A water quality prediction method, device, equipment and storage medium

Abstract

The application relates to the technical field of cloud, and provides a water quality prediction method, device, equipment and storage medium. The method comprises the following steps: acquiring first historical monitoring data sets collected at each location in a target area, each first historical monitoring data set containing first historical monitoring data from different data sources; obtaining, through shallow feature extraction, intraclass correlation features representing the influence degree of the first historical monitoring data of the same data type collected at surrounding locations on the water quality of the corresponding location; and then extracting water quality prediction features from the intraclass correlation features through deep feature extraction, wherein the features represent the first historical monitoring data sets collected at each location in each historical stage and the first historical monitoring data sets collected at the corresponding surrounding locations in each historical stage, and the influence degree of the features on the prediction of the water quality of the corresponding location in a future stage. The spatial correlation and the time correlation of the data are considered, and the water quality prediction accuracy is greatly improved.

Description

Technical Field

[0001] This application relates to the field of cloud technology and provides a water quality prediction method, apparatus, equipment, and storage medium. Background Technology

[0002] Surface water is one of the most important sources of drinking water for humans; therefore, the quality of surface water environment has a significant impact on people's daily lives and production. Currently, water quality prediction methods based on hydrodynamics and water quality prediction methods based on machine learning are commonly used to predict the water quality of surface water.

[0003] Natural evaporation of surface water and fluid reactions between surface water and pipelines can both cause changes in surface water quality. However, in water quality prediction methods based on hydrodynamics, it is difficult to construct mathematical models to simulate changes in surface water quality, making it impossible to accurately predict water quality conditions.

[0004] In machine learning-based water quality prediction methods, either only water quality data is considered, which limits the application scenarios of the water quality prediction model, and the model may gradually become ineffective as factors such as climate and environment change, making it unable to accurately assess the water quality of surface water; or only historical monitoring data of a single water quality monitoring station are considered, making it difficult for the water quality prediction model to accurately assess the water quality of the area where the station is located. Summary of the Invention

[0005] This application provides a water quality prediction method, apparatus, device, and storage medium to address the problem of low accuracy in water quality prediction.

[0006] In a first aspect, embodiments of this application provide a water quality prediction method, including:

[0007] Acquire the first historical monitoring datasets collected at different locations within the target area at various historical stages. Each first historical monitoring dataset contains first historical monitoring data from various data sources.

[0008] The first historical monitoring data corresponding to data sources belonging to the same data type are divided into a group of homogeneous data, and the intra-class correlation features of each group of homogeneous data are extracted. Each intra-class correlation feature represents the degree of influence of homogeneous data collected from surrounding locations on the water quality of the corresponding location.

[0009] Deep feature extraction is performed on the intra-class correlation features to obtain the corresponding water quality prediction features. The water quality prediction features represent the first historical monitoring datasets collected at each location in each historical stage, as well as the first historical monitoring datasets collected at the corresponding surrounding locations in each historical stage, and the degree of influence on predicting the water quality of the corresponding locations in the future stage.

[0010] Based on the obtained water quality prediction characteristics, the water quality prediction results for each location within the target area are determined.

[0011] Secondly, embodiments of this application also provide a water quality prediction device, comprising:

[0012] The data acquisition unit is used to acquire the first historical monitoring datasets collected at different locations within the target area at various historical stages. Each first historical monitoring dataset contains first historical monitoring data from various data sources.

[0013] The shallow feature extraction unit is used to divide the first historical monitoring data corresponding to the data sources belonging to the same data type into a group of homogeneous data, and extract the intra-class correlation features of each group of homogeneous data. Each intra-class correlation feature represents the degree of influence of homogeneous data collected from surrounding locations on the water quality of the corresponding location.

[0014] The deep feature extraction unit is used to perform deep feature extraction on the intra-class related features to obtain the corresponding water quality prediction features. The water quality prediction features represent the first historical monitoring datasets collected at each location in each historical stage, as well as the first historical monitoring datasets collected at the corresponding surrounding locations in each historical stage, and the degree of influence on predicting the water quality of the corresponding location in the future stage.

[0015] The prediction unit is used to determine the water quality prediction results for each location within the target area based on the obtained water quality prediction characteristics.

[0016] Optionally, the deep feature extraction unit is used for:

[0017] Through the processing units in the time recurrent neural network, the corresponding inter-class association features of each dimension in the inter-class association features are read, wherein...

[0018] For each one-dimensional inter-class correlation feature read, based on the one-dimensional inter-class correlation feature and the one-dimensional water quality prediction feature output by the previous processing unit in the time recurrent neural network, feature filtering and feature updating are performed on the one-dimensional inter-class correlation feature read by the current processing unit, and the current processing unit generates the corresponding one-dimensional water quality prediction feature based on the updated one-dimensional inter-class correlation feature.

[0019] Optionally, the target water quality prediction model includes multiple homogeneous data convolutional networks, feature fusion networks, and temporal recurrent neural networks;

[0020] The water quality prediction device further includes a model training unit, which trains the target water quality prediction model by performing the following steps:

[0021] Obtain a training sample set for the target area, where each training sample contains a second historical monitoring dataset collected at various locations in the target area during historical periods;

[0022] The general water quality prediction model is trained twice using a training sample set for the target region in a cyclic iterative manner until the general water quality prediction model meets the stopping condition of the iterative training. The general water quality prediction model output from the last round of iterative training is used as the target water quality prediction model after training is completed. The general water quality prediction model is obtained by training on the training sample set of each region.

[0023] Optionally, during the secondary training of the general water quality prediction model, the iterative training process of the model training unit includes:

[0024] Read training samples of a target region, divide the training samples of the target region into multiple groups of homogeneous data according to the data type of the data source, and use a convolutional network of homogeneous data of the corresponding data type to extract the corresponding intra-class association features from each group of homogeneous data.

[0025] The feature fusion network is used to fuse the intra-class correlation features to obtain the corresponding inter-class correlation features, and the temporal recurrent neural network is used to perform deep feature extraction on the inter-class correlation features to obtain the water quality prediction results for each location in the target area.

