameter abnormality node group includes nodes such as dissolved oxygen control abnormality, sludge concentration abnormality, and hydraulic load abnormality.

[0128] The second layer is the intermediate state node layer, which represents system state changes caused by fundamental causes. It includes the biochemical reaction state node group, the hydraulic state node group, and the solid-liquid separation state node group. The biochemical reaction state node group includes nodes such as nitrification inhibition, denitrification efficiency reduction, and organic matter degradation obstruction; the hydraulic state node group includes nodes such as abnormal hydraulic retention time and uneven mixing; and the solid-liquid separation state node group includes nodes such as sludge settling performance reduction and sludge floating.

[0129] The third layer is the observation feature node layer, which represents abnormal features that can be observed through the sensor network. It includes water quality feature node groups, equipment operation feature node groups, and process feature node groups. The water quality feature node group includes nodes for abnormal effluent COD, abnormal effluent ammonia nitrogen, and abnormal effluent total phosphorus; the equipment operation feature node group includes nodes for abnormal equipment current, abnormal equipment vibration, and abnormal equipment temperature; and the process feature node group includes nodes for abnormal dissolved oxygen concentration, abnormal MLSS, and abnormal SVI.

[0130] In the Bayesian network structure construction process, a method combining structural learning and expert knowledge is employed. First, the initial causal relationship structure between nodes is determined based on prior knowledge provided by domain experts. Then, a scoring-based structural learning algorithm, such as the K² algorithm or the Bayesian Information Criterion (BIC) algorithm, is used to learn and optimize the network structure from a secondary training sample database. During the structural learning process, the strong causal relationships defined by the experts are maintained, and only weak causal relationships are optimized and adjusted.

[0131] Step 43: Based on the labeled data in the second training sample library, the maximum likelihood estimation and expectation maximization algorithms are used to learn the conditional probability table of the causal relationship between each node in the Bayesian network, and the Bayesian estimation method is used to deal with the data sparsity problem;

[0132] Specifically, after determining the Bayesian network structure, based on the labeled data in the second training sample library, it is necessary to learn the conditional probability table of each node, that is, the probability distribution of the current node in each possible state given the parent node state.

[0133] For discrete nodes, conditional probabilities are calculated from the second training sample library using maximum likelihood estimation. For example, the conditional probability table from the "Aeration Equipment Failure" node (with a value of "Yes" or "No") to the "Dissolved Oxygen Insufficient" node (with a value of "Yes" or "No") is determined by counting the proportion of dissolved oxygen insufficiency cases in the second training sample library when the aeration equipment fails and the proportion of dissolved oxygen insufficiency cases when the aeration equipment is functioning properly.

[0134] For conditional probabilities with sparse data, we use Bayesian estimation and introduce a prior distribution to smooth the probability estimate to avoid the zero probability problem. In the specific implementation, a small pseudo-count (such as 0.5) is added to each conditional probability to ensure that even state combinations not observed in the training data have non-zero probability.

[0135] For discrete child nodes with continuous parent nodes, soft discretization is used, mapping the continuous value to the probability of a discrete state using the sigmoid function. For example, to map "Dissolved Oxygen Concentration" (a continuous value) to "Nitrification Inhibition" (a discrete state), the function P(Nitrification Inhibition = "Yes" | Dissolved Oxygen = x) = 1 / (1+exp(α(x-β))) is used, where α and β are parameters learned from the training data.

[0136] In a preferred embodiment of the present invention, conditional probability learning employs the expectation-maximization algorithm, which can handle the problem of missing data in training samples. The EM algorithm iteratively estimates the distribution of missing data in the E step and maximizes the likelihood function that includes the estimated missing data in the M step until convergence.

[0137] Step 44: Based on the real-time operation data as observation evidence, the joint tree algorithm is used for probabilistic reasoning to calculate the posterior probability distribution of each root cause node and output the anomaly cause with the highest probability;

[0138] Specifically, in the actual application stage, when step three determines that the current operating state is in an abnormal condition, the diagnostic model based on the Bayesian network trained on the second training sample library receives the real-time operating data obtained in step one as input information.

