CN116343946A - Water Pollution Decision-Making Method and System Based on Neural Network - Google Patents

Abstract

The invention relates to the technical field of water pollution prevention and control, specifically a neural network-based water pollution decision-making method and system. The method includes the following contents: obtaining current water quality data; calling a preset water quality prediction model, which is based on LSTM and Markov chain construction; the water quality prediction model predicts and generates LSTM prediction results based on the current water quality data, and corrects the LSTM prediction results according to the Markov chain to output water quality prediction trends. Adopting this solution can solve the technical problem of low accuracy existing in water quality prediction using a neural network algorithm in the prior art.

Description

Water Pollution Decision-Making Method and System Based on Neural Network

technical field

The invention relates to the technical field of water pollution prevention and control, in particular to a neural network-based water pollution decision-making method and system.

Background technique

The prevention and control of water pollution is one of the major key issues in environmental protection. Although my country has abundant water resources, water pollution is serious, droughts and floods are frequent, and regional economic development is different, which makes my country's water resources management and water environment protection difficult. The problem is becoming more and more prominent.

With the rapid development and in-depth application of a new round of information technology such as global cloud computing, the Internet of Things, and the mobile Internet, major changes and new breakthroughs are brewing in the development of urban informatization. Globalization has become an inevitable trend. Water environment management is an important part of urban management, informatization will inevitably become a powerful tool for the development of water pollution prevention and control, and the construction of informatization in water pollution prevention and control is imperative.

Big data for water pollution prevention and control mainly senses the operating status of each system in real time through online monitoring equipment such as data acquisition instruments, wireless networks, and water quality and pressure gauges, analyzes and processes massive water environment information in a timely manner, and makes corresponding processing results to assist decision-making It is recommended that water pollution prevention and control be managed in a more granular and dynamic manner. In order to manage the water environment more intelligently, the water quality prediction of the water environment is carried out. The water quality prediction is based on the index data of the actual monitoring factors of the water quality, and the water quality mathematical model is used to predict the changes of the water body in the next few hours or days. Prevent water pollution incidents through water quality forecasting.

In the prior art, a neural network algorithm is used to establish a mathematical model, and the general trend of future changes in water quality is predicted through the mathematical model. In order to ensure the accuracy of the prediction of the mathematical model, the historical monitoring data are used as samples to train the mathematical model. The samples used in the training process are easily affected by external factors, and sometimes there are certain errors, resulting in random fluctuations in the prediction results within a certain range. , which reduces the prediction accuracy.

Contents of the invention

One of the objectives of the present invention is to provide a neural network-based water pollution decision-making method to solve the technical problem of low accuracy in water quality prediction using neural network algorithms in the prior art.

Basic scheme one that the present invention provides: the water pollution decision-making method based on neural network, comprises the following content:

Obtain current water quality data;

Call the preset water quality prediction model, which is constructed based on LSTM and Markov chain;

The water quality prediction model predicts and generates LSTM prediction results based on the current water quality data, and corrects the LSTM prediction results according to the Markov chain to output the water quality prediction trend.

Further, the following content is also included: obtaining historical water quality data, training the water quality prediction model according to the historical water quality data, and storing the trained water quality prediction model.

Further, the following content is also included: analyzing and determining the current water quality according to the water quality prediction trend and preset water quality grade rules to generate a water quality prediction grade.

Furthermore, both the current water quality data and the water quality forecast trend include indicator data of various monitoring factors.

Further, it also includes the following:

Obtain big data on water pollution prevention and control, and establish a water pollution prevention and control model based on the random forest algorithm based on the big data on water pollution prevention and control; the water pollution prevention and control model outputs water quality control measures based on water quality prediction trends.

Further, water quality control measures include one or more of the corresponding pollution types, major categories of control measures, subcategories of control measures, technical processes, construction difficulty, construction cost, operation difficulty, operation cost, and expected results .