[0026] Based on the water quality prediction results and the actual water quality assessment results of each location in the target area, the network parameters of each network in the target water quality prediction model are adjusted a second time.

[0027] Optionally, before acquiring historical monitoring datasets collected at different locations within the target area at various historical stages, the acquisition unit is further configured to:

[0028] The target area is divided into multiple locations based on geographical latitude and longitude.

[0029] At each of the locations, multiple sensor devices were deployed to acquire the historical monitoring datasets collected at each of the respective locations during the respective historical periods.

[0030] Thirdly, embodiments of this application also provide a computer device, including a processor and a memory, wherein the memory stores program code, and when the program code is executed by the processor, the processor performs the steps of any of the above-described water quality prediction methods.

[0031] Fourthly, embodiments of this application also provide a computer-readable storage medium including program code, which, when the program product is run on a computer device, is used to cause the computer device to perform the steps of any of the above-described water quality prediction methods.

[0032] Fifthly, embodiments of this application also provide a computer program product, including computer instructions, which are executed by a processor using the steps of any of the above-described water quality prediction methods.

[0033] The beneficial effects of this application are as follows:

[0034] This application provides a water quality prediction method, apparatus, device, and storage medium. The method includes:

[0035] The first historical monitoring datasets collected at each location within the target area are obtained. Each first historical monitoring dataset contains first historical monitoring data from different data sources. Through shallow feature extraction, intra-class correlation features representing the impact of first historical monitoring data of the same data type collected at surrounding locations on the water quality of the corresponding locations are obtained. Then, through deep feature extraction, corresponding water quality prediction features are extracted from each intra-class correlation feature. Based on the obtained water quality prediction features, the water quality prediction results for each location are obtained.

[0036] In this embodiment, by employing both shallow and deep feature extraction, the influence of first historical monitoring data from different data sources on the prediction of water quality at each location is considered, as well as the influence of first historical monitoring datasets collected from surrounding locations on the prediction of water quality at the corresponding locations. Therefore, by fully considering the spatial and temporal correlations of the data, and by making full use of the collected multi-source heterogeneous data, more expressive intra-class correlation features are extracted, significantly improving the accuracy of water quality prediction.

[0037] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0038] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0039] Figure 1 This is an optional schematic diagram of an application scenario in the embodiments of this application;

[0040] Figure 2a This is a schematic diagram of the structure of the target water quality prediction model in the embodiments of this application;

[0041] Figure 2b A schematic flowchart illustrating the water quality prediction method provided in the embodiments of this application;

[0042] Figure 2c This is a logical diagram illustrating how a gridded target area is processed in this embodiment of the application to obtain the first historical monitoring dataset for each location within that area.

[0043] Figure 2d This is a schematic diagram illustrating the logic of convolutional layer i performing convolution processing on the output data in an embodiment of this application.

[0044] Figure 2e A schematic diagram comparing the data processing of a conventional convolutional network and a homogeneous data convolutional network provided in an embodiment of this application;

[0045] Figure 2f A schematic diagram illustrating the process of obtaining water quality prediction features using a time-recurrent neural network, as provided in an embodiment of this application.

[0046] Figure 2g This is a logical diagram illustrating the acquisition of intra-class association features and inter-class association features in the embodiments of this application;

[0047] Figure 3a This is a schematic diagram of the internal structure of the LSTM in the embodiments of this application;

[0048] Figure 3b This is a logical diagram illustrating the generation of a one-dimensional water quality prediction feature in a cell in an embodiment of this application.

[0049] Figure 4a This is a logical diagram illustrating the training of the target water quality prediction model in an embodiment of this application;

[0050] Figure 4b This is a flowchart illustrating the process of training the target water quality prediction model in an embodiment of this application.

[0051] Figure 4c This is a schematic diagram of one round of iterative training of the target water quality prediction model in the embodiments of this application;

[0052] Figure 5 This is a schematic diagram of the structure of a water quality prediction device provided in an embodiment of this application;

[0053] Figure 6 This is a schematic diagram of the composition structure of a computer device provided in the embodiments of this application;

[0054] Figure 7 This is a schematic diagram of the structure of a computing device in an embodiment of this application. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this application. Obviously, the described embodiments are only some embodiments of the technical solutions of this application, and not all embodiments. Based on the embodiments recorded in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the technical solutions of this application.

[0056] The following explanations of some terms used in the embodiments of this application are provided to facilitate understanding by those skilled in the art.

[0057] 1. Cloud technology:

[0058] This application relates to the field of cloud technology, which refers to a hosting technology that unifies a series of resources such as hardware, software, and networks within a wide area network or local area network to realize the computation, storage, processing, and sharing of data.

[0059] Specifically, cloud technology is a general term for network technology, information technology, integration technology, management platform technology, and application technology that apply the cloud computing business model. The resource pool composed of cloud technology can be used on demand, making it more flexible and convenient.

[0060] Cloud computing technology will become a crucial support for cloud technology, primarily addressing the need for substantial computing and storage resources in the backend services of cloud network systems. These cloud network systems include, but are not limited to, video websites, image websites, and various portal websites. With the rapid development and application of the internet industry, every item in the future may possess its own unique identifier, requiring transmission to a backend system for logical processing. Different levels of data will be processed separately, and various industry data will all require robust system support, which can only be achieved through cloud computing.

[0061] 2. Cloud computing:

[0062] In a narrow sense, cloud computing refers to the delivery and usage model of information technology (IT) infrastructure, which is to obtain the required resources in an on-demand and easily scalable manner through the network; while in a broad sense, cloud computing refers to the delivery and usage model of services, which is to obtain the required services in an on-demand and easily scalable manner through the network. These services can be IT and software, Internet-related, or other services.

