Rapid sewage source tracing method and system

Abstract

The invention relates to the technical field of sewage treatment, in particular to a method and a system for quickly tracing sewage. According to the technical scheme, the system comprises a data acquisition and processing module, a data analysis module and a result display and decision support sub-module. According to the method, the sewage pollution source can be quickly positioned, the discharge intensity can be quickly estimated, and the time interval from pollution discovery to pollution source determination is shortened; according to the method, the position of the pollution source and the emission intensity are accurately determined through comprehensive learning and analysis of multiple factors on the basis of an algorithm combining a convolutional neural network and a long-short-term memory network in deep learning, and the traceability accuracy is remarkably improved; through model training and optimization, the system has the characteristic of data driving, model overfitting is effectively prevented by adopting a regularization technology, the generalization ability of the model is enhanced, meanwhile, the model is continuously evaluated and adjusted according to evaluation indexes, and high traceability accuracy and reliability are kept.

Description

Technical Field

[0001] The present invention relates to the technical field of sewage treatment, and in particular to a method and system for rapid sewage source tracing. Background Art

[0002] With the acceleration of industrialization and urbanization, the problem of sewage discharge is becoming increasingly serious, posing a huge threat to the environment and human health. Accurately and quickly tracing the source of sewage pollution has become a key issue that needs to be urgently solved in the field of environmental protection. Traditional sewage source tracing methods are mainly divided into two categories: one is manual monitoring and analysis methods, and the other is model-based methods. Among them, model-based sewage source tracing methods include water quality model methods, which simulate the diffusion and migration of pollutants in water bodies by constructing hydrodynamic and water quality transmission models. However, this type of model has many limitations. First, the establishment of the model is based on a large number of assumptions and parameter estimates. The actual water environment is complex, and factors such as topography, water flow velocity, and meteorological conditions are constantly changing, resulting in limited model accuracy; secondly, the model requires a large amount of accurate hydrological data, topographic data, pollution source emission data and other basic data support, but the acquisition of these data often faces many difficulties, and the error or loss of data will seriously affect the accuracy and reliability of the model. In addition, when the water environment changes due to river regulation, water conservancy project construction, etc., the model needs to readjust parameters or even rebuild, which makes it difficult for the model to quickly adapt to new situations and cannot meet the needs of real-time tracing. Therefore, we propose a method and system for rapid sewage tracing. Summary of the invention

[0003] The purpose of the present invention is to provide a method and system for rapid sewage source tracing in view of the problems existing in the background technology.

[0004] In the first aspect: the present application provides a method for rapid sewage source tracing, comprising the following steps: data collection and preprocessing, setting sensors in the sewage pipe network and receiving water bodies to collect sewage data, and integrating multi-source information to preprocess the collected data to ensure data quality; An algorithm combining convolutional neural network and long short-term memory network is used to build an artificial intelligence model, and a back propagation algorithm combined with a mean square error function is used to train and optimize the model; Real-time sewage tracing and result evaluation: input real-time sewage data into the artificial intelligence model to output tracing results, evaluate the accuracy of the tracing results, and adjust the optimization model based on the evaluation feedback to ensure the reliability and accuracy of the tracing results; Result display and decision support: The traceability results are integrated and displayed by marking the location of the pollution source on the map of the map information system and calculating and drawing the pollution diffusion range. The results are displayed through charts and the feasibility of the pollution control plan is evaluated based on the traceability results and relevant environmental data.

[0005] Optionally, the data collection and preprocessing includes: Data collection: Collect sewage data through sensors, including water quality sensors, flow sensors and water level sensors, and integrate regional terrain and land use type information obtained by satellite remote sensing, meteorological data from weather stations, including rainfall, wind speed, wind direction, and pollution source information registered by the environmental protection department as the collected data; Preprocessing operation: Use statistical methods to set a reasonable data range for the collected sensor data to remove outliers and noise data. The specific operations are as follows: For different types of sensor data, calculate the mean of their historical data respectively and standard deviation , set a reasonable data range as , where The coefficient is determined based on the data distribution characteristics and actual experience, and takes a value between 2 and 3. Data outside this range are considered outliers and are removed; The data from different sources are standardized according to the normalization formula to have a unified dimension and data format. The normalization formula is as follows: ; in, is the original data, is the minimum value of the data type. is the maximum value, is the normalized data; Missing data are processed by using linear interpolation to supplement a small number of consecutive missing data points. For a large number of missing or non-continuous missing data, a model-based method is used to predict and supplement the data according to the historical data distribution law of the data type and the correlation with other related data, so as to ensure the integrity and continuity of the data and provide a reliable data foundation for subsequent model construction and analysis.

