CN-116304908-B - Water bloom judging method based on space excavation

CN116304908BCN 116304908 BCN116304908 BCN 116304908BCN-116304908-B

Abstract

The invention relates to a water bloom judging method based on space excavation, which comprises the steps of respectively deploying sensors in N areas of a water area to acquire water body data in real time, wherein N is an integer greater than 1, the water body data comprise environmental parameters including water temperature, PH, turbidity, dissolved oxygen, conductivity, high-manganese acid salt index, ammonia nitrogen, total phosphorus and total ammonia, the sensors deployed in the areas to be detected in the water area are used as target sensors, the sensors deployed in the other areas in the water area are used as residual sensors, the water body data acquired by all the residual sensors are sequentially arranged to form a residual water body data sequence, the water body data acquired by the residual sensors and the residual water body data sequence are input into a trained water bloom judging model to predict whether the water bloom occurs in the areas to be detected in the water area, the accuracy of water bloom judgment is improved, emergency measures are timely taken by an environmental management department, and ecological hazards and health risks caused by the water bloom are reduced.

Inventors

YAO CHEN
LIU LI
CHEN LIN

Assignees

重庆邮电大学

Dates

Publication Date: 20260512
Application Date: 20230302

Claims (7)

1. The water bloom judging method based on space excavation is characterized by comprising the following steps of: S1, respectively deploying sensors in N areas of a water area to acquire water body data in real time, wherein N is an integer greater than 1, and the water body data comprise environmental parameters such as water temperature, PH, turbidity, dissolved oxygen, conductivity, high-humate index, ammonia nitrogen, total phosphorus and total ammonia; S2, taking a sensor deployed in a region to be detected in a water area as a target sensor, and taking sensors deployed in other regions in the water area as residual sensors; s3, sequentially arranging the water body data acquired by all the residual sensors to form a residual water body data sequence; S4, inputting the water body data and the residual water body data sequences acquired by the residual sensors into a trained water bloom judgment model to predict whether the water bloom occurs in the area to be detected in the water area; the predicting whether the water bloom occurs in the area to be measured in the water area comprises the following steps: S41 will be The water body data collected by the time object sensor is input into a local attention mechanism module, the local attention mechanism module is based on Hidden state of time of day LSTM network And memory state of the cell unit Weighting each parameter of the water body data to obtain A first intermediate feature matrix of the moment; S42, will be The time remaining water body data sequence is input into a global attention mechanism module, the global attention mechanism module is based on Hidden state of time of day LSTM network And memory state of the cell unit Weighting the water body data acquired by each other sensor to obtain A second intermediate feature matrix of the moment; S43, will First intermediate feature matrix sum of moments Splicing and generating a second intermediate feature matrix at moment in feature dimension A third intermediate feature matrix of the moment; S44, will be The third intermediate feature matrix of the moment is input into the LSTM network to be calculated Hidden state of time of day LSTM network And memory state of the cell unit ; S45, outputting the hidden state of the LSTM network at all moments in a preset time period And inputting the comprehensive feature vector into a full-connection layer for linear combination, and inputting the combined feature vector into a softmax classifier to output whether the region to be detected has water bloom or not.
2. The water bloom determination method based on space mining of claim 1, wherein the water bloom determination model comprises an LSTM network, a fully connected layer, a softmax classifier, a local attention mechanism module and a global attention mechanism module.
3. The water bloom determination method based on space mining as claimed in claim 1, wherein the water bloom determination method is characterized in that The first intermediate feature matrix of the time instant comprises: Wherein, the Representation of The first intermediate feature matrix of the time instant, 、、 And Representing a matrix of parameters that can be learned, The activation function is represented as a function of the activation, Representation of The hidden state of the LSTM network at the moment, Representation of The water body data collected by the time target sensor; representing the number of environmental parameters in the water body data acquired by the target sensor, Representing the presentation to be And Splicing the components to be spliced, Representation of Memory state of LSTM network cell unit at time.
4. The water bloom determination method based on space mining as claimed in claim 1, wherein the water bloom determination method is characterized in that The second intermediate feature matrix of the time instant includes: Wherein, the The activation function is represented as a function of the activation, 、、、、 And The parameters that are to be learned are represented, Represents a sequence of environmental parameters in the water body data, Representing the presentation to be And Splicing the components to be spliced, Representation of The memory state of the LSTM network cell unit at the moment, Representation of The hidden state of the LSTM network at the moment, Representing a data sequence of the remaining body of water, Indicating the number of remaining sensors.
5. The water bloom determination method based on space mining as claimed in claim 1, wherein the loss function of the water bloom determination model comprises: wherein L represents a loss function of the water bloom judgment model, Is a binary label, i.e. whether bloom occurs, a specific value is expressed as 0 or 1, and a is the probability of output belonging to the y label.
6. The method for judging water bloom based on space excavation according to claim 1, wherein the step of respectively disposing the sensors in N areas of the water area comprises dividing the water area into N areas, wherein the areas of the N areas are the same, and the sensors are disposed in the center of each area.
7. The water bloom determination method based on space mining as claimed in claim 1, wherein the sequentially arranging the water body data collected by all the remaining sensors comprises: and sequencing the water body data acquired by the residual sensors according to the distances from the central point of the rest areas in the water area to the central point of the area to be tested, wherein the sequencing is more forward when the distances are smaller.