[0063] Cloud computing is a product of the convergence of traditional computer and network technologies such as grid computing, distributed computing, parallel computing, utility computing, network storage technologies, virtualization, and load balancing. Driven by the development of the internet, real-time data streams, the diversification of connected devices, and the demands of search services, social networks, mobile commerce, and open collaboration, cloud computing has rapidly developed. Unlike previous parallel and distributed computing, the emergence of cloud computing has revolutionized the entire internet model and enterprise management model from a conceptual perspective.

[0064] 3. Artificial Intelligence Cloud Services:

[0065] Artificial intelligence cloud services, also known as AI as a Service (AIaaS), are currently a mainstream service model for artificial intelligence platforms.

[0066] Specifically, AIaaS platforms break down several common AI services and provide them as standalone or packaged services in the cloud. This service model is similar to opening an AI-themed marketplace: all developers can access and use one or more AI services provided by the platform through API interfaces, and some experienced developers can also use the AI ​​framework and AI infrastructure provided by the platform to deploy and maintain their own dedicated cloud AI services.

[0067] 4. Surface water:

[0068] Also known as terrestrial water, it refers to water that exists on the Earth's surface and is exposed to the atmosphere. It is a general term for four types of water bodies: rivers, glaciers, lakes, and swamps. Surface water is one of the important sources of water for human life and a major component of water resources in various countries.

[0069] 5. Latitude and longitude:

[0070] It is a collective term for longitude and latitude. Longitude refers to the degree measure of a location on Earth east or west of a north-south line called the Prime Meridian; latitude refers to the angle between the line connecting a point to the Earth's center and the Earth's equatorial plane, with a value between 0 and 90 degrees. Together, they form a geographic coordinate system that can mark any location on Earth.

[0071] 6. Recurrent Neural Networks:

[0072] Recurrent neural networks are a branch of neural networks. Specifically, recurrent neural networks can be divided into temporal recurrent neural networks and structural recurrent neural networks (also known as recurrent neural networks, RNNs).

[0073] Temporal recurrent neural networks (RNNs) process time-series data by iteratively transmitting and processing the data flow. This demonstrates that RNNs possess both persistence and memory, enabling them to describe dynamic temporal behavior and addressing the gradient vanishing or exploding problems that occur in RNNs due to increasing recursion counts. Consequently, they are widely used in machine translation, speech recognition, personalized recommendations, and many other fields.

[0074] 7. Long Short-Term Memory (LSTM) Network:

[0075] LSTM belongs to the category of temporal recurrent neural networks and is mainly used to solve the problems of gradient vanishing and gradient explosion during long sequence training. It is widely used in tasks such as machine translation, image analysis, document summarization, speech recognition, image recognition, robot control, and music synthesis.

[0076] Unlike the RNN network architecture, LSTM adds a "processor" to the network architecture to determine whether the data information is useful. This processor is also called a cell.

[0077] A cell contains three gates, called the forget gate, input gate, and output gate, which correspond to the three important stages inside the LSTM: the forget stage, the selective memory stage, and the output stage.

[0078] For example, when a piece of data enters an LSTM cell, the cell can determine whether the data is useful based on rules. The input gate only retains data that meets the rule requirements, while the forget gate forgets data that does not meet the rule requirements. Finally, the output gate determines which data will be used as the output of the current state.

[0079] To make it easier to remember, the working principle of LSTM can be summarized as "one in, two out". Although the working principle of LSTM is not difficult to understand, it can solve a long-standing problem in RNNs through repeated computation.

[0080] Currently, a large amount of experimental data has demonstrated that LSTM is one of the effective techniques for solving long sequence dependency problems. Moreover, LSTM has very high versatility, and researchers have proposed various LSTM-based variant networks, enabling the use of LSTM to handle a wide range of problems.

[0081] 8. Transfer learning:

[0082] Transfer learning is a major class of learning methods in machine learning, which refers to the reuse of a model trained on one task on another related task.

[0083] Since most data or tasks are correlated, transfer learning algorithms can be used to transfer learned network parameters from related tasks and share them with the neural network that needs to learn. In this way, the new neural network does not need to learn from scratch like most neural networks, but can learn in a targeted manner based on a certain foundation, thus accelerating and optimizing the learning efficiency of the new neural network.

[0084] The design concept of the embodiments of this application is briefly introduced below:

[0085] Surface water is one of the most important sources of drinking water for humans; therefore, the quality of surface water environment has a significant impact on people's daily lives and production. Currently, water quality prediction methods based on hydrodynamics and water quality prediction methods based on machine learning are commonly used to predict the water quality of surface water.

[0086] Natural evaporation of surface water and fluid reactions between surface water and pipelines can both cause changes in surface water quality. However, in water quality prediction methods based on hydrodynamics, it is difficult to construct mathematical models to simulate changes in surface water quality, making it impossible to accurately predict water quality conditions.

[0087] In machine learning-based water quality prediction methods, either only water quality data is considered, which limits the application scenarios of the water quality prediction model, and the model may gradually become ineffective as factors such as climate and environment change, making it unable to accurately assess the water quality of surface water; or only historical monitoring data of a single water quality monitoring station are considered, making it difficult for the water quality prediction model to accurately assess the water quality of the area where the station is located.

[0088] To address the technical problems existing in the aforementioned related technologies, this application provides a water quality prediction method. The method includes:

[0089] The first historical monitoring dataset collected at each location within the target area is obtained. Each first historical monitoring dataset contains first historical monitoring data from different data sources. Through shallow feature extraction, intra-class association features representing the impact of first historical monitoring data of the same data type collected at surrounding locations on the water quality of the corresponding locations are obtained. Then, through deep feature extraction, corresponding water quality prediction features are further extracted from each intra-class association feature. These water quality prediction features represent the impact of the first historical monitoring datasets collected at each location at each historical stage, as well as the first historical monitoring datasets collected at the corresponding surrounding locations at each historical stage, on the predicted water quality of the corresponding locations in the future stage.

[0090] In this embodiment, by employing shallow and deep feature extraction, the influence of first historical monitoring data from different data sources on the prediction of water quality at each location is considered, as well as the influence of first historical monitoring datasets collected from surrounding locations on the prediction of water quality at the corresponding locations. Therefore, by fully considering the spatial and temporal correlation of the data, the water quality prediction method proposed in this embodiment significantly improves the accuracy of water quality prediction.