[0006] Optionally, the building of an artificial intelligence model includes: The algorithm data input layer based on the combination of convolutional neural network and long short-term memory network in deep learning converts the preprocessed water quality data sequence into , flow data series , meteorological data series , pollution source information coding and geospatial information coding As input to the model, represents a time series; Convolution layer operation: extract the local features of the data through multiple convolution kernels. The convolution operation formula is: ; in, Output feature map of the convolutional layer, For input data, is the convolution kernel, is the index of the corresponding dimension; Pooling: The pooling layer performs a pooling operation on the feature map output by the convolutional layer to reduce the data dimension. The pooling operation adopts maximum pooling, and the maximum pooling formula is as follows: ; in, is the output of the maximum pooling layer, is the index within the pooling window, Long short-term memory network layer: The feature sequence output by the pooling layer is input into the long short-term memory network layer to learn the temporal dependency of the data. The calculation of the long short-term memory network unit includes a forget gate. , Input Gate , cell state update , cell status , output gate and hidden state ; in, for function, represents element-wise multiplication, and are the weight matrix and bias vector respectively, is the hidden state at the last moment, is the current input, and the number of units in the long short-term memory network layer is determined according to the temporal characteristics of the sewage data and the model training effect; Fully connected layer integration and mapping: The output of the long short-term memory network layer is integrated and mapped through the fully connected layer, and finally the sewage source tracing results, the location of the pollution source and the possible pollution emission intensity estimation are output.

[0007] Optionally, the model training and optimization includes the following steps: Data division: The collected historical data is divided into training set, verification set and test set according to the ratio of 70% for training, 20% for verification and 10% for testing. The training set is used for learning and adjusting model parameters, the verification set is used to verify the generalization ability of the model and adjust hyperparameters during the training process, and the test set is used to finally evaluate the performance of the model on unseen data; Back propagation training: The model is trained using the back propagation algorithm with the mean square error as the error function. The mean square error formula is: ; in, is the sample size, is the true value, is the predicted value; In the back propagation process, the gradient of the loss to the weights of each layer is calculated according to the error function, and the weights are updated according to the gradient descent method. The weight update formula is: ; in, is the updated weight, is the current weight, is the learning rate. The initial learning rate is set between 0.001 and 0.1. During the training process, a learning rate decay strategy is used to balance the model training speed and convergence effect. The learning rate decay strategy includes an exponential decay method. The exponential decay formula is: ; in, is the initial learning rate, is the attenuation factor, is the number of training rounds and iterations Set between 100-1000 times; Regularization constraint: L1 and L2 regularization are used to constrain the model weights. The regularization constraint is used to prevent the model from overfitting. The L1 regularization is used to make the model weights sparse, and the L2 regularization is used to limit the size of the weights. Among them, the L1 regularization term is expressed as: ; The L2 regularization term is expressed as: ; in, , is the regularization coefficient, is the model weight, the regularization coefficient , Determined by cross-validation method, the value range is between 0.001-0.1; Random dropout: Random dropout is used to increase the randomness of the model. It uses random dropout to randomly discard some neuron connections during training with a probability of , the probability Set between 0.2-0.5.

[0008] Optionally, the real-time sewage source tracing and result evaluation includes: Real-time source tracing calculation; inputting the sewage data into the artificial intelligence model to calculate and output the pollution source location and pollution emission intensity estimation results; Calculation of result evaluation indicators: The result evaluation indicators include accuracy, recall rate, and F1 value. The traceability results are evaluated according to the result evaluation indicators, and the model is adjusted and optimized according to the evaluation results; Accuracy calculation: The accuracy is used to reflect the correct prediction ratio of the model as a whole. The accuracy is calculated using the accuracy formula, which is as follows: ; Recall rate calculation: The recall rate is used to measure the ability of the model to correctly identify the actual pollution source. The recall rate formula is as follows: ; F1 value calculation: The F1 value formula is as follows: ; in, True positives are the number of pollution sources that the model correctly predicts and are actually pollution sources; is the true negative, that is, the number of non-polluting sources correctly predicted by the model and actually non-polluting sources; is the false positive, which is the number of sources that the model incorrectly predicts to be pollution sources but are actually non-pollution sources; False negatives, that is, the number of sources that are incorrectly predicted by the model as non-pollution sources but are actually pollution sources; Root mean square error calculation: The root mean square error is used to evaluate the accuracy of the pollution emission intensity estimate. The formula is: ; in, is the actual pollution emission intensity, To predict the intensity of pollution emissions.