[0139] First, map the real-time operating data into evidence at the observation feature node level in the Bayesian network. For example, if the real-time operating data shows that the effluent ammonia nitrogen concentration is 15 mg / L, exceeding the standard limit of 10 mg / L, then set the "Outlet Ammonia Nitrogen Abnormal" node as "Yes" evidence; if the dissolved oxygen concentration is 0.8 mg / L, below the normal range of 2-4 mg / L, then set the "Dissolved Oxygen Concentration Abnormal" node as "Yes" evidence.

[0140] Then, based on the provided evidence, the posterior probability distribution of each node in the root cause node layer is calculated using the Bayesian network inference algorithm. In an embodiment of the present invention, a joint tree algorithm is used for precise inference, and probability propagation calculation is implemented by converting the Bayesian network into a joint tree structure.

[0141] Finally, based on the posterior probabilities of each node in the root cause layer, the most probable anomaly cause is identified and output. Specifically, the root cause node with the highest posterior probability is selected as the primary anomaly cause. Other root cause nodes with posterior probabilities exceeding a preset threshold, such as 0.3, are considered secondary anomaly causes. These anomaly causes and their corresponding posterior probabilities are directly output, with the one with the highest posterior probability being the primary anomaly cause.

[0142] It should be noted that the diagnostic model based on the Bayesian network has the following advantages: first, the graph structure of the Bayesian network intuitively expresses the causal relationship between the components and parameters in the sewage treatment system, making the diagnostic results interpretable; second, the Bayesian network can process incomplete observation data and can still perform effective reasoning even if some sensor data is missing; third, through the probabilistic reasoning mechanism, the Bayesian network can quantify uncertainty and provide a confidence assessment for each possible cause of the abnormality, avoiding the misjudgment that may be caused by deterministic diagnostic methods; finally, the Bayesian network model can continuously update the conditional probability table as new data accumulates, thereby achieving continuous optimization of diagnostic capabilities.

[0143] Furthermore, the integration of multiple decision trees to classify different abnormality types and generate targeted control strategies based on historical processing experience and current operating parameters includes the following steps:

[0144] Step 51: constructing a third training sample library containing historical abnormal events and successful handling experiences for the strategy generation model;

[0145] Specifically, during the third training sample library construction phase, all successfully handled abnormal event data was screened from the historical operational data of the sewage treatment system. For each abnormal event, the following information was simultaneously supplemented: the key characteristics of the abnormality, the cause of the abnormality, the control strategy adopted to address the abnormality, and an evaluation of the effectiveness of the control strategy after implementation.

[0146] In a preferred embodiment of the present invention, the control strategies can be divided into three categories based on the target of action: water quality parameter control strategies, equipment operation control strategies, and process parameter control strategies. Water quality parameter control strategies include, but are not limited to, adjusting the influent pH, increasing the carbon source dosage, and increasing the phosphorus removal agent dosage; equipment operation control strategies include, but are not limited to, adjusting the aeration equipment operating frequency, adjusting the stirring equipment operating time, and adjusting the return pump flow rate; and process parameter control strategies include, but are not limited to, adjusting the dissolved oxygen setpoint, adjusting the sludge concentration, adjusting the hydraulic retention time, and adjusting the return flow ratio.

[0147] In the third training sample library, each sample record contains the following fields: timestamp, anomaly type, anomaly cause, key feature vector when the anomaly occurs, control strategy adopted, and control effect evaluation indicators (such as the degree of improvement in effluent water quality after control, energy consumption changes, treatment efficiency improvement, etc.).

[0148] Step 52: Construct a strategy generation model based on random forest, including a feature preprocessing module, an anomaly classification module, and a strategy generation module;

[0149] Specifically, in the feature preprocessing module, the input key feature vector and anomaly cause are combined to form the input features for strategy generation. This processing involves concatenating the 64-dimensional key feature with the one-hot encoded vector of the anomaly cause to form an enhanced feature vector; applying principal component analysis (PCA) to the enhanced feature vector for dimensionality reduction; and normalizing the reduced features to ensure consistent dimension across all features.

