CN118966712B

CN118966712B - Intelligent control method and system for aquaculture based on machine learning

Info

Publication number: CN118966712B
Application number: CN202411378876.XA
Authority: CN
Inventors: 蔡玉刚; 彭滔; 杨涛; 彭淇
Original assignee: Hunan Junshan Ecological Fishery Group Co ltd
Current assignee: Hunan Junshan Ecological Fishery Group Co ltd
Priority date: 2024-09-30
Filing date: 2024-09-30
Publication date: 2025-03-07
Anticipated expiration: 2044-09-30
Also published as: CN118966712A

Abstract

The present invention discloses an intelligent control method and system for aquaculture based on machine learning, which relates to the field of aquaculture. The method comprises: obtaining water quality detection information by using a water body detection device; obtaining fish detection information by using an underwater image detection device; inputting water quality detection information and fish detection information into a target control model obtained based on machine learning to obtain target control information; wherein the distance between the predicted water quality obtained by inputting the target control information into the target water quality prediction model and the target water quality is less than a preset threshold, the target water quality is determined based on the fish detection information, and the target control model is obtained by joint training with the target water quality prediction model; according to the target control information, the working state of the intelligent breeding device is regulated, and the intelligent breeding device includes at least one of an oxygenating device, a feeding device, and a water circulation treatment device. It can effectively ensure that the regulated water quality can meet good breeding conditions and improve breeding benefits.

Description

Intelligent aquaculture regulation and control method and system based on machine learning

Technical Field

The invention relates to the field of aquaculture, in particular to an intelligent aquaculture regulation and control method and system based on machine learning.

Background

The uncontrollable factors of environmental factors in the traditional aquaculture process are more, so that the traditional aquaculture mode is eating by the day, the hidden danger of diseases is serious, and the yield mode is unstable.

Along with the development of science and technology, some intelligent breeding devices are favored by farmers, however, the control of these breeding devices is still relatively primary, for example, only the timing feeding of the feeding device, the timing starting of the oxygen charging machine and the like can be realized, and the requirements of the current more refined aquaculture cannot be met.

Disclosure of Invention

The embodiment of the invention provides an aquaculture intelligent regulation and control method and system based on machine learning.

In a first aspect, an embodiment of the present invention provides an intelligent aquaculture regulation and control method based on machine learning, where the method includes:

acquiring water quality detection information by utilizing a water body detection device, wherein the water quality detection information comprises at least one of water body temperature, pH value, oxygen content, nitrate content, nitrite content and feed residue proportion;

Acquiring fish school detection information by using an underwater image detection device, wherein the fish school detection information comprises at least one of fish school liveness, fish school quantity and average body length;

Inputting the water quality detection information and the fish shoal detection information into a target regulation and control model obtained based on machine learning to obtain target regulation and control information, wherein the distance between the predicted water quality obtained by inputting the target regulation and control information into a target water quality prediction model and the target water quality is smaller than a preset threshold value, the target water quality is determined based on the fish shoal detection information, and the target regulation and control model is obtained by combined training with the target water quality prediction model;

And regulating and controlling the working state of the intelligent culture device according to the target regulation and control information, wherein the intelligent culture device comprises at least one of an oxygenation device, a feeding device and a water circulation treatment device.

Optionally, the joint training of the target regulation model and the target water quality prediction model comprises the following steps:

Acquiring a target training set, wherein the target training set comprises a preprocessed historical regulation data sequence and a corresponding historical detection data sequence, the historical regulation data sequence comprises regulation information of a plurality of regulation periods on the intelligent culture device, and the historical detection sequence comprises water quality detection information and fish shoal detection information detected by the intelligent culture device in a preset time period after executing the regulation information;

extracting a plurality of pieces of regulation data from the historical regulation data sequence and extracting one piece of detection data from the historical detection data sequence to form sample data and labeling information, wherein the labeling information is used for indicating regulation data corresponding to the detection data in the sample data;

Training to obtain a target water quality prediction model based on the sample data, wherein the target water quality prediction model is used for predicting the regulated water quality according to regulation information of the intelligent culture device;

For each piece of detection data in the historical detection data sequence, determining a first target water quality according to fish shoal detection information in the detection data;

inputting the detection data into an initial regulation model to obtain prediction regulation information;

Inputting the prediction regulation information into the target water quality prediction model to obtain second predicted water quality;

determining a second loss parameter based on the second predicted water quality and the first target water quality;

and updating the parameters of the initial regulation model according to the second loss parameters to obtain the target regulation model.