[0009] Optionally, in the result display and decision support, the calculation and drawing of the pollution diffusion range is calculated using a Gaussian diffusion model, and the formula is expressed as: ; in, A point in space The pollutant concentration at is the emission intensity of pollution sources, are the lateral and vertical diffusion coefficients, respectively, is the average wind speed, The effective height of the pollution source is used to calculate and draw the pollution diffusion range in combination with geographic information; The trend of sewage quality indicators in different time periods is displayed in a bar chart. At the same time, the pollution source information is displayed in a table, including the type, name, location coordinates, historical violation records, and basic information of the affiliated enterprise or region; According to the traceability results and relevant environmental data, a pollution control plan is formulated using a cost-benefit analysis algorithm, the costs and expected environmental benefits of different control measures are calculated, and the feasibility of the plan is evaluated using a cost-benefit ratio formula, which is as follows: ; in, For the cost-effectiveness ratio, For the Annual environmental benefits, For the The annual cost, The implementation cycle of the governance plan; Based on the traceability results and historical data of pollution sources, the risk assessment algorithm is used to adjust the pollution source supervision strategy and to perform risk scoring on the pollution sources. The risk scoring formula is: ; in, Score the risk, For the The weight of the factors, For the The score of the factor, Based on the number of factors, pollution sources are divided into different levels according to their scores. The supervision frequency is increased and monitoring methods are strengthened for high-risk pollution sources, while the supervision process is simplified for low-risk pollution sources.

[0010] Second aspect: This application provides a system for rapid sewage tracing, including: A data acquisition and processing module, wherein the data acquisition and processing module comprises a data acquisition submodule and a data preprocessing submodule, wherein the data acquisition submodule is used to acquire relevant data, and the data preprocessing submodule is used to preprocess the acquired data to ensure data quality; A data analysis module, which includes a model training and management submodule and a source tracing analysis submodule. The model training and management submodule is used to train the model, and the source tracing analysis submodule performs sewage source tracing analysis based on the model and outputs sewage source tracing results; Result display and decision support submodule, the result display interface and decision support submodule include result display interface and decision support submodule, the result display interface displays sewage source tracing results, water quality change trend, pollution source information in the form of charts, the decision support submodule is used to evaluate the feasibility of pollution control plans for environmental protection departments based on source tracing results and related environmental data: Optionally, the data acquisition submodule includes a sensor network and a data acquisition terminal, wherein the sensor network is composed of water quality sensors, flow sensors, and water level sensors distributed in the sewage pipe network and the receiving water body, and the collected data is transmitted to the data processing center in real time through wireless transmission technology, and the wireless transmission technology includes Wi-Fi, Bluetooth or 4G / 5G network; the data acquisition terminal is used to receive and preliminarily process the sensor data, compress and encrypt the data, and connect to the satellite remote sensing receiving equipment and the meteorological data interface to obtain geographic space information and meteorological data and then transmit them to the data processing center; The data preprocessing submodule is used to perform preprocessing operations on the collected data, and the preprocessing operations include data cleaning, data standardization, and missing data processing. The data cleaning is used to regard the data in the sensor data that exceeds the set range as abnormal values ​​and eliminate them. The data standardization is used to standardize the data through a normalization formula so that the data has a unified dimension and data format.

[0011] Optionally, the model training and management submodule constructs an artificial intelligence model and uses a back propagation algorithm to train, optimize and save the model with a mean square error as an error function; The source tracing analysis submodule performs sewage source tracing analysis by inputting real-time data into the model, outputs the pollution source location and pollution emission intensity estimation results and visualizes them on the map of the map information system, wherein the source tracing analysis submodule uses an algorithm combining convolutional neural network and long short-term memory network in deep learning to perform sewage source tracing analysis, and the input real-time data includes pre-processed water quality data sequence, flow data sequence, meteorological data sequence, pollution source information code and geographic space information code.