Beneficial effects of basic plan one:

1. In this program, the water quality prediction model based on LSTM and Markov chain is used to analyze the fluctuation range of the LSTM prediction result error through the Markov chain, and the development trend of fluctuation is predicted, and the LSTM prediction result is analyzed through the Markov chain Corrections are made to finely optimize the trend of water quality prediction, so as to better eliminate the prediction error caused by external factors. Therefore, the establishment of a water quality prediction model combined with LSTM and Markov chain can reduce to a certain extent The error caused by the value can improve the accuracy of the prediction result.

2. Use this program to predict the trend of water quality and know the evolution trend of water quality, so as to analyze the factors causing water quality changes in advance and make early decisions on polluted river waters in time. Through the water quality forecast level, understand the change of water quality level, so as to find and solve problems in time.

3. In this scheme, the current water quality data and water quality prediction trends include multiple monitoring factors. Considering the correlation of river water pollutants, multi-factor prediction is adopted, and the interaction of multiple factors is used to jointly predict a certain level in the next moment. The index data of factors can improve the accuracy of prediction results.

4. At the same time, this program also uses a water pollution prevention and control model based on the random forest algorithm to match water quality control measures through water quality prediction trends. The random forest algorithm has the advantages of high accuracy, can effectively run on large data sets, and is not easy to overfit. When matching water quality control measures, it can automatically implement the recommendation of control measures for polluted water quality in different river basins, and provide a basis for subsequent prevention and control measures. Provide effective reference for decision-making, and realize the recommendation of water pollution control measures in different scenarios. At the same time, the water environment management tasks and goals can also be formed according to the water quality control measures, and the implementation and completion of the tasks and the effect can be tracked and assessed.

The second object of the present invention is to provide a water pollution decision system based on neural network.

The present invention provides the second basic solution: a neural network-based water pollution decision-making system, using the above-mentioned neural network-based water pollution decision-making method.

Further, including:

Data acquisition module, used to acquire current water quality data;

The water quality prediction module is preset with a water quality prediction model; the water quality prediction model is used to generate LSTM prediction results based on current water quality data, and to generate water quality prediction trends based on Markov chain correction of LSTM prediction results;

The water quality prediction module is used to obtain the water quality prediction trend output by the water quality prediction model based on the current water quality data.

Further, the data acquisition module is also used to acquire historical water quality data, including:

The model generation and training module is used to establish a water quality prediction model based on LSTM and Markov chain, train the water quality prediction model according to historical water quality data, and store the trained water quality prediction model in the water quality prediction module.

Further, it also includes:

The water pollution prevention and control module is preset with a water pollution prevention and control model;

The water pollution prevention and control module is used to obtain the water quality control measures output by the water pollution prevention and control model according to the water quality prediction trend.

Beneficial effects of the second basic plan:

1. In this program, the setting of the water quality prediction module analyzes the fluctuation range of the LSTM prediction result error through the Markov chain, and predicts the development trend of the fluctuation, and corrects the LSTM prediction result through the Markov chain to predict the water quality. The fine optimization of the trend can better eliminate the forecast error caused by external factors, and then reduce the error caused by the specific value to a certain extent, and improve the accuracy of the forecast result. At the same time, this program uses the water quality prediction module to predict the water quality prediction trend, know the water quality evolution trend, analyze the factors that cause water quality changes in advance, and make early decisions on polluted river waters in a timely manner.

2. In this plan, the setting of the water pollution prevention and control module matches the water quality control measures for the water quality prediction trend through the water pollution prevention and control model based on the random forest algorithm. The random forest algorithm has the advantages of high accuracy, can effectively run on large data sets, and is not easy to overfit. Different decision-making measures for water environment governance, thus forming water environment governance tasks and goals, and tracking and evaluating the implementation and completion of tasks and their effects.

Description of drawings

Fig. 1 is the flow chart of the water pollution decision-making system based on neural network of the present invention;

Fig. 2 is the structural representation of LSTM memory unit of the present invention;

Fig. 3 is the comparison chart of pH value prediction data and real data of the present invention;

Fig. 4 is the comparison chart of the prediction data and real data of water quality ammonia nitrogen content of the present invention;

Fig. 5 is the comparison chart of the prediction data and real data of water quality total phosphorus content of the present invention;

FIG. 6 is an integrated architecture diagram of Embodiment 2 of the present invention.