Optionally, the training based on the sample data to obtain a target water quality prediction model includes:

constructing a regulation and control feature matrix and a detection data feature matrix according to the sample data;

performing feature cascading according to the regulation feature matrix and the detection data feature matrix to obtain a first cascade matrix;

determining an intermediary feature vector from the first cascade matrix and a predefined plurality of process feature vectors;

Determining a first regulation feature matrix and a first detection feature matrix according to the regulation feature matrix, the detection data feature matrix, the intermediate feature vector and a predefined filtering matrix, wherein the predefined filtering matrix is used for filtering the regulation feature matrix and the detection data feature matrix;

determining a model updating function according to the regulation feature matrix, the detection data feature matrix, the intermediate feature vector, the first regulation feature matrix and the first detection feature matrix;

training the initial water quality prediction model according to the model updating function, so that the trained initial water quality prediction model is used for learning the association influence parameters between the regulation feature matrix and the detection data feature matrix of sample data, and data decoupling between the regulation feature matrix and the detection data feature matrix.

Optionally, the determining an intermediary feature vector according to the first cascade matrix and a predefined plurality of process feature vectors includes:

Determining a first matrix distance between the first cascade matrix and each of the process feature vectors;

And carrying out weighted numerical sum determination on the plurality of process feature vectors according to the first matrix distance to obtain the intermediate feature vector.

Optionally, the regulation feature matrix includes a plurality of first regulation data matrices, the detection data feature matrix includes a first detection data matrix, and feature cascading is performed according to the regulation feature matrix and the detection data feature matrix to obtain a first cascade matrix, including:

performing mean value aggregation on the plurality of first regulation and control data matrixes to obtain a first regulation and control sequence feature matrix;

and performing feature cascade according to the first regulatory sequence feature matrix and the first detection data matrix to obtain a first cascade matrix.

Optionally, the determining a model update function according to the regulation feature matrix, the detection data feature matrix, the intermediate feature vector, the first regulation feature matrix, and the first detection feature matrix includes:

determining whether a positive class rate corresponds to the first regulation feature matrix and the first detection feature matrix according to the first regulation feature matrix, the first detection feature matrix and predefined weight distribution information;

and determining a first error index according to the labeling information corresponding to the positive class rate and the sample data pair, and determining whether a model updating function corresponds according to the first error index.

Optionally, the method further comprises:

acquiring a sample data set comprising a plurality of sample data;

Determining a first regulation feature matrix group and a first detection feature matrix group based on the regulation feature matrix, the detection data feature matrix, the intermediate feature vector and the predefined filtering matrix corresponding to a plurality of sample data pairs in the sample data group, wherein the first regulation feature matrix group contains a first regulation feature matrix corresponding to each sample data and a first detection feature matrix corresponding to each sample data, each first detection feature matrix in the first detection feature matrix group comprises a second detection data matrix, and each first regulation feature matrix in the first regulation feature matrix group comprises a plurality of second regulation data matrices;

Determining a target detection data matrix from a plurality of second detection data matrixes corresponding to the first detection feature matrix group according to the first regulation sequence feature matrix;

Respectively carrying out mean value aggregation on a plurality of second regulation and control data matrixes corresponding to each first regulation and control feature matrix in the first regulation and control feature matrix group to obtain a second regulation and control sequence feature matrix corresponding to each first regulation and control feature matrix in the first regulation and control feature matrix group;

And determining a target regulation and control feature matrix from a plurality of second regulation and control sequence feature matrices corresponding to the first regulation and control feature matrix group according to the first detection data matrix.