[0012] In summary, the present application includes at least one of the following beneficial technical effects: 1. The present invention uses multi-source data collected in real time and artificial intelligence algorithms to quickly locate sewage pollution sources and estimate emission intensity. Compared with the traditional method that relies on manual sampling and laboratory analysis, it greatly shortens the time interval from discovering pollution to determining the pollution source. It can respond promptly at the early stage of sewage pollution incidents, effectively avoid further expansion of the pollution scope, and reduce the difficulty and cost of subsequent governance; 2. Through the algorithm based on the combination of convolutional neural network and long short-term memory network in deep learning, through the comprehensive learning and analysis of multiple factors such as sewage pipe network structure, geographical spatial distribution of water bodies, sewage flow and water quality changes over time, and meteorological conditions, it is possible to accurately determine the location of pollution sources and emission intensity. The accuracy of tracing the source has been significantly improved, reducing the possibility of misjudgment and missed judgment, and providing a reliable basis for formulating scientific and reasonable pollution control plans; 3. The present invention performs model training and optimization based on historical data and real-time collected data, so that the system has data-driven characteristics. With the accumulation and updating of data, the model can continuously learn and adapt to new data patterns, thereby improving its own performance. It also effectively prevents model overfitting by adopting regularization technology, thereby enhancing the generalization ability of the model. At the same time, the model is continuously evaluated and adjusted according to the evaluation indicators, so that it can be continuously optimized in practical applications and maintain a high traceability accuracy and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A flow chart of a method and system for rapid sewage source tracing of the present invention is provided. DETAILED DESCRIPTION