Detailed ways

The following is further described in detail through specific implementation methods:

Embodiment one

The water pollution decision-making method based on neural network, as shown in Figure 1, includes the following contents:

S1: Obtain historical water quality data, train the water quality prediction model according to the historical water quality data, and store the trained water quality prediction model.

S2: Get the current water quality data; call the preset water quality prediction model, the water quality prediction model is constructed based on LSTM (neural network) and Markov chain; the water quality prediction model generates LSTM prediction results according to the current water quality data, and according to Markov The chain corrects the LSTM prediction results to output the water quality prediction trend.

Step S1 specifically includes the following:

S101: Obtain historical water quality data. Historical water quality data includes index data of various monitoring factors collected at each section in history. Monitoring factors include water quality data including sampling time, section (location of data sampling), location (upstream and downstream of the river) One or more of water flow, rainfall, cross-section catchment area population, TP (total phosphorus content of water quality), NH4-N (ammonia nitrogen content of water quality), COD (dissolved oxygen of water quality), and pH value.

S102: Establish a water quality prediction model based on LSTM (neural network) and Markov chain. As shown in Figure 2, each circle in the figure is a point-by-point operation, that is, a corresponding operation is performed on the vector; each rectangle represents a neural network layer, and its internal characters represent the activation function used by the corresponding neural network.

Each cell of the LSTM neural network has three gates, the input gate, the forget gate and the output gate.

The forget gate takes the output data of the previous unit and the input data of this unit as the input sigmoid function, which can generate a value in the range [0, 1] for any item in the memory unit at the previous moment, through this value To control the degree to which the previous unit is forgotten.

The input gate is coupled with a tanh function to control how much new information is added. The tanh function will generate the current candidate memory cell value, and the input gate will generate a value in the range [0,1] for any of the current candidate cells, thereby controlling the degree to which new information is added.

The output gate is to control how much the current unit state is forgotten. After the unit state is activated, the output gate will generate a value in the range [0,1] for any of them, and use this value to control the unit state to be filtered. degree.

Construct an LSTM-based mathematical model according to the following formula:

f _t ＝σ(W _fx x _t +W _fh h _t-1 +b _f ) (1)

i _t ＝σ(W _ix x _t +W _ih h _t-1 +b _i ) (2)

o _t ＝σ(W _ox x _t +W _oh h _t-1 +b _o ) (3)

Among them, x _t represents the input data at the time, h _t-1 represents the output value of the LSTM unit at the previous time, C _t-1 represents the value of the memory unit at the previous time, C _t represents the value of the memory unit at time t; W _* is the weight coefficient (for example W _xi represents the weight between the corresponding input data and the input gate); b _* is the bias vector (for example, _bi is the bias vector of the input gate). σ is a sigmoid function, and its value is [0,1]. When it takes a value of 0, it means that the gate control is closed, and when it takes a value of 1, it means that the gate control is open. The formula is as in formula (4).

C′ _t ＝tanh(W _xc x _t +W _hc h _t -1+b _c ) (5)

C _t ＝f _t C _t -1+i _t C′ _t (6)

Among them, C′ _t represents the value of the current candidate memory unit, and the iterative formula for calculating the state value C _t of the memory unit at the current moment is shown in formula (6), tanh is the hyperbolic tangent activation function; W _xc is the relationship between the corresponding input data and the memory unit The weight between, W _hc is the weight between the hidden layer and the memory unit.

Suppose n(n>0) dimensional input x ₁ ,x ₂ ,...,x _n ,m(m>0) hidden layer state sequence h ₁ ,h ₂ ,...,h _m ,k (k>0)-dimensional output sequence y ₁ , y ₂ ,...,y _k , y _k is the output of the LSTM unit at time t, and the calculation formula is shown in formula (8).

yt=o _t tanh(C _t ) (8)

Although the LSTM cycle neural network model will use many sample training and test data to ensure accuracy during simulation, but these data are affected by external factors, sometimes there will be certain errors, which will lead to random fluctuations in the prediction results within a certain range , reducing the prediction accuracy. The use of Markov chain can better eliminate the prediction error caused by external factors. Therefore, the establishment of a combined prediction model of neural network and Markov chain can obtain more accurate prediction results.