Determining whether a positive class rate corresponds to the first regulation feature matrix, the first detection feature matrix and the target detection data matrix, and whether a positive class rate corresponds to the first detection feature matrix and the target regulation feature matrix according to the first regulation feature matrix, the first detection feature matrix and the predefined weight distribution information;

And determining a first error index according to whether the first regulation feature matrix corresponds to the first detection feature matrix, whether the first regulation feature matrix corresponds to the target detection feature matrix, whether the first detection feature matrix corresponds to the target regulation feature matrix, the labeling information corresponding to the first regulation feature matrix and the first detection feature matrix, the labeling information corresponding to the first regulation feature matrix and the target detection feature matrix, and the labeling information between the first detection feature matrix and the target regulation feature matrix, and determining a model updating function according to the first error index.

Determining a second matrix distance between the first regulatory sequence feature matrix and the first detection data matrix;

Determining a first control sequence feature matrix and a third matrix distance between second detection data matrixes which do not correspond to the first control sequence feature matrix in the plurality of second detection data matrixes;

determining a second error index according to the second matrix distance and the third matrix distance, and determining a model update function according to the second error index;

The determining a model update function according to the regulation feature matrix, the detection data feature matrix, the intermediate feature vector, the first regulation feature matrix, and the first detection feature matrix includes:

Determining a fourth matrix distance between the first detection data matrix and the first regulatory sequence feature matrix;

determining a fifth matrix distance between the first detection data matrix and a second regulatory sequence feature matrix which does not correspond to the first detection data matrix in the plurality of second regulatory sequence feature matrices;

and determining a third error index according to the fourth matrix distance and the fifth matrix distance, and determining a model updating function according to the third error index.

In a second aspect, an embodiment of the present invention provides an intelligent aquaculture regulation and control system based on machine learning, where the system includes a computer device and an intelligent aquaculture device;

The computer device comprises a processor and a nonvolatile memory storing computer instructions, wherein the computer device executes the aquaculture intelligent regulation and control method based on machine learning in the first aspect when the computer instructions are executed by the processor.

Compared with the prior art, the method has the beneficial effects that the fish shoal detection information and the water quality detection information are obtained through the water body detection device and the underwater image detection device, and then the collected information is analyzed and determined based on the target regulation and control model, so that the regulated and controlled water quality can be effectively ensured to meet good culture conditions, and the culture benefit is improved.

Drawings

In order to more clearly illustrate the technical solution of the embodiments of the present invention, the drawings that are required to be used in the embodiments will be briefly described. It is appreciated that the following drawings depict only certain embodiments of the invention and are therefore not to be considered limiting of its scope. Other relevant drawings may be made by those of ordinary skill in the art without undue burden from these drawings.

FIG. 1 is a flow chart illustrating a machine learning based intelligent regulation method for aquaculture in accordance with an embodiment of the present invention;

fig. 2 is a schematic diagram of a computer device, according to an embodiment of the invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. It will be apparent that the described embodiments are some, but not all, embodiments of the invention. The components of the embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.

Thus, the following detailed description of the embodiments of the invention, as presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

It should be noted that like reference numerals and letters refer to like items in the following figures, and thus once an item is defined in one figure, no further definition or explanation thereof is necessary in the following figures.

In the description of the present invention, it should be understood that the directions or positional relationships indicated by the terms "upper", "lower", "inner", "outer", "left", "right", etc. are based on the directions or positional relationships shown in the drawings, or the directions or positional relationships conventionally put in place when the product of the application is used, or the directions or positional relationships conventionally understood by those skilled in the art are merely for convenience of describing the present invention and simplifying the description, and do not indicate or imply that the apparatus or elements to be referred to must have a specific direction, be constructed and operated in a specific direction, and thus should not be construed as limiting the present invention.

Furthermore, the terms "first," "second," and the like, are used merely to distinguish between descriptions and should not be construed as indicating or implying relative importance.

In the description of the present invention, it should also be noted that, unless explicitly specified and limited otherwise, terms such as "disposed," "connected," and the like are to be construed broadly, and for example, "connected" may be a fixed connection, a detachable connection, or an integral connection, may be a mechanical connection or an electrical connection, may be a direct connection, may be an indirect connection via an intermediary, or may be a communication between two elements. The specific meaning of the above terms in the present invention can be understood by those of ordinary skill in the art according to the specific circumstances.

The following describes specific embodiments of the present invention in detail with reference to the drawings.