[0091] The preferred embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit this application. Furthermore, the embodiments and features in the embodiments of this application can be combined with each other without conflict.

[0092] The embodiments of this application can be applied to various scenarios, including but not limited to cloud technology, artificial intelligence, smart transportation, and assisted driving. Figure 1 The diagram illustrates one application scenario, which includes two physical terminal devices 110 and a server 130. The physical terminal devices 110 and the server 130 establish a communication connection using a wired or wireless network.

[0093] In this embodiment, the physical terminal device 110 is a computer device used by the user. The computer device can be a personal computer, mobile phone, tablet computer, laptop, e-book reader, smart home device, or other computer device with certain computing capabilities.

[0094] The server 130 in this application embodiment can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms. This application does not impose any restrictions on these services.

[0095] Users upload the first historical monitoring datasets collected at different locations in the target area at various historical stages to the server 130 through the access interface 120 of the physical terminal device 110. A data intelligence algorithm platform is deployed on the server 130, which contains multiple models for performing different tasks. The server 130 calls the target water quality prediction model of the data intelligence algorithm platform, inputs the first historical monitoring datasets collected at different locations in various historical stages uploaded by the physical terminal device 110 into the target water quality prediction model, and obtains the water quality prediction results for each location in the target area after multiple feature extractions within the model.

[0096] Figure 2a The diagram shows the structure of the target water quality prediction model, which consists of three parts: multiple homogeneous data convolutional networks, feature fusion networks, and temporal recurrent neural networks.

[0097] Homogeneous data convolutional networks are convolutional networks consisting of multiple convolutional layers, each composed of multiple convolutional kernels. The size of the convolutional layers is set according to actual needs; for example, convolutional layer 1 consists of c1 convolutional kernels of size m1*m1, and convolutional layer 2 consists of c2 convolutional kernels of size m2*m2.

[0098] Homogeneous data convolutional networks can employ any one or a combination of convolutional networks with different structures, such as Resnext and ResNest, to perform shallow feature extraction on the first historical monitoring data of the same data type in order to obtain the corresponding intra-class correlation features.

[0099] The feature fusion network is also a convolutional network used to fuse the obtained intra-class correlation features to obtain the corresponding inter-class correlation features. These features can characterize the degree of influence of the first historical monitoring dataset collected at each location and the first historical monitoring dataset collected at the corresponding surrounding locations on the water quality of the corresponding location.

[0100] Temporal recurrent neural networks can describe dynamic temporal behavior. LSTM, Gated Recurrent Unit (GRU), and other temporal recurrent neural networks can be used. By performing deep feature extraction on inter-class correlation features that integrate intra-class correlation features, a corresponding water quality prediction model can be obtained. This feature can characterize the first historical monitoring dataset collected at each location in each historical period, as well as the first historical monitoring dataset collected at the corresponding surrounding locations in each historical period, and the degree of influence on predicting the water quality of the corresponding location in the future.

[0101] After introducing the model architecture of the target water quality prediction model, please refer to... Figure 2b The flowchart shown will further explain the process of using the target water quality prediction model.

[0102] S201: Obtain the first historical monitoring datasets collected at different locations within the target area at various historical stages. Each first historical monitoring dataset contains first historical monitoring data from various data sources.

[0103] Before performing step 201, the target area is first gridded according to geographical latitude and longitude, so that it is divided into multiple locations;

[0104] The water quality at each location is affected by a variety of factors, including meteorological conditions, air quality changes, local climate, and geographical location, and exhibits non-linear changes. Since the amount of information contained in water quality data alone is very limited, in order to collect other data that may affect water quality, multiple sensor devices were deployed at each location to obtain historical monitoring datasets collected at each historical stage. Each first historical monitoring dataset contains first historical monitoring data from various data sources.

[0105] The data sources in this application embodiment include, but are not limited to, pH value data, ammonia nitrogen concentration data, permanganate concentration data, fine particulate matter (PM2.5) data, sulfur dioxide (SO2) data, carbon monoxide (CO) data, wind direction, and wind speed.

[0106] As shown in 2c, the thumbnail map of Beijing North City was gridded according to geographical latitude and longitude, resulting in multiple locations. At each location, multiple sensor devices were deployed to collect water quality data such as pH value data, ammonia nitrogen concentration data, and permanganate concentration data, as well as sensor devices to collect air quality data such as PM2.5 data, SO2 data, and CO data.

[0107] Perform the following operations for each location separately to generate the first historical monitoring dataset for that location from the previous day:

[0108] The initial historical monitoring data collected by each sensor device from a location the previous day is obtained separately. All the initial historical monitoring data are divided into multiple groups according to the data source type. In this way, each group will contain initial historical monitoring data from one or more data sources from the same data source.

[0109] The median value was calculated for each group of historical monitoring data, and the result of each calculation was used as the first historical monitoring data collected at that location corresponding to a certain data source.

[0110] After performing the above operations at each location, you can obtain the following results: Figure 2c The first historical monitoring dataset for each data source is shown.

[0111] S202: Divide the first historical monitoring data corresponding to data sources belonging to the same data type into a group of homogeneous data, and extract the intra-class correlation features of each group of homogeneous data. Each intra-class correlation feature represents the degree of influence of the first historical monitoring data collected from surrounding locations on the water quality of the corresponding location.

[0112] After gridding, both the temporal correlation between multi-source data from the same location and the geospatial correlation between multi-source data from different locations are preserved. To fully explore the multi-source data collected from various locations at different historical stages, multiple homogeneous data convolutional networks are used for shallow feature extraction. The specific process is as follows:

[0113] First, the first historical monitoring data corresponding to data sources belonging to the same data type are divided into a group of homogeneous data.