[0014] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0015] Example

[0016] The present invention proposes a method for rapid sewage source tracing, such as Figure 1 As shown, the following steps are included: 1. System deployment and data collection Data collection: Several groups of water quality sensors, flow sensors and water level sensors are set up around key nodes of a city's sewage pipe network and major receiving water bodies (including rivers and lakes) to collect sewage data. At the same time, the data collection terminal is connected to the satellite remote sensing receiving equipment to obtain the terrain data and land use type information of the urban area. At the same time, it is connected to the meteorological data interface to receive meteorological data provided by the local meteorological station, including rainfall, wind speed and wind direction. In addition, the production process of industrial enterprises, the location and discharge of sewage outlets, the operation of sewage treatment facilities, and pollution source information, including the distribution of domestic sewage discharge points and discharge estimation data, are imported from the pollution source database of the environmental protection department. Data preprocessing: After receiving the sensor data, the data preprocessing submodule first cleans the data and calculates the mean of its historical data for water quality data. and standard deviation , set a reasonable data range as , where The coefficient is determined based on the data distribution characteristics and actual experience, and the value is preferably 2.5. Data outside this range are considered outliers and are removed; Then the data from different sources are standardized. Taking traffic data as an example, it is converted into a value between 0 and 1 through the normalization formula, so that it has a unified dimension and data format with other data, which is convenient for subsequent model processing. The normalization formula is as follows: ; in, is the original data, is the minimum value of the data type. is the maximum value, is the normalized data; For a small number of missing flow data points, linear interpolation is used to supplement them. If multiple data points are missing continuously within a certain period of time, a time series model based on the rules of historical flow data is used for prediction and supplementation; 2. Data Analysis The preprocessed data is divided into training set, validation set and test set in the proportion of 70%, 20% and 10% through the model training and management submodule, and a model combining convolutional neural network and long short-term memory network is constructed using the deep learning framework; In model construction, the preprocessed water quality data series , flow data series , meteorological data series , pollution source information coding and geospatial information coding The model is introduced as input for convolution operation, where represents a time series; Convolution layer operation: The convolution layer sets two scales of convolution kernels, 3x3 and 5x5, to extract the multi-scale spatial characteristics of pollutant distribution in sewage pipe networks and water bodies. The convolution operation formula is: ; in, Output feature map of the convolutional layer, For input data, is the convolution kernel, is the index of the corresponding dimension; The pooling layer uses the maximum pooling operation to reduce the data dimension. The maximum pooling formula is as follows: ; in, is the output of the maximum pooling layer, is the index within the pooling window; Long short-term memory network layer: The feature sequence output by the pooling layer is input into the long short-term memory network layer to learn the temporal dependency of the data. The calculation of the long short-term memory network unit includes the forget gate , Input Gate , cell state update , cell status , output gate and hidden state ; in, for function, represents element-wise multiplication, and are the weight matrix and bias vector respectively, is the hidden state at the last moment, For the current input, the number of units in the long short-term memory network layer is preferably 64 units, which are used to learn the temporal dependency of the data; Fully connected layer integration and mapping: The output of the long short-term memory network layer is integrated and mapped through the fully connected layer, and the final output is the sewage source tracing results, the location of the pollution source and the possible pollution emission intensity estimation; Model training: The back propagation algorithm is used to train the model with mean square error as the error function. The mean square error formula is: ; in, is the sample size, is the true value, is the predicted value; In the back propagation process, the gradient of the loss to the weights of each layer is calculated according to the error function, and the weights are updated according to the gradient descent method. The weight update formula is: ; in, is the updated weight, is the current weight, is the learning rate. During the training process, a learning rate decay strategy is used to balance the model training speed and convergence effect. The learning rate decay strategy includes an exponential decay method. The exponential decay formula is: ; in, is the initial learning rate, is the attenuation factor, is the number of training rounds, and the initial learning rate is Set to 0.005, the attenuation factor The value is 0.95, the number of iterations is set to 500, and the early stopping method is used to prevent overfitting. The training is stopped when the loss on the validation set does not decrease for 8 consecutive epochs. At the same time, L1 and L2 regularization are used to constrain the model weights, where the L1 regularization term is expressed as: ; The L2 regularization term is expressed as: ; in, , is the regularization coefficient, is the model weight, L1 regularization coefficient is 0.01, L2 regularization coefficient The random dropout technique is used to set the dropout probability to 0.3. The hyperparameters such as the convolution kernel size, the number of long short-term memory network units, the learning rate, and the regularization coefficient are optimized through grid search to improve the model performance. During the model training process, the model parameters are regularly updated according to the newly collected data, and the model version is managed, and the parameter settings and evaluation index results of each training are recorded. Functional realization of the source analysis submodule: The real-time collected data is input into the trained model through the source analysis submodule. When the sewage quality in a certain area fluctuates abnormally, its COD concentration suddenly increases, and the flow data shows that the sewage flow increases, the model combines the current meteorological data, geographic spatial information and pollution source information to quickly calculate and output the possible pollution source location and pollution emission intensity estimation results, and visualize the results on the map of the map information system. The latitude and longitude coordinates of the pollution source are converted into map projection coordinates for precise marking using the geodetic coordinate system conversion formula. At the same time, the pollution diffusion range is calculated based on the Gaussian diffusion model combined with water flow direction, terrain and other geographic information, and the corresponding area is drawn on the map. The calculation formula is expressed as: ; in, A point in space The pollutant concentration at is the emission intensity of pollution sources, , are the lateral and vertical diffusion coefficients, respectively, is the average wind speed, is the effective height of the pollution source; 3. Results presentation and decision support The result display interface presents content: The result display interface displays the sewage source tracing results, water quality change trends, and pollution source information in the form of charts. On the map of the map information system, the location of the pollution source and the pollution diffusion range are marked, and different colors are used to represent different levels of pollution concentration. At the same time, a bar chart is used to show the change trend of the water quality indicators of the sewage in the area in the past week. The water quality indicators include COD, ammonia nitrogen, and total phosphorus content, and the pollution source information is listed in detail in the form of a table. The pollution source information includes the type of the chemical enterprise as an industrial pollution source, the name, the precise location coordinates, the historical violation records in the past year, and the basic production information of the enterprise to which it belongs; The decision support submodule assists the decision-making process: The decision support submodule works based on the traceability results and relevant environmental data. It uses a cost-benefit analysis algorithm to develop pollution control plans for the pollution sources of the above-mentioned chemical companies, calculates the costs of different control measures, including the construction costs and operating costs of upgrading and renovating the company's sewage treatment facilities, and the social costs that may be incurred on the surrounding environment and residents, and evaluates the expected environmental benefits, including the amount of pollutant emissions reduced and the degree of water quality improvement. The cost-benefit ratio of different plans is calculated through the cost-benefit ratio formula, and the environmental protection department is provided with governance plan recommendations. The cost-benefit ratio formula is as follows: ; in, For the cost-effectiveness ratio, For the Annual environmental benefits, For the The annual cost, The implementation cycle of the governance plan; Based on the traceability results and historical data of pollution sources, the risk assessment algorithm is used to adjust the pollution source supervision strategy and to perform risk scoring on the chemical enterprise. Factors considered include pollution emission intensity, emission stability, historical violation frequency, and surrounding environmental sensitivity. The risk score is calculated using the risk scoring formula. The risk scoring formula is: ; in, Score the risk, For the The weight of the factors, For the The score of the factor, Based on the number of factors, the pollution sources are divided into different levels according to the scores. The supervision frequency of high-risk pollution sources is increased and monitoring methods are strengthened. The supervision process is simplified for low-risk pollution sources. The enterprise is listed as a high-risk pollution source according to the risk score, and the supervision frequency is increased from the original monthly inspection to weekly inspection. The monitoring methods are strengthened, and online monitoring equipment is installed to monitor its sewage discharge in real time to ensure that the enterprise strictly abides by environmental protection laws and regulations and reduce the risk of illegal sewage discharge; In a second aspect, the present application provides a system for rapid sewage tracing, comprising: The data acquisition and processing module includes a data acquisition submodule and a data preprocessing submodule. The data acquisition submodule is used to acquire relevant data, and the data preprocessing submodule is used to preprocess the acquired data to ensure data quality. The data acquisition submodule includes a sensor network and a data acquisition terminal. The sensor network consists of water quality sensors, flow sensors, and water level sensors distributed in the sewage pipe network and receiving water bodies. The collected data is transmitted to the data processing center in real time through wireless transmission technology. The wireless transmission technology includes Wi-Fi, Bluetooth, or 4G / 5G network. The data acquisition terminal is used to receive and preliminarily process sensor data, compress and encrypt the data, and connect to the satellite remote sensing receiving equipment and the meteorological data interface to obtain geographic space information and meteorological data, and then transmit it to the data processing center. The data preprocessing submodule is used to perform preprocessing operations on the collected data. The preprocessing operations include data cleaning, data standardization, and missing data processing. Data cleaning is used to treat the data that exceeds the set range in the sensor data as abnormal values ​​and remove them. Data standardization is used to standardize the data through the normalization formula to make the data have a unified dimension and data format; Data analysis module: The data analysis module includes a model training and management submodule and a source tracing analysis submodule. The model training and management submodule is used to train the model. The source tracing analysis submodule performs sewage source tracing analysis based on the model and outputs the sewage source tracing results. Among them, the model training and management submodule builds an artificial intelligence model and uses the back propagation algorithm to train, optimize and save the model with the mean square error as the error function; In the source tracing analysis submodule, real-time data is input into the model to perform sewage source tracing analysis, and the pollution source location and pollution emission intensity estimation results are output and visualized on the map of the map information system. The source tracing analysis submodule uses an algorithm combining convolutional neural network and long short-term memory network in deep learning to perform sewage source tracing analysis. The input real-time data includes pre-processed water quality data sequence, flow data sequence, meteorological data sequence, pollution source information coding and geographic space information coding; Result display and decision support submodule,The result display interface and decision support submodule include result display interface and decision support submodule, the result display interface displays sewage source tracing results, water quality change trends, and pollution source information in the form of charts, and the decision support submodule is used to evaluate the feasibility of pollution control plans for environmental protection departments based on source tracing results and related environmental data.