FIG. 1 is a flow chart of a machine learning based intelligent regulation method for aquaculture, which can be executed by a computer device, according to an embodiment of the present application, as shown in FIG. 1, and includes:

Step S101, acquiring water quality detection information by utilizing a water body detection device.

The water quality detection information comprises at least one of water temperature, pH value, oxygen content, nitrate content, nitrite content and feed residue proportion.

Specifically, the water body detection device may include sensing devices such as a feed allowance sensor, a dissolved oxygen sensor, a water temperature and air temperature sensor, a PH sensor, and a nitrite sensor.

Wherein, can install a plurality of water detection device in the internal different positions of aquatic in order to more comprehensively detect water quality.

Step S102, acquiring fish school detection information by using an underwater image detection device.

The fish school detection information comprises at least one of fish school liveness, fish school quantity and average body length.

Specifically, the image acquired by the image detection device can be detected by an image recognition technology, so that information such as the liveness of the fish shoal, the number of the fish shoals, the average body length and the like can be obtained. For example, the number and the body length of fish in the image may be analyzed according to the image data detected by the image detection device, so as to predict the total number and the average body length of the fish in the water, and further, the moving speed of the fish may be detected according to a plurality of images continuously detected by the image detection device, so as to determine the activity of the fish.

Step S103, inputting the water quality detection information and the fish school detection information into a target regulation and control model obtained based on machine learning to obtain target regulation and control information.

The distance between the predicted water quality obtained by inputting the target regulation and control information into the target water quality prediction model and the target water quality is smaller than a preset threshold value, the target water quality is determined based on the fish shoal detection information, and the target regulation and control model is obtained by combined training with the target water quality prediction model.

Step S104, regulating and controlling the working state of the intelligent culture device according to the target regulation and control information.

Wherein, intelligent breeding device includes at least one in oxygenation device, the device of throwing something and feeding, hydrologic cycle processing apparatus.

It is worth noting that when different breeding devices are regulated and controlled, various different influences may be caused to the quality of the water body, for example, when the nitrate content nitrite in the water body is regulated by using the water circulation treatment device, feed residues and oxygen content in the water body may be reduced, if the feed is insufficient, a good living environment cannot be obtained for the fish shoal, and further, the breeding effect is poor, at the moment, the working states of the oxygenation device and the feeding device also need to be regulated, and further, the feed residues and the oxygen content in the water body are increased.

The target water quality prediction model can predict a water quality detection result after a period of time in the future according to regulation and control information of the intelligent culture device, and can learn the influence on water quality possibly caused in different dimensionalities when regulating and controlling different culture devices through machine learning.

It can be understood that the accuracy of the target water quality prediction model greatly influences the accuracy of the target regulation model, so that in the embodiment of the application, a combined training mode is adopted, the influence of the target water quality prediction model on the water quality is learned and regulated, and the target regulation model learns what cultivation device should be controlled under different water qualities so that the water quality can reach an ideal state.

In the embodiment of the disclosure, the water body detection device and the underwater image detection device are used for acquiring the shoal detection information and the water quality detection information, and further the collected information is analyzed and determined based on the target regulation and control model to determine the target regulation and control information, so that the regulated and controlled water quality can be effectively ensured to meet good culture conditions, and the culture benefit is improved.

In some embodiments, the joint training of the target regulation model and the target water quality prediction model comprises the steps of:

Acquiring a target training set, wherein the target training set comprises a preprocessed historical regulation data sequence and a corresponding historical detection data sequence, the historical regulation data sequence comprises regulation information of a plurality of regulation periods on the intelligent culture device, and the historical detection sequence comprises water quality detection information and fish shoal detection information detected by the intelligent culture device in a preset time period after the regulation information is executed;

Extracting a plurality of pieces of regulation data from the historical regulation data sequence, extracting one piece of detection data from the historical detection data sequence to form sample data and labeling information, wherein the labeling information is used for indicating regulation data corresponding to the detection data in the sample data;

Training based on sample data to obtain a target water quality prediction model, wherein the target water quality prediction model is used for predicting the regulated water quality according to regulation information of the intelligent culture device;

Inputting the predicted regulation and control information into a target water quality prediction model to obtain second predicted water quality;

determining a second loss parameter according to the second predicted water quality and the first target water quality;

And updating the parameters of the initial regulation model according to the second loss parameters to obtain a target regulation model.