[0114] Data sources can include, but are not limited to, water quality data, air quality data, and meteorological data. First historical monitoring data belonging to the same data type are called homogeneous data, while first historical monitoring data belonging to different data types are called heterogeneous data.

[0115] For example, pH data, ammonia nitrogen concentration data, and permanganate concentration data belong to the same type of water quality data, while PM2.5 data, SO2 data, and CO data belong to the same type of air quality data.

[0116] After dividing the data into multiple groups of homogeneous data, perform the following operations for each group of homogeneous data:

[0117] An iterative approach is adopted, in which a set of homogeneous data is sequentially input into multiple convolutional layers of a homogeneous data convolutional network of the same data type for feature extraction, and the fused feature output by the last convolutional layer is used as the corresponding intra-class association feature;

[0118] In this process, after each convolutional layer i receives the output data from the previous convolutional layer, such as... Figure 2d As shown, multiple convolutional kernels contained in convolutional layer i are used to perform convolution processing on the output data respectively, and the obtained convolutional features are fused within the class to obtain the corresponding fused features. This enables the homogeneous data convolutional network to extract fused features that characterize the water quality of each location and the degree of influence of the homogeneous data of the surrounding locations based on the geospatial correlation between the homogeneous data of the surrounding locations and the homogeneous data of the corresponding locations.

[0119] When convolutional layer i is the first convolutional layer, the received output data is the first historical monitoring data in this set of homogeneous data;

[0120] When convolutional layer i is not the first convolutional layer, the received output data is the fused feature of the output of the previous convolutional layer.

[0121] Although homogeneous data convolutional networks belong to the category of convolutional networks, such as Figure 2e As shown, unlike the full-channel convolution operation of conventional convolutional networks (left figure), each homogeneous data convolutional network (right figure) only performs group convolution operations on homogeneous data of the same data type in order to obtain inter-class correlation features with stronger expressive power, which is beneficial to improving the accuracy of water quality prediction.

[0122] S203: Deep feature extraction is performed on the intra-class correlation features to obtain the corresponding water quality prediction features. The water quality prediction features represent the first historical monitoring datasets collected at each location at each historical stage, as well as the first historical monitoring datasets collected at the corresponding surrounding locations at each historical stage, and the degree of influence on the prediction of the water quality of the corresponding locations in the future stage.

[0123] See Figure 2f The flowchart shown illustrates the process of obtaining water quality prediction features using a time recurrent neural network:

[0124] S2031: Perform feature fusion on the intra-class correlation features to obtain the corresponding inter-class correlation features. The inter-class correlation features represent the degree of influence of the first historical monitoring dataset collected at each location and the first historical monitoring dataset collected at the corresponding surrounding locations on the water quality of the corresponding location.

[0125] As shown in 2g, homogeneous air quality data is input into two convolutional layers of homogeneous data convolutional network 1 for shallow feature extraction to obtain intra-class correlation feature 1, and homogeneous water quality data is input into two convolutional layers of homogeneous data convolutional network 2 for shallow feature extraction to obtain intra-class correlation feature 2; then, the two intra-class correlation features are fused by a 1*1 dot product convolutional layer to obtain the corresponding inter-class correlation feature.

[0126] In this embodiment, the target water quality prediction model uses multiple homogeneous data convolutional networks for shallow feature extraction and a feature fusion network to fuse various intra-class correlation features, so that the model can simultaneously consider intra-class correlation between homogeneous data and inter-class correlation between heterogeneous data, thereby obtaining inter-class correlation features with stronger expressive power, which is beneficial to improving the accuracy of water quality prediction.

[0127] S2032: Input the inter-class correlation features into the time recurrent neural network for deep feature extraction to obtain the corresponding water quality prediction features.

[0128] By using each processing unit (i.e., cell) in the time recurrent neural network, the corresponding inter-class association features in each dimension are read from the inter-class association features, where...

[0129] For each one-dimensional inter-class correlation feature read, based on the one-dimensional inter-class correlation feature and the one-dimensional water quality prediction feature output by the previous processing unit in the time recurrent neural network, feature filtering and feature updating are performed on the one-dimensional inter-class correlation feature read by the current processing unit, and the current processing unit generates the corresponding one-dimensional water quality prediction feature based on the updated one-dimensional inter-class correlation feature.

[0130] like Figure 3a As shown, each cell of the LSTM contains three gates: a forget gate, an input gate, an output gate, and a tanh layer. By utilizing the gate structure to eliminate or add information to the cell, the entire LSTM can remember long-term information.

[0131] See Figure 3b The diagram shown illustrates the data processing procedure inside a cell.

[0132] First, the one-dimensional inter-class association feature x at the current time... t The one-dimensional water quality prediction feature h output by the previous cell t-1 All data is input into the forget gate, which determines the memory cell state C of the previous cell. t-1 Which state information is retained and which is forgotten? Using Formula 1, we can obtain the memory cell state C of the previous cell. t-1 The probability f of being forgotten t ;f t =1 indicates that f is completely retained. t =0 indicates that it is completely discarded.

[0133]

[0134] Next, the one-dimensional inter-class association feature x at the current time will be... t The one-dimensional water quality prediction feature h output by the previous cell t-1All information is input into the input gate, which determines which features can be saved to the current memory cell state C'. t First, use Formula 2 to obtain the memory cell state C where the feature information is saved to the current cell. t The probability i t Then, using Formula 3, generate the candidate memory cell state C' for the current cell. t .

[0135]

[0136] C' t =tanh(W C *[h t-1 ,x t ]+b C ) Formula 3;

[0137] As shown in Equation 4, the memory cell state C of the current cell is determined simultaneously by the forget gate and the input gate. t Status information.

[0138] C t =f t *C t-1 +i t *C' t Formula 4;

[0139] Finally, as shown in Equations 5 and 6, the one-dimensional inter-class correlation feature x at the current time is... t The one-dimensional water quality prediction feature h output by the previous cell t-1 All inputs are fed into the output gate, which determines the current cell's memory state C. t Which state information is used as the one-dimensional water quality prediction feature h in the current cell's output? t .