[0017] The above specific embodiments are only several optional embodiments of the present invention. Based on the technical solutions of the present invention and the relevant inspirations of the above embodiments, those skilled in the art can make various alternative improvements and combinations to the above specific embodiments.

Claims

1. A method for rapid sewage tracing, comprising the following steps: Data collection and preprocessing: sensors are installed in the sewage network and receiving water bodies to collect sewage data, and multi-source information is integrated to preprocess the collected data to ensure data quality; An algorithm combining convolutional neural network and long short-term memory network is used to build an artificial intelligence model, and a back propagation algorithm combined with a mean square error function is used to train and optimize the model; Real-time sewage tracing and result evaluation: input real-time sewage data into the artificial intelligence model to output tracing results, evaluate the accuracy of the tracing results, and adjust the optimization model based on the evaluation feedback to ensure the reliability and accuracy of the tracing results; Result display and decision support: The traceability results are integrated and displayed by marking the location of the pollution source on the map of the map information system and calculating and drawing the pollution diffusion range. The results are displayed through charts and the feasibility of the pollution control plan is evaluated based on the traceability results and relevant environmental data.

2. A method for rapid sewage source tracing according to claim 1, characterized in that: The data collection and preprocessing include: Data collection: Collect sewage data through sensors, including water quality sensors, flow sensors and water level sensors, and integrate regional terrain and land use type information obtained by satellite remote sensing, meteorological data from weather stations, including rainfall, wind speed, wind direction, and pollution source information registered by the environmental protection department as the collected data; Preprocessing operation: Use statistical methods to set a reasonable data range for the collected sensor data to remove outliers and noise data. The specific operations are as follows: For different types of sensor data, calculate the mean of their historical data respectively and standard deviation , set a reasonable data range as , where The coefficient is determined based on the data distribution characteristics and actual experience, and takes a value between 2 and 3. Data outside this range are considered outliers and are removed; The data from different sources are standardized according to the normalization formula to have a unified dimension and data format. The normalization formula is as follows: ； in, is the original data, is the minimum value of the data type. is the maximum value, is the normalized data; Missing data are processed by using linear interpolation to supplement a small number of consecutive missing data points. For a large number of missing or non-continuous missing data, a model-based method is used to predict and supplement the data according to the historical data distribution law of the data type and the correlation with other related data, so as to ensure the integrity and continuity of the data and provide a reliable data foundation for subsequent model construction and analysis.