It will be appreciated that the regulation period may be calibrated according to actual requirements, which is not limited by the embodiments of the present disclosure. It should be noted that each piece of detection data may include water quality detection information and fish school detection information corresponding to one regulation period.

In some embodiments, training a target water quality prediction model based on the sample data may include the steps of:

Step 201, constructing a regulation feature matrix and a detection data feature matrix according to the sample data.

The initial water quality prediction model provided by the embodiment of the application can comprise a regulation data encoder, a detection data encoder, a memory module and a cross-dimension encoder. The detection data encoder, the regulation data encoder and the cross-dimension encoder can be realized based on a transducer architecture. The system comprises a regulation and control data encoder, a detection data encoder, a cross-dimension encoder, a memory module and a data processing module, wherein the regulation and control data encoder is used for extracting a regulation and control data matrix, the detection data encoder is used for extracting detection data characteristics, the cross-dimension encoder is used for performing cross-dimension characteristic encoding or decoding, the memory module is used for storing process characteristic vectors, and the process characteristic vectors are used for representing rich intermediate dimension information.

In the embodiment of the application, a plurality of regulation data in sample data can be encoded through a regulation data encoder and is encoded based on a filtering matrix in the regulation data encoder so as to extract characteristics of the regulation data in the plurality of regulation data in the sample data to obtain a regulation characteristic matrix, and detection data in the sample data is encoded through a detection data encoder and is encoded based on the filtering matrix in the detection data encoder so as to extract detection data characteristics in the sample data.

And step 302, performing feature cascade according to the regulation feature matrix and the detection data feature matrix to obtain a cascade matrix.

In the embodiment of the application, the regulation and control feature matrix comprises a plurality of first regulation and control data matrixes, the detection data feature matrix comprises a first detection data matrix, and the step 302 mainly comprises the steps of carrying out mean value aggregation on the plurality of first regulation and control data matrixes to obtain a first regulation and control sequence feature matrix, and carrying out feature cascading according to the first regulation and control sequence feature matrix and the first detection data matrix to obtain a first cascade matrix.

Specifically, the first regulatory sequence feature matrix and the first detection data matrix may be fused by an MLP (Multi-layerPerceptron ) model.

Step 303, determining an intermediate feature vector from the cascade matrix and the predefined plurality of process feature vectors.

Specifically, the regulation feature matrix, the intermediate feature vector and the detection data feature matrix are spliced together and input into a cross-dimension encoder. In the cross-dimension encoder, the regulation feature matrix and the detection data feature matrix interact with the intermediate feature vector. It can be considered that the intermediate feature vector has certain relevance to the regulation feature matrix and the detection data feature matrix. Therefore, in the embodiment of the application, according to the cascade matrix, some information more similar to the cascade matrix can be obtained in the memory module to obtain the intermediate feature vector.

During the pre-training process, the memory storage stores a large number of memory representations for dimensional interactions to enhance the representation capability of cross-dimensional bridging and to promote model robustness.

In some embodiments, step 303 may include determining a first matrix distance between the first cascade matrix and each of the process feature vectors, and performing a weighted numerical sum determination on the plurality of process feature vectors based on the first matrix distance to obtain an intermediate feature vector.

Specifically, the first matrix distance can be obtained by determining the cosine distance between the first cascade matrix and each process feature vector, and the first matrix distance between the first cascade matrix and each process feature vector can also be directly obtained by inputting the first cascade matrix into a network model. For example, after the first cascade matrix is obtained, the cosine distance between the first cascade matrix and each process feature vector in the memory can be determined, and the intermediate feature vector more similar to the cascade matrix can be obtained by linear addition of the cosine distances between each process feature vector and the first cascade matrix as weights.

Step 304, determining a first regulation feature matrix and a first detection feature matrix according to the regulation feature matrix, the detection data feature matrix, the intermediate feature vector and a predefined filtering matrix, wherein the filtering matrix is used for filtering the regulation feature matrix and the detection data feature matrix.