[0140] Wherein, C represents the memory cell state of the current cell. t The probability of being retained is o t .

[0141]

[0142] h t =o t *tanh(C t ) Formula 6;

[0143] However, in addition to using a time recurrent neural network to extract the temporal correlation between data, this application embodiment also supports using an attention-based neural network to extract the temporal correlation between data.

[0144] S204: Based on the obtained water quality prediction characteristics, determine the water quality prediction results for each location within the target area.

[0145] In the embodiments of this application, regression problems or classification problems can be established according to the actual needs of the scenario.

[0146] For regression problems, the obtained water quality prediction features are used as input to the fully connected layer. After the fully connected layer integrates the water quality prediction features, the corresponding prediction label (e.g., the pH value of the predicted water quality) is output.

[0147] For classification problems, the obtained water quality prediction features are used as input to the fully connected layer, which then performs feature integration processing and mapping processing in sequence. The output is the predicted probability of water quality belonging to each category (e.g., normal water quality, slightly polluted, moderately polluted, heavily polluted).

[0148] After introducing the usage process of the target water quality prediction model, we will continue to introduce the training process of the target water quality prediction model.

[0149] The amount of effective data that can be collected in a single city is relatively small, and the main factors affecting the water quality of each city are also different. As a result, water quality prediction models trained on datasets from a single city do not have universality.

[0150] Therefore, refer to Figure 4a The schematic diagram shown illustrates that, according to this application embodiment, the idea of ​​transfer learning is used to first train a general water quality prediction model with generalization ability based on datasets from multiple cities, and then train the general water quality prediction model a second time based on partial data from the target city to obtain a target water quality prediction model for the target city.

[0151] The training process of the target water quality prediction model is as follows: Figure 4b As shown, it specifically includes:

[0152] S401: Obtain a training sample set for the target area. Each training sample contains a second historical monitoring dataset collected from various locations in the target area during historical periods.

[0153] S402: Using an iterative approach, the general water quality prediction model is trained twice using the training sample set for the target area until the general water quality prediction model meets the stopping condition of iterative training. The general water quality prediction model output from the last round of iterative training is used as the target water quality prediction model after training. The general water quality prediction model is trained based on the training sample set of each area.

[0154] One round of iterative training is as follows: Figure 4c As shown, it specifically includes:

[0155] S4021: Read training samples of a target region, divide the training samples of a target region into multiple groups of homogeneous data according to the data type of the data source, and use a convolutional network of homogeneous data of the corresponding data type to extract the corresponding intra-class association features from each group of homogeneous data.

[0156] S4022: Use a feature fusion network to fuse intra-class correlation features to obtain corresponding inter-class correlation features, and use a temporal recurrent neural network to perform deep feature extraction on the inter-class correlation features to obtain the water quality prediction results for each location in the target area.

[0157] S4023: Based on the water quality prediction results and the actual water quality assessment results of each location in the target area, adjust the network parameters of each network in the target water quality prediction model for the second time.

[0158] However, in addition to using the idea of ​​transfer learning to obtain a target water quality prediction model for the target city, the embodiments of this application also support the customized development of the model. First, a target water quality prediction model is built for the actual business scenario, and then the target water quality prediction model is trained based on the business data in that scenario.

[0159] For example, based on the business data on sewage and exhaust gas from concentrated and non-concentrated industrial zones in urban areas provided by the environmental protection department, a sewage prediction model for industrial zones can be trained. Then, the trained model can be used to obtain the sewage assessment results for each industrial zone, and based on the sewage assessment results, each industrial zone can be managed in a hierarchical and zoned manner.

[0160] Based on the same inventive concept as the above-described method embodiments, this application also provides a water quality prediction device. (See also...) Figure 5 The schematic diagram shown indicates that the water quality prediction device 500 may include:

[0161] The acquisition unit 501 is used to acquire the first historical monitoring datasets collected at different locations within the target area at various historical stages. Each first historical monitoring dataset contains first historical monitoring data from various data sources.

[0162] The shallow feature extraction unit 502 is used to divide the first historical monitoring data corresponding to the data source belonging to the same data type into a group of homogeneous data, and extract the intra-class correlation features of each group of homogeneous data. Each intra-class correlation feature represents the degree of influence of homogeneous data collected from surrounding locations on the water quality of the corresponding location.

[0163] The deep feature extraction unit 503 is used to perform deep feature extraction on the intra-class related features to obtain the corresponding water quality prediction features. The water quality prediction features represent the first historical monitoring datasets collected at each location at each historical stage, as well as the first historical monitoring datasets collected at the corresponding surrounding locations at each historical stage, and the degree of influence on predicting the water quality of the corresponding locations in the future stage.

[0164] Prediction unit 504 is used to determine the water quality prediction results for each location within the target area based on the obtained water quality prediction characteristics.

[0165] Optionally, the shallow feature extraction unit 502 is used for:

[0166] For each group of homogeneous data, perform the following operations respectively:

[0167] An iterative approach is adopted, in which a set of homogeneous data is sequentially input into multiple convolutional layers in a homogeneous data convolutional network for feature extraction, and the fused feature output by the last convolutional layer is used as the corresponding intra-class association feature. Specifically, after each convolutional layer receives the output data of the previous convolutional layer, multiple convolutional kernels contained in the convolutional layer are used to perform convolution processing on the output data respectively, and the obtained convolutional features are fused to obtain the corresponding fused feature.

[0168] Optionally, each output data point can be either the first historical monitoring data point from a set of homogeneous data, or a fused feature from the output of the previous convolutional layer.

[0169] Optionally, the deep feature extraction unit 503 is used for:

[0170] Feature fusion is performed on the intra-class correlation features to obtain the corresponding inter-class correlation features. The inter-class correlation features represent the degree of influence of the first historical monitoring dataset collected at each location and the first historical monitoring dataset collected at the corresponding surrounding locations on the water quality of the corresponding location.

[0171] Inter-class correlation features are input into a time recurrent neural network for deep feature extraction to obtain corresponding water quality prediction features.