[0150] In the anomaly classification module, a random forest classifier is used to identify fine-grained anomaly types based on the features output by the feature preprocessing module. This random forest classifier consists of 100 decision trees. Each decision tree randomly selects a subset of features and samples during training and uses voting to determine the final anomaly type. The maximum depth of the decision tree is set to 10 to balance model complexity and generalization. The output of the anomaly classification module is a fine-grained anomaly classification, such as "insufficient nitrification due to high ammonia nitrogen loading" or "insufficient dissolved oxygen due to aeration equipment failure."

[0151] In the strategy generation module, a random forest regressor is used to generate specific control parameter values ​​for the fine-grained anomaly types identified by the anomaly classification module. This random forest regressor consists of 100 regression decision trees, each of which independently predicts the value of each control parameter, ultimately averaging the predicted values ​​to obtain the final parameter values. The maximum depth of the regression decision trees is set to 12 to capture the complex nonlinear relationships between parameters. The output of the strategy generation module is a set of specific control parameter values, such as "adjust the dissolved oxygen setpoint in the aeration tank to 3.5 mg / L" and "adjust the reflux ratio to 150%."

[0152] Step 53: Based on the third training sample library, an ensemble learning method is used to train the random forest strategy generation model;

[0153] Specifically, the training process of the random forest strategy generation model is divided into the following steps:

[0154] First, the third training sample library is divided into a training set, a validation set, and a test set in a ratio of 8:1:1. The training set is used to learn model parameters, the validation set is used for model selection and hyperparameter adjustment, and the test set is used to evaluate the performance of the final model.

[0155] During training of the anomaly classification module, a cross-entropy loss function is used to measure the difference between the predicted anomaly type and the actual anomaly type. During training, each decision tree is constructed using a randomly selected feature subset and a randomly selected sample subset. The feature subset size is set to the square root of the total number of features. The sample subset is obtained through sampling with replacement and has the same size as the original training set. Each decision tree is trained independently, and the predictions of all decision trees are ultimately integrated through majority voting.

[0156] During training of the strategy generation module, a mean squared error loss function is used to measure the difference between the predicted and actual control parameter values. Each regression decision tree is constructed using a random feature subset and a sample subset. The feature subset size is set to one-third of the total number of features, and the sample subset is obtained through sampling with replacement. Each regression decision tree is trained independently, and the predictions of all the regression decision trees are finally averaged.

[0157] During the model hyperparameter optimization phase, we used a grid search combined with cross-validation to optimize key hyperparameters of the random forest, including the number of decision trees, maximum depth, and minimum number of leaf node samples. For each hyperparameter combination, we evaluated model performance on the validation set, and selected the hyperparameter combination with the best performance as the final model configuration.

[0158] During the model evaluation phase, the test set was used to assess the performance of the final model. For the anomaly classification module, classification accuracy, precision, recall, and F1 score were calculated; for the strategy generation module, mean squared error, mean absolute error, and R² score were calculated. Simultaneously, wastewater treatment experts manually evaluated the control strategies generated by the model to ensure their rationality and operability.

[0159] Step 54: Input the abnormal causes and key features into the trained random forest strategy generation model to generate targeted control strategies;

[0160] Specifically, in the actual application stage, the abnormal causes output by the Bayesian network diagnosis model in step four and the key features extracted by the multi-source data fusion network in step two are input into the trained random forest strategy generation model. After processing by the feature preprocessing module, the abnormality classification module and the strategy generation module, a specific control strategy for the current abnormal working condition is generated.