Markov chain is a sequence of random variables with Markov properties, and Markov chain prediction method is a scientific and effective dynamic prediction method suitable for random processes. This method is mainly divided into two processes: one is to determine the state space of the Markov chain, and the other is to determine the state transition probability and state transition matrix through calculation.

Build a water quality prediction model based on LSTM and Markov chain according to the following formula:

In the process of event development and change, the probability of transitioning to state j in the next step in the process of state i, referred to as the state transition probability, is shown in formula (9):

P _ij ＝P{X _n+1 ＝j|X _n ＝i} (9)

Among them, P _ij is the state transition probability; X _n =i means that the process is in state i at time n, called {0, 1, 2, ...) is the state space of the process, denoted as S, and S is the general name of the state space. For a Markov chain, given the past states X ₁ , X ₂ ,...,X _n-1 , and the current state X _n .

Through the Markov chain model, it is possible to analyze the fluctuation range of the LSTM cyclic neural network simulation prediction result error, and predict the development trend of the fluctuation, and further refine the results predicted by the LSTM cyclic neural network through the state transition probability matrix of the error.

The Markov chain has the so-called "no aftereffect", that is, to determine the state of the process moment, it is only necessary to know the situation at the moment, and it is not necessary to have a complete understanding of the situation at the moment. In addition, the final prediction result of the final constructed model is not a specific value, but a set of prediction interval values with different probabilities is generated, so as to reduce the error caused by the specific value to a certain extent and improve the prediction accuracy.

S103: Train the water quality prediction model according to the historical water quality data.

In this embodiment, multi-factor prediction is adopted, so the monitoring factors include pH value, TP and NH3-N, and the indicator data of the monitoring factors in the historical water quality data are used as samples, and the water quality based on LSTM and Markov chain is established according to the samples The predictive model is trained.

Multi-factor prediction means that the index data of multiple monitoring factors contained in the water quality at the same time are interacted and affected by other factors. Considering the correlation of river water pollutants, multi-factor prediction is adopted, and the interaction of multiple factors is used to jointly predict the index data of a certain factor at the next moment, so as to improve the accuracy of the prediction results.

S104: Store the trained water quality prediction model for use in subsequent water quality predictions.

Step S2 specifically includes the following:

S201: Obtain current water quality data. The current water quality data includes index data of various monitoring factors. In this embodiment, the monitoring factors include pH value, TP and NH3-N.

S202: calling a preset water quality prediction model, and calling the water quality prediction model stored in step S104.

S203: Inputting the current water quality data into the water quality prediction model to obtain the water quality prediction trend output by the water quality prediction model. That is, the water quality prediction model predicts and generates LSTM prediction results according to the current water quality data, and corrects the LSTM prediction results according to the Markov chain to output the water quality prediction trend.

The water quality prediction trend includes indicator data of various monitoring factors. In this embodiment, the monitoring factors include pH value, TP and NH3-N. The water quality prediction trend is displayed for management personnel to view, and based on this, water quality control measures can be set, and the water quality prediction trend can also be input into a third-party model for further analysis and judgment.

In other embodiments, S2 further includes: analyzing and determining the current water quality according to the water quality forecast trend and preset water quality level rules to generate a water quality forecast level. Specifically: the water quality grade rule adopts the existing water quality grade standard, analyzes and determines the water quality grade of the water quality prediction trend according to the water quality grade rule, thereby generating the water quality prediction grade. For example, in February 2019, a certain section had Class 2 water. Combined with the monitoring factors from March to December in previous years, the water quality level at the end of 2019 is predicted.