[0172] Optionally, the deep feature extraction unit 503 is used for:

[0173] By using the processing units in the time-recurrent neural network, the corresponding inter-class association features in each dimension are read from the inter-class association features.

[0174] For each one-dimensional inter-class correlation feature read, based on the one-dimensional inter-class correlation feature and the one-dimensional water quality prediction feature output by the previous processing unit in the time recurrent neural network, feature filtering and feature updating are performed on the one-dimensional inter-class correlation feature read by the current processing unit, and the current processing unit generates the corresponding one-dimensional water quality prediction feature based on the updated one-dimensional inter-class correlation feature.

[0175] Optionally, the target water quality prediction model includes multiple homogeneous data convolutional networks, feature fusion networks, and temporal recurrent neural networks;

[0176] The water quality prediction device 500 also includes a model training unit 505, which trains the target water quality prediction model by performing the following steps:

[0177] Obtain a training sample set for the target area, where each training sample contains a second historical monitoring dataset collected from various locations in the target area during historical periods.

[0178] The general water quality prediction model is trained twice using a training sample set for the target region in a cyclical iterative manner until the general water quality prediction model meets the stopping condition of iterative training. The general water quality prediction model output from the last round of iterative training is used as the target water quality prediction model after training is completed. The general water quality prediction model is trained based on the training sample set of each region.

[0179] Optionally, during the secondary training of the general water quality prediction model, the iterative training process of model training unit 505 includes:

[0180] Read training samples from a target region, divide the training samples from the target region into multiple groups of homogeneous data according to the data type of the data source, and use a convolutional network of homogeneous data of the corresponding data type to extract the corresponding intra-class association features from each group of homogeneous data.

[0181] The feature fusion network is used to fuse the intra-class correlation features to obtain the corresponding inter-class correlation features. Then, a time recurrent neural network is used to perform deep feature extraction on the inter-class correlation features to obtain the water quality prediction results for each location in the target area.

[0182] Based on the water quality prediction results and the actual water quality assessment results of each location in the target area, the network parameters of each network in the target water quality prediction model are adjusted for the second time.

[0183] Optionally, before acquiring historical monitoring datasets collected at different locations within the target area at various historical stages, the acquisition unit 501 is also used for:

[0184] The target area is divided into multiple locations based on geographical latitude and longitude.

[0185] Multiple sensor devices were deployed at various locations to obtain historical monitoring datasets collected at different historical stages.

[0186] For ease of description, the above sections are divided into modules (or units) according to their functions and described separately. Of course, in implementing this application, the functions of each module (or unit) can be implemented in one or more software or hardware components.

[0187] Having introduced the service platform access method and apparatus according to exemplary embodiments of this application, the computer device according to another exemplary embodiment of this application will now be described.

[0188] Those skilled in the art will understand that various aspects of this application can be implemented as a system, method, or program product. Therefore, various aspects of this application can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software implementations, collectively referred to herein as a "circuit," "module," or "system."

[0189] Based on the same inventive concept as the above-described method embodiments, this application also provides a computer device, see reference. Figure 6 As shown, the computer device 600 may include at least a processor 601 and a memory 602. The memory 602 stores program code, which, when executed by the processor 601, causes the processor 601 to perform the steps of any of the aforementioned water quality prediction methods.

[0190] In some possible implementations, the computing device according to this application may include at least one processor and at least one memory. The memory stores program code that, when executed by the processor, causes the processor to perform the steps in the water quality prediction methods according to the various exemplary embodiments of this application described above. For example, the processor may perform actions such as... Figure 2b The steps are shown in the figure.

[0191] The following reference Figure 7 To describe a computing device 700 according to this embodiment of the present application. Figure 6 The computing device 700 is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0192] like Figure 7As shown, the computing device 700 is presented in the form of a general-purpose computing device. The components of the computing device 700 may include, but are not limited to: at least one processing unit 701, at least one storage unit 702, and a bus 703 connecting different system components (including storage unit 702 and processing unit 701).

[0193] Bus 703 represents one or more of several bus architectures, including a memory bus or memory controller, peripheral bus, processor, or a local bus using any of the various bus architectures.

[0194] Storage unit 702 may include a readable medium in the form of volatile memory, such as random access memory (RAM) 7021 and / or cache memory unit 7022, and may further include read-only memory (ROM) 7023.

[0195] Storage unit 702 may also include a program / utility 7025 having a set (at least one) program module 7024, such program module 7024 including but not limited to: operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.

[0196] The computing device 700 can also communicate with one or more external devices 704 (e.g., keyboard, pointing device, etc.), one or more devices that enable a user to interact with the computing device 700, and / or any device that enables the computing device 700 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 705. Furthermore, the computing device 700 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 706. As shown, network adapter 706 communicates with other modules used in the computing device 700 via bus 703. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with the computing device 700, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0197] Based on the same inventive concept as the above-described method embodiments, various aspects of the water quality prediction method provided in this application can also be implemented as a program product, which includes program code. When the program product is run on a computer device, the program code is used to cause the computer device to perform the steps in the water quality prediction method according to the various exemplary embodiments of this application described above. For example, the computer device can perform actions such as... Figure 2bThe steps are shown in the figure.

[0198] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0199] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0200] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A water quality prediction method, characterized in that, include: Acquire the first historical monitoring datasets collected at various locations within the target area at various historical stages. Each first historical monitoring dataset contains first historical monitoring data from various data sources. The first historical monitoring data corresponding to data sources belonging to the same data type are divided into a group of homogeneous data, and the intra-class correlation features of each group of homogeneous data are extracted. Each intra-class correlation feature represents the degree of influence of homogeneous data collected from surrounding locations on the water quality of the corresponding location. Feature fusion is performed on the intra-class correlation features to obtain the corresponding inter-class correlation features. The inter-class correlation features represent the degree of influence of the first historical monitoring dataset collected at each location and the first historical monitoring dataset collected at the corresponding surrounding locations on the water quality of the corresponding location. Deep feature extraction is performed on the inter-class correlation features to obtain corresponding water quality prediction features. The water quality prediction features represent the first historical monitoring datasets collected at each location in each historical stage, as well as the first historical monitoring datasets collected at the corresponding surrounding locations in each historical stage, and the degree of influence on predicting the water quality of the corresponding locations in the future stage. Based on the obtained water quality prediction features, the water quality prediction results for each location within the target area are determined.