Specifically, in order to establish an association relationship between the dimension of the regulation and control information and the dimension of the water quality detection result and ensure decoupling between the dimension of the regulation and control information and the dimension of the water quality detection result, the embodiment of the application introduces the intermediate feature vector for learning the regulation and control dimension information and the water quality dimension information. The regulation dimension and the water quality dimension are filtered through the filtering matrix, and fine-grained interaction is carried out on the regulation dimension and the water quality dimension and the intermediate feature vector respectively, so that the model can learn the association influence parameters among the cross dimensions, and the dimension separability among different dimensions is ensured.

In the embodiment of the application, after the regulation feature matrix and the detection data feature matrix are obtained, the regulation feature matrix, the detection data feature matrix and the intermediate feature vector are cascaded, and the cascaded features are input into a cross-dimension encoder. In the cross-dimension encoder, the embodiment of the application applies a novel inter-dimension interaction mechanism, wherein the regulation feature matrix and the detection data feature matrix are not directly related, but are respectively related to the intermediate feature vector. Specifically, in the self-attention module of the cross-dimensional encoder, the intermediate feature vector performs attention computation with the regulatory feature matrix and the detection data feature matrix, respectively.

Specifically, the regulation feature matrix, the detection data feature matrix and the intermediate feature vector are encoded through a cross-dimensional encoder, and encoding processing is performed based on a predefined filtering matrix in the cross-dimensional encoder, so that a first regulation feature matrix, a first detection feature matrix and a target intermediate feature vector are obtained.

Specifically, as the only way for information comparison between regulation and water quality dimensions, intermediate feature vectors learn intermediate dimension information that facilitates interaction of the two original dimensions. Dimension separability is ensured through mutual filtering, and a first regulation feature matrix and a first detection feature matrix which keep respective dimension characteristics after interaction are obtained.

Step 305, determining a model update function according to the regulation feature matrix, the detection data feature matrix, the intermediate feature vector, the first regulation feature matrix and the first detection feature matrix.

Specifically, in order to improve the feature reliability after passing through the cross-dimension encoder, matrix distance matching can be performed on the first regulation feature matrix and the first detection feature matrix obtained after passing through the cross-dimension encoder, an error index is constructed according to corresponding labeling information, and the initial water quality prediction model is trained.

In some embodiments, step 305 may include determining whether a positive class rate corresponds between the first regulation feature matrix and the first detection feature matrix according to the first regulation feature matrix, the first detection feature matrix, and predefined weight distribution information, determining a first error index according to labeling information corresponding to the pair of positive class rate and sample data, and determining a model update function according to the first error index, where the labeling information is used to label whether the sample data in the pair of sample data corresponds to the sample data.

Specifically, the weight distribution information refers to parameter values learned during training. For weight distribution information, the values may be updated in an iterative manner starting from a set of random values, which are then updated as the network learns. In the embodiment of the application, two weight distribution information is introduced to predict the positive class rate, specifically, after each dimension characteristic corresponding to the sample data pair passes through a cross-dimension encoder, matrix distance matching is needed.

Specifically, for each first regulation and control feature matrix output by the cross-dimension encoder, the target regulation and control sequence feature matrix can be obtained by means of average value aggregation of regulation and control data matrixes in each first regulation and control feature matrix. And regarding the detection data matrix in the first detection feature matrix output by the cross-dimension encoder as the overall detection feature of the first detection feature matrix.

Specifically, the first error indicator may be determined according to a cross entropy between the predicted value and the labeling information.

Specifically, when training the initial water quality prediction model, a sample data set containing a large amount of sample data is generally adopted, and the target regulatory sequence feature matrix and the target detection data feature corresponding to each pair of sample data pairs can be regarded as one sample pair. For each object detection data feature, there are multiple negative samples. When the first regulation and control feature matrix and the first detection feature matrix are subjected to matrix distance matching, positive class rate between each positive sample pair can be determined, a cross entropy loss function is constructed according to the positive class rate between each positive sample pair and labeling information corresponding to each positive sample pair, positive class rate between target detection data features and negative sample cases is determined, a cross entropy loss function is constructed according to the positive class rate between the target detection data features and the negative sample cases and labeling information corresponding to each negative sample pair, and an initial water quality prediction model is trained according to the cross entropy function, so that the positive class rate between each positive sample pair and the positive class rate between each negative sample pair are closer to the labeling information.