[0161] The control strategy includes at least the following: a description of the anomaly type, a list of recommended parameters to be adjusted and their target values, a ranking of adjustment priorities, and the expected control effect. The anomaly type description provides an accurate description of the current anomaly; the list of recommended parameters to be adjusted includes the water quality parameters, equipment operating parameters, or process control parameters that need to be adjusted, along with their specific target values; the ranking of adjustment priorities is based on the importance and urgency of the parameters to guide the operator's control sequence; and the expected control effect predicts the improvement in system performance after the adjustment.

[0162] In a preferred embodiment of the present invention, the random forest strategy generation model generates different types of control strategies for different types of anomalies. For example, for an abnormality of effluent COD exceeding the standard due to a sudden increase in influent COD, the generated control strategies may include: increasing the aeration volume, extending the hydraulic retention time, adjusting the recirculation ratio, etc.; for an abnormality of insufficient dissolved oxygen due to aeration equipment failure, the generated control strategies may include: activating backup aeration equipment, reducing the influent flow rate, increasing the mixed liquor recirculation, etc.; for an abnormality of sludge bulking due to abnormal sludge concentration, the generated control strategies may include: adjusting the sludge discharge volume, adding flocculants, adjusting the aeration mode, etc. Example 2

[0163] like Figure 2 As shown, the artificial intelligence-based sewage treatment data management system includes a real-time data collection module, a key feature extraction module, an abnormal condition judgment module, an abnormal cause identification module, and a strategy generation module; wherein each module is electrically connected;

[0164] Real-time data collection module, which obtains the real-time operation data of the sewage treatment system and sends the real-time operation data to the key feature extraction module;

[0165] A key feature extraction module extracts key features of the sewage treatment system from the real-time operation data through a pre-trained multi-source data fusion network, and sends the key features to the abnormal operating condition judgment module;

[0166] The abnormal operating condition judgment module inputs the key features into a pre-trained state assessment model to generate a sewage treatment state; based on the sewage treatment state, it determines whether the current operating state is in an abnormal operating condition. If it is determined to be an abnormal operating condition, the real-time operating data and sewage treatment state are sent to the abnormal cause identification module;

[0167] The abnormality cause identification module uses a Bayesian network-based diagnostic model to identify the abnormality cause based on real-time operation data and sewage treatment status, and sends the abnormality cause and real-time operation data to the strategy generation module;

[0168] The strategy generation module inputs the key features into a random forest-based strategy generation model based on the abnormal cause. The model classifies different abnormal types by integrating multiple decision trees, and generates targeted control strategies based on historical processing experience and current operating parameters.

[0169] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A sewage treatment data management method based on artificial intelligence, characterized in that: The following steps are involved: Step 1: Obtain real-time operating data of the sewage treatment system; Step 2: extracting key features of the sewage treatment system from the real-time operation data through a pre-trained multi-source data fusion network; the multi-source data fusion network model includes a feature extraction module, a time series modeling module, a multi-source fusion module and a feature output module; In the feature extraction module, a one-dimensional convolutional neural network structure consisting of three consecutive one-dimensional convolutional blocks is used for the data stream of water quality parameters; a gated recurrent unit network consisting of a two-layer bidirectional GRU structure is used for the data stream of process control parameters; and a multi-scale residual network structure consisting of four residual blocks is used for the data stream of operating parameters. In the time series modeling module, the temporal attention mechanism is applied to the features extracted from the three data streams to learn the importance weights of the features at different time points. The weighted sum of the features at each time point is then performed to obtain a feature representation with temporal importance. In the multi-source fusion module, an adaptive feature fusion strategy is adopted, which includes three steps: feature alignment, feature interaction and feature fusion; The feature alignment step maps the features output by the three data streams through the time series modeling module to the same feature space through a fully connected layer; The feature interaction step calculates the mutual influence between the feature spaces corresponding to different data streams through the cross-stream attention mechanism; The feature fusion step uses a gated fusion unit to adaptively determine the fusion weights based on the correlation of the feature spaces output by the feature interaction step of each data stream feature to generate a comprehensive feature representation. In the feature output module, the comprehensive feature representation is reduced in dimension and feature selection is performed according to the fusion weights through a two-layer fully connected network to extract key features; Step 3: Input the key features into a pre-trained state assessment model to generate a sewage treatment state; and determine whether the current operating state is in an abnormal condition based on the sewage treatment state; Step 4: If the system is in an abnormal operating condition, the cause of the abnormality is identified based on the real-time operating data and sewage treatment status through a Bayesian network-based diagnostic model; Step 5: Based on the cause of the abnormality, the key features are input into a strategy generation model based on random forest. The model classifies different abnormality types by integrating multiple decision trees, and generates targeted control strategies based on historical processing experience and current operating parameters.