In other embodiments, S1 also includes: acquiring big data on water pollution prevention and control, and establishing a water pollution prevention and control model based on a random forest algorithm according to the big data on water pollution prevention and control. Specifically: the existing random forest algorithm is used to establish a water pollution prevention and control model, the water pollution prevention and control model is trained according to the big data of water pollution prevention and control, and the trained water pollution prevention and control model is stored. The input of the water pollution prevention and control model is water quality data, and the output is the water quality control measures corresponding to the water quality data. Water quality control measures include one or more of the pollution type corresponding to the measure, major category of control measures, subcategory of control measures, technical process, construction difficulty, construction cost, operation difficulty, operation cost, and expected results.

S2 also includes: water quality control measures output by the water pollution prevention and control model based on the water quality prediction trend. Specifically: the water quality prediction trend is essentially water quality data, and the water quality prediction trend is used as input to the water pollution prevention and control model, and the water quality control measures that are matched and output by the water pollution prevention and control model are obtained. Through the water pollution prevention and control model, the corresponding water quality control measures are automatically recommended for water quality data matching, including the pollution type corresponding to the measures, the major categories of control measures, the subcategories of control measures, technical processes, construction difficulty, construction cost, and operation difficulty , operating costs, expected results, etc. According to the control measures and characteristic indicators, it can be manually modified and added, and finally determine the appropriate water environment control decision-making measures for the basin at this time, form the annual target task, and track and evaluate the implementation and effect of the task.

The water pollution decision-making system based on the neural network uses the above-mentioned water pollution decision-making method based on the neural network. It includes a data acquisition module, a model generation and training module and a water quality prediction module.

The data acquisition module is used to obtain historical water quality data. Historical water quality data includes index data of various monitoring factors collected by each section in history. Monitoring factors include water quality data including sampling time, section (location of data sampling), location (in the form of river Upstream and downstream), flow, rainfall, population of cross-section catchment area, TP (total phosphorus content of water quality), NH4-N (ammonia nitrogen content of water quality), COD (dissolved oxygen of water quality), and pH value .

The model generation and training module is used to establish a water quality prediction model based on LSTM and Markov chain, train the water quality prediction model according to historical water quality data, and store the trained water quality prediction model in the water quality prediction module. specific:

The model generation and training module is used to construct an LSTM-based mathematical model according to formulas (1)-(4).

The iterative formula for calculating the state value C _t of the memory unit at the current moment is shown in formulas (6) and (7).

Suppose n(n>0) dimensional input x ₁ ,x ₂ ,...,x _n ,m(m>0) hidden layer state sequence h ₁ ,h ₂ ,...,h _m ,k (k>0)-dimensional output sequence y ₁ , y ₂ ,...,y _k , y _k is the output of the LSTM unit at time t, and the calculation formula is shown in formula (8).

Although the LSTM cycle neural network model will use many sample training and test data to ensure accuracy during simulation, but these data are affected by external factors, sometimes there will be certain errors, which will lead to random fluctuations in the prediction results within a certain range , reducing the prediction accuracy. The use of Markov chain can better eliminate the prediction error caused by external factors. Therefore, the establishment of a combined prediction model of neural network and Markov chain can obtain more accurate prediction results.

Markov chain is a sequence of random variables with Markov properties, and Markov chain prediction method is a scientific and effective dynamic prediction method suitable for random processes. This method is mainly divided into two processes: one is to determine the state space of the Markov chain, and the other is to determine the state transition probability and state transition matrix through calculation.

The model generation and training module is also used to construct a water quality prediction model based on LSTM and Markov chain according to the following formula:

In the process of event development and change, the probability of transitioning to state j in the next step in the process of state i, referred to as the state transition probability, is shown in formula (9).

In this embodiment, multi-factor prediction is adopted, so the monitoring factors include pH value, TP and NH3-N, and the model generation and training module is also used to use the indicator data of the monitoring factors in the historical water quality data as samples, and according to the sample training based on LSTM and the water quality prediction model of the Markov chain, and store the trained water quality prediction model in the water quality prediction module.

The data acquisition module is also used to acquire current water quality data, which includes index data of various monitoring factors. In this embodiment, the monitoring factors include pH value, TP and NH3-N.