Description

Water bloom judging method based on space excavation Technical Field The invention belongs to the technical field of environment detection, and particularly relates to a water bloom judging method based on space excavation. Background The eutrophication of lakes and the occurrence of water bloom are serious environmental problems commonly faced worldwide at present, the water bloom can destroy the erbium material foundation of a fishing ground, thereby reducing the yield of fishery, and when treating the water bloom, the basic rule of the water bloom formation and some physical and chemical factors influencing the water bloom formation must be known first to develop a sensitive lake area. Along with the rapid development of the deep learning field, the machine learning is applied to the basic law of the water bloom formation, whether the water bloom occurs in the current water area is judged, emergency measures are timely taken for environmental management departments, and the method has great significance in reducing the ecological harm and health risk caused by the water bloom. The existing water bloom judgment model mainly only considers the water body data of the characteristic data at the target site, but the data of each area in the water area are dynamically related due to the fluidity of the water area, and the water body condition of the current area can not be accurately expressed only by adopting the characteristic information on the target site, so that the water bloom judgment result of the model is inaccurate, and the management department is caused to make a decision to be wrong, thereby wasting resources. Disclosure of Invention In order to solve the problems in the background technology, the invention provides a water bloom judging method based on space excavation, so as to improve the accuracy of water bloom judgment, enable an environment management department to timely take emergency measures, reduce ecological harm and health risk caused by water bloom, and comprise the following steps: S1, respectively deploying sensors in N areas of a water area to acquire water body data in real time, wherein N is an integer greater than 1, and the water body data comprise environmental parameters such as water temperature, PH, turbidity, dissolved oxygen, conductivity, high-humate index, ammonia nitrogen, total phosphorus and total ammonia; S2, taking a sensor deployed in a region to be detected in a water area as a target sensor, and taking sensors deployed in other regions in the water area as residual sensors; s3, sequentially arranging the water body data acquired by all the residual sensors to form a residual water body data sequence; And S4, inputting the water body data and the residual water body data sequences acquired by the residual sensors into a trained water bloom judging model to predict whether the water bloom occurs in the region to be detected in the water area. Preferably, the water bloom judging model comprises an LSTM network, a full connection layer, a softmax classifier, a local attention mechanism module and a global attention mechanism module. Preferably, predicting whether the water bloom occurs in the area to be measured in the water area comprises: S41, inputting water body data acquired by a k-moment target sensor into a local attention mechanism module, wherein the local attention mechanism module obtains a first intermediate feature matrix at k moment by giving weight to each parameter of the water body data according to a hidden state h k-1 of an LSTM network at k-1 moment and a memory state S k-1 of a cell unit; S42, inputting the k-moment residual water body data sequence into a global attention mechanism module, and enabling the global attention mechanism module to give weight to water body data acquired by each other sensor according to a hidden state h k-1 of the k-1 moment LSTM network and a memory state S k-1 of the cell unit so as to obtain a k-moment second intermediate feature matrix; S43, splicing the first intermediate feature matrix at the moment k and the second intermediate feature matrix at the moment t on the feature dimension to generate a third intermediate feature matrix at the moment k; S44, inputting the third intermediate feature matrix at the k moment into the LSTM network to calculate to obtain a hidden state h k of the LSTM network at the k moment and a memory state S k of the cell unit; S45, splicing the hidden states h k output by the LSTM network at all times in the characteristic dimension within a preset time period to generate a comprehensive characteristic vector, inputting the comprehensive characteristic vector into a full-connection layer for linear combination, and inputting the comprehensive characteristic vector into a softmax classifier to output whether the region to be detected has water bloom. The invention has at least the following beneficial effects According to the invention, the influence of different learned environment