2. The artificial intelligence-based sewage treatment data management method according to claim 1, characterized in that: The acquisition of real-time operating data of the sewage treatment system comprises the following steps: Step 11: Collect real-time parameter data of each process unit of the sewage treatment system through a distributed sensor network; Step 12: The field control unit using the edge computing architecture structures the collected real-time parameter data to form structured real-time operation data.

3. The artificial intelligence-based sewage treatment data management method according to claim 2, characterized in that: The method of extracting key features of the sewage treatment system from the real-time operation data using a pre-trained multi-source data fusion network comprises the following steps: Step 21: Collect historical operation data of the sewage treatment system and build a multi-source heterogeneous data training sample library; Step 22: Construct a multi-source data fusion network model; Step 23: Based on the constructed multi-source heterogeneous data training sample library, an end-to-end supervised learning method is used to train the multi-source data fusion network model and optimize the network parameters; Step 24: Use the trained multi-source data fusion network to process the real-time operation data and extract the key features of the sewage treatment system.

4. The artificial intelligence-based sewage treatment data management method according to claim 3, characterized in that: In the structure of the multi-source data fusion network model: Each convolution block consists of a one-dimensional convolution layer, a batch normalization layer, a ReLU activation function, and a maximum pooling layer; Each residual block contains two one-dimensional convolutional layers and a skip connection to capture device operation characteristics at different time scales; The GRU network can effectively model the long-term dependencies and temporal trends of process control parameters, while the bidirectional structure considers both historical and future information, improving the comprehensiveness of feature extraction. The temporal attention mechanism generates attention weights by calculating the similarity between the query vector and the features at each time point.

5. The artificial intelligence-based sewage treatment data management method according to claim 4, characterized in that: Inputting the key features into a pre-trained state assessment model to generate the sewage treatment state comprises the following steps: Step 31: Synchronously construct a training sample library for the state assessment model with a multi-source heterogeneous data training sample library; the training sample library contains two types of data: normal operating condition data and abnormal operating condition data; Step 32: Design a state assessment model with a network structure including a feature extraction layer, a time series modeling layer, and a state classification layer; Step 33: Based on the constructed training sample library, a supervised learning method is used to train the state assessment model with the optimization of state classification as the goal; Step 34: The state assessment model receives the key features from step 2 and generates a state classification result of the sewage treatment state in real time.

6. The artificial intelligence-based sewage treatment data management method according to claim 5, characterized in that: The structure of the state assessment model includes: In the feature extraction layer, a multi-layer perceptron network is used to perform nonlinear transformation and dimensionality reduction on the key features of the input. The feature extraction layer contains three fully connected layers, and the ReLU function is used as the activation function to enhance the state assessment model's ability to express nonlinear relationships. Dropout layers are added between the fully connected layers to prevent model overfitting. In the time series modeling layer, a bidirectional long short-term memory network is used to capture the temporal dependencies of sewage treatment system parameters; the time series modeling layer contains two bidirectional long short-term memory networks; In the state classification layer, the output features of the time series modeling layer are passed through two fully connected layers, and finally the state classification results of the sewage treatment system are output through the Softmax function.