The method comprises the steps of obtaining a sample data set comprising a plurality of sample data pairs, determining a first regulation and control feature matrix set and a first detection feature matrix set based on regulation and control feature matrixes, detection data feature matrixes, intermediate feature vectors and predefined filtering matrixes corresponding to the plurality of sample data pairs in the sample data set, wherein the first regulation and control feature matrix set comprises first regulation and control feature matrixes corresponding to the sample data and first detection feature matrixes corresponding to the sample data, each first detection feature matrix in the first regulation and control feature matrix set comprises a second detection data matrix, each first regulation and control feature matrix in the first regulation and control feature matrix set comprises a plurality of second regulation and control data matrixes, determining a target detection data matrix from the plurality of second detection data matrixes corresponding to the first detection feature matrix set according to a regulation and control sequence feature matrix, conducting mean value aggregation on the plurality of second data matrixes corresponding to the first regulation and control feature matrixes in the first regulation and control feature matrix set respectively to obtain second regulation and control feature matrixes corresponding to the first sample data, and determining a target detection feature matrix from the first regulation and control feature matrixes according to a regulation and control sequence feature matrix.

Specifically, step 305 may include determining, according to the first regulation feature matrix, the first detection feature matrix, and the predefined weight distribution information, whether a positive class ratio between the first regulation feature matrix and the first detection feature matrix corresponds, whether a positive class ratio between the first regulation feature matrix and the target detection feature matrix corresponds, and whether a positive class ratio between the first detection feature matrix and the target regulation feature matrix corresponds, determining, according to the positive class ratio between the first regulation feature matrix and the first detection feature matrix, the positive class ratio between the first regulation feature matrix and the target detection feature matrix, and the labeling information corresponding to the first regulation feature matrix and the first detection feature matrix, the labeling information corresponding to the first regulation feature matrix and the target detection feature matrix, and the labeling information corresponding to the first detection feature matrix and the target regulation feature matrix, and determining a first error index, and determining an update function according to the first error index.

After the first regulatory sequence feature matrix and the first detection data matrix are obtained, the similarity of the first regulatory sequence feature matrix and the first detection data matrix is used as a normalized probability, and a target detection data matrix and a target regulatory feature matrix are respectively sampled for the first regulatory sequence feature matrix and the first detection data matrix corresponding to each sample data pair in the sample data set.

In some embodiments, to improve the underlying representation capability, training tasks may be provided that align between the underlying dimensions, functioning as a first alignment prior to fusion.

Specifically, step 305 may mainly include determining a second matrix distance between the first regulatory sequence feature matrix and the first detection data matrix, determining the first regulatory sequence feature matrix and a third matrix distance between second detection data matrices of the plurality of second detection data matrices that do not correspond to the first regulatory sequence feature matrix, determining a second error index according to the second matrix distance and the third matrix distance, and determining a model update function according to the second error index.

Similarly, step 305 may further include determining a fourth matrix distance between the first detection data matrix and the first regulatory sequence feature matrix, determining the first detection data matrix and a fifth matrix distance between a second regulatory sequence feature matrix of the plurality of second regulatory sequence feature matrices that does not correspond to the first detection data matrix, determining a third error indicator based on the fourth matrix distance and the fifth matrix distance, and determining a model update function based on the third error indicator.

For example, a contrast learning error index may be employed to align the underlying dimensions. Specifically, the first regulatory sequence feature matrix and the first detection data matrix are aligned with the aim of maximizing the matrix distance between the first regulatory sequence feature matrix and the first detection data matrix and minimizing the matrix distance between the first regulatory sequence feature matrix and the second detection data matrix corresponding to other sample pairs, and the first regulatory sequence feature matrix and the first detection data matrix are aligned with the aim of maximizing the matrix distance between the first regulatory sequence feature matrix and the first detection data matrix and minimizing the matrix distance between the first detection data matrix and the second regulatory sequence feature matrix corresponding to other sample pairs

And 306, training the initial water quality prediction model according to the model updating function, so that the trained initial water quality prediction model is used for learning the association influence parameters between the regulation feature matrix and the detection data feature matrix of the sample data pair and data decoupling between the regulation feature matrix and the detection data feature matrix.