The water quality prediction module is preset with a water quality prediction model. The water quality prediction model is used to generate LSTM prediction results based on current water quality data, and to generate water quality prediction trends based on Markov chain correction of LSTM prediction results. Specifically: the water quality prediction module is used to obtain the water quality prediction trend output by the water quality prediction model based on the current water quality data. The water quality prediction trend includes indicator data of various monitoring factors. In this embodiment, the monitoring factors include pH value, TP and NH3-N.

In other embodiments, the neural network-based water pollution decision-making system further includes a water pollution prevention and control module.

The data acquisition module is also used to acquire big data on water pollution prevention and control, which includes water quality data and corresponding water quality control measures. Water quality control measures include one or more of the pollution type corresponding to the measure, major category of control measures, subcategory of control measures, technical process, construction difficulty, construction cost, operation difficulty, operation cost, and expected results.

The model generation and training module is also used to establish a water pollution prevention and control model based on the random forest algorithm, specifically using the existing random forest algorithm. The model generation and training module is also used to use the water pollution prevention and control big data as samples, train the water pollution prevention and control model according to the samples, and store the trained water pollution prevention and control model in the water pollution prevention and control module.

The water pollution prevention and control module is preset with a water pollution prevention and control model, and the water pollution prevention and control module is used to call the water pollution prevention and control model to obtain the water quality control measures output by the water pollution prevention and control model according to the water quality prediction trend.

Predict the water quality to obtain the trend of water quality prediction, and collect real water quality data in real time. Experiments have proved that the predicted results of this program are consistent with the real values, and the prediction results are relatively accurate. The accuracy of river water quality prediction can reach 80%, which can be applied to In the prediction of river water quality factors, predictions are made for possible water pollution in rivers. The experimental results are shown in Figures 3, 4, and 5.

Embodiment two

The differences between this embodiment and Embodiment 1 are:

As shown in Figure 6, the neural network-based water pollution decision-making system includes a big data intelligent analysis and decision-making platform for water pollution prevention and control, and an intelligent algorithm platform. The intelligent analysis of big data for water pollution prevention and control includes a configuration module, several application modules, visualization components and an application interface layer. The intelligent algorithm platform includes algorithm encapsulation and unified algorithm interface service layer.

In this application, the algorithms used include big data algorithms related to different pressure sources and water quality, multi-source heterogeneous data knowledge feature extraction and fusion algorithms, and small watershed water pollution decision-making self-learning algorithms. In this embodiment, python is used To achieve algorithm construction, the specific algorithm encapsulation strategy is: use the python process to run the model trained in deep learning, and call the service provided by the python process in the Java application. The python application and the Java application can run on different servers. Remote access calls. After the algorithm is encapsulated, the system platform transmits data in a predetermined data format, such as Word, PDF, etc., through the HTTP protocol.

In this application, the interface between the algorithm and the application is realized through the application interface layer and the unified and unified algorithm interface service layer.

The unified algorithm interface service layer includes algorithm parameter configuration interface, situation configuration interface, algorithm driver interface and algorithm result message interface, etc. Wherein, the algorithm driving interface is a core interface, which includes the driving interfaces of each algorithm in the algorithm package.

Algorithm parameter configuration interface: 1. Realize the basic technical parameter configuration function of the basic algorithm and three application algorithms. 2. Through the web service publishing interface based on the http/https protocol; 3. It is called by the configuration module of the big data intelligent analysis and decision-making platform for water pollution prevention and control.

Situation configuration interface: 1. Realize the self-learning algorithm of small watershed water pollution decision-making based on deep neural network, the relationship model algorithm of pressure source and water quality, the knowledge feature extraction and fusion algorithm of multi-source heterogeneous data of pressure source and water quality, and the basic configuration of situation configuration parameters Function. 2. Through the web service publishing interface based on the http/https protocol; 3. It is called by the configuration module of the big data intelligent analysis and decision-making platform for water pollution prevention and control.