7. The artificial intelligence-based sewage treatment data management method according to claim 6, characterized in that: The method of identifying the cause of an abnormality based on the Bayesian network-based diagnostic model and real-time operation data and sewage treatment status includes the following steps: Step 41: Based on the historical operation data of the sewage treatment system and the abnormal event records annotated by experts, a second training sample library of the Bayesian network diagnosis model including the abnormal type, cause and symptom causal relationship is constructed; Step 42: Based on the second training sample library, a method combining structural learning and expert knowledge is used to construct a three-layer hierarchical Bayesian network structure including a root cause node layer, an intermediate state node layer, and an observation feature node layer; Step 43: Based on the labeled data in the second training sample library, the maximum likelihood estimation and expectation maximization algorithms are used to learn the conditional probability table of the causal relationship between each node in the Bayesian network, and the Bayesian estimation method is used to deal with the data sparsity problem; Step 44: Based on the real-time operation data as observation evidence, the joint tree algorithm is used for probabilistic reasoning to calculate the posterior probability distribution of each root cause node and output the abnormal cause with the highest probability.

8. The artificial intelligence-based sewage treatment data management method according to claim 7, characterized in that: The Bayesian network structure adopts a hierarchical design and includes three layers of nodes: The first layer is the root cause node layer, which represents the root cause of abnormalities with a probability greater than the preset probability threshold, including the inlet water quality abnormality node group, equipment failure node group, and process parameter abnormality node group; The second layer is the intermediate state node layer, which represents the system state changes caused by the root cause, including the biochemical reaction state node group, the hydraulic state node group and the solid-liquid separation state node group; The third layer is the observation feature node layer, which represents the abnormal features that can be observed through the sensor network, including the water quality feature node group, the equipment operation feature node group and the process feature node group.

9. The artificial intelligence-based sewage treatment data management method according to claim 8, characterized in that: The integration of multiple decision trees to classify different abnormality types and generate targeted control strategies based on historical processing experience and current operating parameters includes the following steps: Step 51: constructing a third training sample library containing historical abnormal events and successful handling experiences for the strategy generation model; Step 52: Construct a strategy generation model based on random forest, including a feature preprocessing module, an anomaly classification module, and a strategy generation module; In the feature preprocessing module, the input key feature vector and the abnormal cause are combined to form the input feature for strategy generation; In the anomaly classification module, a random forest classifier is used to perform fine-grained anomaly type identification on the features output by the feature preprocessing module; In the strategy generation module, a random forest regressor is used to generate specific control parameter values ​​for the fine-grained anomaly types identified by the anomaly classification module; Step 53: Based on the third training sample library, an ensemble learning method is used to train the random forest strategy generation model; Step 54: Input the abnormal causes and key features into the trained random forest strategy generation model to generate targeted control strategies.

10. A sewage treatment data management system based on artificial intelligence, which is used to implement the sewage treatment data management method based on artificial intelligence according to any one of claims 1 to 9, characterized in that: It includes a real-time data collection module, a key feature extraction module, an abnormal operating condition judgment module, an abnormal cause identification module, and a strategy generation module; wherein each module is electrically connected; Real-time data collection module, which obtains the real-time operation data of the sewage treatment system and sends the real-time operation data to the key feature extraction module; A key feature extraction module extracts key features of the sewage treatment system from the real-time operation data through a pre-trained multi-source data fusion network, and sends the key features to the abnormal operating condition judgment module; The abnormal operating condition judgment module inputs the key features into a pre-trained state assessment model to generate a sewage treatment state; based on the sewage treatment state, it determines whether the current operating state is in an abnormal operating condition. If it is determined to be an abnormal operating condition, the real-time operating data and sewage treatment state are sent to the abnormal cause identification module; The abnormality cause identification module uses a Bayesian network-based diagnostic model to identify the abnormality cause based on real-time operation data and sewage treatment status, and sends the abnormality cause and real-time operation data to the strategy generation module; The strategy generation module inputs the key features into a random forest-based strategy generation model based on the abnormal cause. The model classifies different abnormal types by integrating multiple decision trees, and generates targeted control strategies based on historical processing experience and current operating parameters.

Images