3. A method for rapid sewage source tracing according to claim 1, characterized in that: The constructing of the artificial intelligence model comprises: The algorithm data input layer based on the combination of convolutional neural network and long short-term memory network in deep learning converts the preprocessed water quality data sequence into , flow data series , meteorological data series , pollution source information coding and geospatial information coding As input to the model, represents a time series; Convolution layer operation: extract the local features of the data through multiple convolution kernels. The convolution operation formula is: ； in, Output feature map of the convolutional layer, For input data, is the convolution kernel, is the index of the corresponding dimension; Pooling: The pooling layer performs a pooling operation on the feature map output by the convolutional layer to reduce the data dimension. The pooling operation adopts maximum pooling, and the maximum pooling formula is as follows: ； in, is the output of the maximum pooling layer, is the index within the pooling window; Long short-term memory network layer: The feature sequence output by the pooling layer is input into the long short-term memory network layer to learn the temporal dependency of the data. The calculation of the long short-term memory network unit includes a forget gate. , Input Gate , cell state update , cell status , output gate and hidden state ; in, for function, represents element-wise multiplication, and are the weight matrix and bias vector respectively, is the hidden state at the last moment, is the current input, and the number of units in the long short-term memory network layer is determined according to the temporal characteristics of the sewage data and the model training effect; Fully connected layer integration and mapping: The output of the long short-term memory network layer is integrated and mapped through the fully connected layer, and finally the sewage source tracing results, the location of the pollution source and the possible pollution emission intensity estimation are output.

4. A method for rapid sewage source tracing according to claim 1, characterized in that: The model training and optimization includes the following steps: Data division: The collected historical data is divided into training set, verification set and test set according to the ratio of 70% for training, 20% for verification and 10% for testing. The training set is used for learning and adjusting model parameters, the verification set is used to verify the generalization ability of the model and adjust hyperparameters during the training process, and the test set is used to finally evaluate the performance of the model on unseen data; Back propagation training: The model is trained using the back propagation algorithm with the mean square error as the error function. The mean square error formula is: ； in, is the sample size, is the true value, is the predicted value; In the back propagation process, the gradient of the loss to the weights of each layer is calculated according to the error function, and the weights are updated according to the gradient descent method. The weight update formula is: ； in, is the updated weight, is the current weight, is the learning rate. The initial learning rate is set between 0.001 and 0.

1. During the training process, a learning rate decay strategy is used to balance the model training speed and convergence effect. The learning rate decay strategy includes an exponential decay method. The exponential decay formula is: ； in, is the initial learning rate, is the attenuation factor, is the number of training rounds and iterations Set between 100-1000 times; Regularization constraint: L1 and L2 regularization are used to constrain the model weights. The regularization constraint is used to prevent the model from overfitting. The L1 regularization is used to make the model weights sparse, and the L2 regularization is used to limit the size of the weights. Among them, the L1 regularization term is expressed as: ； The L2 regularization term is expressed as: ； in, , is the regularization coefficient, is the model weight, the regularization coefficient , Determined by cross-validation method, the value range is between 0.001-0.1; Random dropout: Random dropout is used to increase the randomness of the model. It uses random dropout to randomly discard some neuron connections during training with a probability of , the probability Set between 0.2-0.

5.

5. A method for rapid sewage source tracing according to claim 1, characterized in that: The real-time sewage tracing and result evaluation includes: Real-time source tracing calculation; inputting the sewage data into the artificial intelligence model to calculate and output the pollution source location and pollution emission intensity estimation results; Calculation of result evaluation indicators: The result evaluation indicators include accuracy, recall rate, and F1 value. The traceability results are evaluated according to the result evaluation indicators, and the model is adjusted and optimized according to the evaluation results; Accuracy calculation: The accuracy is used to reflect the correct prediction ratio of the model as a whole. The accuracy is calculated using the accuracy formula, which is as follows: ； Recall rate calculation: The recall rate is used to measure the ability of the model to correctly identify the actual pollution source. The recall rate formula is as follows: ； F1 value calculation: The F1 value formula is as follows: ； in, True positives are the number of pollution sources that the model correctly predicts and are actually pollution sources; is the true negative, that is, the number of non-polluting sources correctly predicted by the model and actually non-polluting sources; is the false positive, which is the number of sources that the model incorrectly predicts to be pollution sources but are actually non-pollution sources; False negatives, that is, the number of sources that are incorrectly predicted by the model as non-pollution sources but are actually pollution sources; Root mean square error calculation: The root mean square error is used to evaluate the accuracy of the pollution emission intensity estimate. The formula is: ； in, is the actual pollution emission intensity, To predict the intensity of pollution emissions.