2. The method as described in claim 1, characterized in that, The step of extracting the intra-class association features of each group of homogeneous data includes: For each group of homogeneous data, perform the following operations respectively: An iterative approach is adopted, where a set of homogeneous data is sequentially input into multiple convolutional layers of a homogeneous data convolutional network for feature extraction, and the fused feature output by the last convolutional layer is used as the corresponding intra-class association feature. Each convolutional layer, upon receiving input data, uses multiple convolutional kernels to perform convolution processing on the input data, and then fuses the obtained convolutional features to obtain the corresponding fused feature. Specifically, when the convolutional layer is the first convolutional layer, the input data is a first historical monitoring data point from the set of homogeneous data; when the convolutional layer is not the first convolutional layer, the input data is the fused feature output by the previous convolutional layer.

3. The method as described in claim 1, characterized in that, The step of performing deep feature extraction on the inter-class correlation features to obtain corresponding water quality prediction features includes: The inter-class association features of each dimension are read from the inter-class association features through the processing units in the time recurrent neural network. For each one-dimensional inter-class correlation feature read, based on the one-dimensional inter-class correlation feature and the one-dimensional water quality prediction feature output by the previous processing unit in the time recurrent neural network, feature filtering and feature updating are performed on the one-dimensional inter-class correlation feature read by the current processing unit, and the current processing unit generates the corresponding one-dimensional water quality prediction feature based on the updated one-dimensional inter-class correlation feature.

4. The method as described in claim 3, characterized in that, Construct a target water quality prediction model that includes multiple homogeneous data convolutional networks, feature fusion networks, and temporal recurrent neural networks; The target water quality prediction model is obtained by training it through the following steps: Obtain a training sample set for the target area, where each training sample contains a second historical monitoring dataset collected at various locations in the target area during historical periods; The general water quality prediction model is trained twice using a training sample set for the target region in a cyclic iterative manner until the general water quality prediction model meets the stopping condition of the iterative training. The general water quality prediction model output from the last round of iterative training is used as the target water quality prediction model after training is completed. The general water quality prediction model is obtained by training on the training sample set of each region.

5. The method as described in claim 4, characterized in that, During the secondary training of the general water quality prediction model, one round of iterative training includes: Read training samples of a target region, divide the training samples of the target region into multiple groups of homogeneous data according to the data type of the data source, and use a convolutional network of homogeneous data of the corresponding data type to extract the corresponding intra-class association features from each group of homogeneous data. The feature fusion network is used to fuse the intra-class correlation features to obtain the corresponding inter-class correlation features, and the temporal recurrent neural network is used to perform deep feature extraction on the inter-class correlation features to obtain the water quality prediction results for each location in the target area. Based on the water quality prediction results and the actual water quality assessment results of each location in the target area, the network parameters of each network in the target water quality prediction model are adjusted a second time.

6. The method according to any one of claims 1 to 3, characterized in that, Before obtaining the first historical monitoring datasets collected at various locations within the target area at various historical stages, the following steps are also included: The target area is divided into multiple locations based on geographical latitude and longitude. At each of the locations, multiple sensor devices were deployed to acquire the first historical monitoring dataset collected at each of the respective locations during the respective historical periods.

7. A water quality prediction device, characterized in that, include: The data acquisition unit is used to acquire the first historical monitoring datasets collected at various locations within the target area at various historical stages. Each first historical monitoring dataset contains first historical monitoring data from various data sources. The shallow feature extraction unit is used to divide the first historical monitoring data corresponding to the data sources belonging to the same data type into a group of homogeneous data, and extract the intra-class correlation features of each group of homogeneous data. Each intra-class correlation feature represents the degree of influence of homogeneous data collected from surrounding locations on the water quality of the corresponding location. The deep feature extraction unit is used to perform feature fusion on the intra-class correlation features to obtain the corresponding inter-class correlation features. The inter-class correlation features represent the degree of influence of the first historical monitoring dataset collected at each location and the first historical monitoring dataset collected at the corresponding surrounding locations on the water quality of the corresponding location. Deep feature extraction is performed on the inter-class correlation features to obtain corresponding water quality prediction features. The water quality prediction features represent the first historical monitoring datasets collected at each location in each historical stage, as well as the first historical monitoring datasets collected at the corresponding surrounding locations in each historical stage, and the degree of influence on predicting the water quality of the corresponding locations in the future stage. The prediction unit is used to determine the water quality prediction results for each location within the target area based on the obtained water quality prediction characteristics.

8. The apparatus as claimed in claim 7, characterized in that, The shallow feature extraction unit is used for: For each group of homogeneous data, perform the following operations respectively: An iterative approach is adopted, where a set of homogeneous data is sequentially input into multiple convolutional layers of a homogeneous data convolutional network for feature extraction, and the fused feature output by the last convolutional layer is used as the corresponding intra-class association feature. Each convolutional layer, upon receiving input data, uses multiple convolutional kernels to perform convolution processing on the input data, and then fuses the obtained convolutional features to obtain the corresponding fused feature. Specifically, when the convolutional layer is the first convolutional layer, the input data is a first historical monitoring data point from the set of homogeneous data; when the convolutional layer is not the first convolutional layer, the input data is the fused feature output by the previous convolutional layer.

9. A computer device, characterized in that, It includes a processor and a memory, wherein the memory stores program code that, when executed by the processor, causes the processor to perform the steps of the method according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, It includes program code that, when the program product is run on a computer device, causes the computer device to perform the steps of the method according to any one of claims 1 to 6.

11. A computer program product, characterized in that, It includes computer instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 6.

Images