The algorithm-driven interface is the core interface of the interface layer, which realizes the connection of characteristic data and parameters with the algorithm, including: small water pollution self-learning algorithm-driven interface based on deep neural network, and algorithm-driven interface of the relationship model between water quality and pressure source . The relationship model between water quality and pressure source. Algorithm-driven interface. The interface function requirements are as follows: 1. To realize the task-driven invocation of the three application algorithms, which is the entry point for the application system to call the algorithm. 2. Through the web service publishing interface based on the http/https protocol; 3. It is called by the algorithm calling interface of the big data intelligent analysis and decision-making platform for water pollution prevention and control.

Algorithm result message interface: 1. After the algorithm calculation is completed, the calculation result message is returned to the called application module. 2. This interface is the calling port. 3. The interface calls back the decision analysis business module of the big data intelligent analysis and decision-making platform for water pollution prevention and control to complete the transfer of task messages through the result response interface.

The application interface layer includes general web call interface, algorithm call interface and result response interface, etc.

General web call interface: 1. Use the general http/https protocol to realize the call through the web service; 2. This interface is the call interface. Such as calling the algorithm parameter configuration interface and the scenario configuration interface.

Algorithm call interface: 1. This interface calls the algorithm-driven interface of the algorithm platform, which is the entry point for the application system to call algorithm execution; 2. This interface is the call interface.

Result response interface: 1. After the algorithm operation task is completed, receive the message returned by the algorithm platform, and write the status identifier into the database; 2. Publish the interface through the web service based on the http/https protocol; 3. It is the algorithm result message of the algorithm platform called by the interface.

The application interface layer and the unified algorithm interface service layer are constructed using unified specifications. The interface protocol, interface data format, data encoding, and encapsulation method specification requirements are as follows: http/https to call and respond. Interface data format: JSON format is used for data interaction response. Interface encoding: UTF-8 encoding.

The connection between the algorithm and the platform adopts the method of separating the algorithm and the platform. A monitoring process is established on the algorithm server, and the platform invokes the algorithm through remote access of the process. The http hypertext transfer protocol is used between the platform and the algorithm server, and the information exchange between the system platform and the algorithm server is completed by using the GET and POST request methods in HTTP.

In this application, the unified data read-write interface adopts a unified non-relational data and relational data read-write interface as a unified interface between algorithms and data resources, applications and data resource read-write calls. Relational data is read or stored for application platforms and algorithm platforms through relational database drivers (JDBC, ODBC, etc. drivers). Non-relational data (plans, files, etc.) are read or stored for the application platform and algorithm platform through the file interface.

In this application, the configuration definition is used to realize data-driven, 1. Parameter configuration (integration of model basic data): when designing the system, it is necessary to realize the overall configuration of the model parameters to achieve the scheduling integration between the application and the algorithm, including various initialization conditions of the model (meshes, etc.), model input data files, etc. The parameter configuration is realized through the interface layer, and the application system connects with the "algorithm parameter configuration interface" of the unified algorithm interface service layer through the "universal web call interface". 2. Scenario configuration (customization and integration of application scenarios): When designing the system, it is necessary to realize the configuration of the situation parameters, so that the system can start to predict. The parameter configuration is realized through the interface layer, and the application system is connected with the "situation configuration interface" of the unified algorithm interface service layer through the "general web call interface".

In this application, data model design is used to realize data normalization and fusion. Data standardization is the basis of data fusion, and effective data planning (that is, data model design) is an important method for the fusion of data and applications/algorithms. The target result database (that is, decision analysis data) is oriented by analysis and decision-making, and realizes the planning and design of the target result data (that is, decision analysis data) database through target data modeling; source data modeling is the basis of algorithm source data, and realizes algorithm data integration. A fusion foundation for entry, cleansing and integration.

During application, the configuration module is used to perform parameter configuration and scenario configuration through the general web call interface, the application module is used to call the required algorithm through the algorithm call interface for analysis, and transmit the analysis results through the result response interface, and analyze the analysis results through the visualization component show.

The neural network-based water pollution decision-making method uses the above-mentioned neural network-based water pollution decision-making system.