6. A method for rapid sewage source tracing according to claim 1, characterized in that: In the result display and decision support, the calculation and drawing of the pollution diffusion range is calculated using the Gaussian diffusion model, and the formula is expressed as: ； in, A point in space The pollutant concentration at is the emission intensity of pollution sources, , are the lateral and vertical diffusion coefficients, respectively, is the average wind speed, The effective height of the pollution source is used to calculate and draw the pollution diffusion range in combination with geographic information; The trend of sewage quality indicators in different time periods is displayed in a bar chart. At the same time, the pollution source information is displayed in a table, including the type, name, location coordinates, historical violation records, and basic information of the affiliated enterprise or region; According to the traceability results and relevant environmental data, a pollution control plan is formulated using a cost-benefit analysis algorithm, the costs and expected environmental benefits of different control measures are calculated, and the feasibility of the plan is evaluated using a cost-benefit ratio formula, which is as follows: ； in, For the cost-effectiveness ratio, For the Annual environmental benefits, For the The annual cost, The implementation cycle of the governance plan; Based on the traceability results and historical data of pollution sources, the risk assessment algorithm is used to adjust the pollution source supervision strategy and to perform risk scoring on the pollution sources. The risk scoring formula is: ； in, Score the risk, For the The weight of the factors, For the The score of the factor, Based on the number of factors, pollution sources are divided into different levels according to their scores. The supervision frequency is increased and monitoring methods are strengthened for high-risk pollution sources, while the supervision process is simplified for low-risk pollution sources.

7. A system for rapid sewage tracing, characterized in that: include: A data acquisition and processing module, wherein the data acquisition and processing module comprises a data acquisition submodule and a data preprocessing submodule, wherein the data acquisition submodule is used to acquire relevant data, and the data preprocessing submodule is used to preprocess the acquired data to ensure data quality; A data analysis module, which includes a model training and management submodule and a source tracing analysis submodule. The model training and management submodule is used to train the model, and the source tracing analysis submodule performs sewage source tracing analysis based on the model and outputs sewage source tracing results; The result display and decision support submodule includes a result display interface and a decision support submodule. The result display interface displays the sewage source tracing results, water quality change trends, and pollution source information in the form of charts. The decision support submodule is used to evaluate the feasibility of pollution control plans for environmental protection departments based on source tracing results and related environmental data.

8. A rapid sewage tracing system according to claim 7, characterized in that: The data acquisition submodule includes a sensor network and a data acquisition terminal, wherein the sensor network is composed of water quality sensors, flow sensors, and water level sensors distributed in the sewage pipe network and the receiving water body, and the collected data is transmitted to the data processing center in real time through wireless transmission technology, and the wireless transmission technology includes Wi-Fi, Bluetooth or 4G / 5G network; the data acquisition terminal is used to receive and preliminarily process the sensor data, compress and encrypt the data, and connect with the satellite remote sensing receiving equipment and the meteorological data interface to obtain geographic space information and meteorological data and then transmit them to the data processing center; The data preprocessing submodule is used to perform preprocessing operations on the collected data, and the preprocessing operations include data cleaning, data standardization, and missing data processing. The data cleaning is used to regard the data in the sensor data that exceeds the set range as abnormal values ​​and eliminate them. The data standardization is used to standardize the data through a normalization formula so that the data has a unified dimension and data format.

9. A rapid sewage tracing system according to claim 7, characterized in that: The model training and management submodule constructs an artificial intelligence model and uses a back propagation algorithm to train, optimize and save the model with mean square error as the error function; The source tracing analysis submodule performs sewage source tracing analysis by inputting real-time data into the model, outputs the pollution source location and pollution emission intensity estimation results and visualizes them on the map of the map information system, wherein the source tracing analysis submodule uses an algorithm combining convolutional neural network and long short-term memory network in deep learning to perform sewage source tracing analysis, and the input real-time data includes pre-processed water quality data sequence, flow data sequence, meteorological data sequence, pollution source information code and geographic space information code.

Images