With this solution, the integration of algorithms and applications is realized through interfaces, and the connection between algorithms and applications is effectively realized based on the application interface layer and the unified algorithm interface service layer. Through source data modeling and target data modeling design to complete the data fusion of algorithms and applications, realize the integration of application calls, algorithm operations, and data presentation, and improve integration capabilities through parameter configuration and scenario configuration.

What is described above is only an embodiment of the present invention, and the common knowledge such as the specific structure and characteristics known in the scheme is not described too much here, and those of ordinary skill in the art know all the common knowledge in the technical field to which the invention belongs before the filing date or the priority date Technical knowledge, being able to know all the existing technologies in this field, and having the ability to apply conventional experimental methods before this date, those of ordinary skill in the art can improve and implement this plan based on their own abilities under the inspiration given by this application, Some typical known structures or known methods should not be obstacles for those of ordinary skill in the art to implement the present application. It should be pointed out that for those skilled in the art, under the premise of not departing from the structure of the present invention, some modifications and improvements can also be made, which should also be regarded as the protection scope of the present invention, and these will not affect the implementation of the present invention. Effects and utility of patents. The scope of protection required by this application shall be based on the content of the claims, and the specific implementation methods and other records in the specification may be used to interpret the content of the claims.

Claims (10)

1. The water pollution decision-making method based on neural network, is characterized in that, comprises the following content: Obtain current water quality data; Call the preset water quality prediction model, which is constructed based on LSTM and Markov chain; The water quality prediction model predicts and generates LSTM prediction results based on the current water quality data, and corrects the LSTM prediction results according to the Markov chain to output the water quality prediction trend. 2. the water pollution decision-making method based on neural network according to claim 1, is characterized in that, also comprises the following content: Obtain historical water quality data, train the water quality prediction model according to the historical water quality data, and store the trained water quality prediction model. 3. the water pollution decision-making method based on neural network according to claim 1 and 2, is characterized in that, also comprises the following content: According to the water quality prediction trend and the preset water quality grade rules, the current water quality is analyzed and judged to generate the water quality prediction grade. 4. The neural network-based water pollution decision-making method according to claim 1, characterized in that: the current water quality data and the water quality prediction trend both include index data of multiple monitoring factors. 5. the water pollution decision-making method based on neural network according to claim 1, is characterized in that, also comprises the following content: Obtain big data on water pollution prevention and control, and establish a water pollution prevention and control model based on the random forest algorithm based on the big data on water pollution prevention and control; The water quality control measures output by the water pollution prevention and control model according to the water quality prediction trend. 6. The neural network-based water pollution decision-making method according to claim 5, characterized in that: water quality control measures include pollution types corresponding to measures, major categories of control measures, small categories of control measures, technical processes, construction difficulty, One or more of construction cost, operational difficulty, operating cost, and expected results. 7. A water pollution decision-making system based on a neural network, characterized in that: the water pollution decision-making method based on a neural network according to any one of claims 1-6 is used. 8. the neural network-based water pollution decision-making system according to claim 7, is characterized in that, comprising: Data acquisition module, used to acquire current water quality data; The water quality prediction module is preset with a water quality prediction model; the water quality prediction model is used to generate LSTM prediction results based on current water quality data, and to generate water quality prediction trends based on Markov chain correction of LSTM prediction results; The water quality prediction module is used to obtain the water quality prediction trend output by the water quality prediction model based on the current water quality data. 9. The neural network-based water pollution decision-making system according to claim 8, wherein: the data acquisition module is also used to obtain historical water quality data, and also includes: The model generation and training module is used to establish a water quality prediction model based on LSTM and Markov chain, train the water quality prediction model according to historical water quality data, and store the trained water quality prediction model in the water quality prediction module. 10. the neural network-based water pollution decision-making system according to claim 8, is characterized in that, also comprises: The water pollution prevention and control module is preset with a water pollution prevention and control model; The water pollution prevention and control module is used to obtain the water quality control measures output by the water pollution prevention and control model according to the water quality prediction trend.

Images