Specifically, an Adam optimizer may be used to train the model in two steps, first, training the dispatch data encoder and the detection data encoder using only the alignment tasks between the underlying dimensions, and then training the model using all of the pre-training tasks.

In an embodiment of the present application, step 305 may include determining at least one of the second error index and the third error index, or a sum of the second error index and the third error index, as a first model update function, determining a sum of the first error index, the second error index, the third error index, the fourth error index, and the fifth error index as a second model update function, and step 306 may include performing one-stage training on the initial water quality prediction model according to the first model update function, and performing two-stage training on the initial water quality prediction model according to the second model update function.

The method comprises the steps of obtaining sample data marked in pairs, carrying out feature extraction on the sample data to obtain a regulation feature matrix of the sample data in the sample data pair and a detection data feature matrix of the sample data in the sample data pair, carrying out feature cascading according to the regulation feature matrix and the detection data feature matrix to obtain a first cascade matrix, determining an intermediate feature vector according to the first cascade matrix and a plurality of predefined process feature vectors, determining a first regulation feature matrix and a first detection feature matrix according to the regulation feature matrix, the detection data feature matrix, the intermediate feature vector and a predefined filtering matrix, wherein the predefined filtering matrix is used for filtering the regulation feature matrix and the detection data feature matrix, determining a model updating function according to the regulation feature matrix, the detection data feature matrix, the intermediate feature vector, the first regulation feature matrix and the first detection feature matrix, training an initial water quality prediction model according to the model updating function, and enabling the trained initial water quality prediction model to be used for learning association influence parameters between the regulation feature matrix and the detection data feature matrix of the sample data pair and data feature matrix and carrying out data decoupling between the regulation feature matrix and the detection data feature matrix.

The embodiment of the disclosure also provides an aquaculture intelligent regulation and control system based on machine learning, which comprises computer equipment and an intelligent aquaculture device;

The computer device may be used to execute the machine learning-based aquaculture intelligent regulation and control method involved in the above embodiment to control the working state of the intelligent aquaculture device.

In some embodiments, the computer device includes a processor and a non-volatile memory storing computer instructions that, when executed by the processor, perform the machine learning based intelligent regulation method of aquaculture referred to in the above embodiments.

Referring now to fig. 2, a schematic diagram of a computer device 200 suitable for use in implementing embodiments of the present disclosure is shown, which may be, for example, a network device in an embodiment of the method corresponding to fig. 1. The terminal devices in the embodiments of the present disclosure may include, but are not limited to, mobile terminals such as mobile phones, notebook computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and the like, and stationary terminals such as digital TVs, desktop computers, and the like. The computer device illustrated in fig. 2 is merely an example and should not be construed as limiting the functionality and scope of use of embodiments of the present disclosure.

As shown in fig. 2, the computer apparatus 200 may include a processing device 201, and the processing device 201 may be, for example, a central processing unit, an image processor, or the like, which may perform various appropriate actions and processes according to a program stored in a read only memory 202 or a program loaded from a storage device 208 into a random access memory 203. In the random access memory 203, various programs and data required for the operation of the computer device 200 are also stored. The processing means 201, the read only memory 202 and the random access memory 203 are connected to each other by a bus 204. An input/output interface 205 is also connected to the bus 204.

In general, devices may be connected to the input/output interface 205 including input devices 206 such as a touch screen, touch pad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc., output devices 207 including a Liquid Crystal Display (LCD), speaker, vibrator, etc., storage devices 208 including, for example, magnetic tape, hard disk, etc., and communication devices 209. The communication means 209 may allow the computer device 200 to communicate with other devices wirelessly or by wire to exchange data. While FIG. 2 illustrates a computer apparatus 200 having various devices, it is to be understood that not all illustrated devices are required to be implemented or provided. More or fewer devices may be implemented or provided instead.

In particular, according to embodiments of the present disclosure, the processes described above with reference to flowcharts may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program embodied on a non-transitory computer readable medium, the computer program comprising program code for performing the method shown in the flow chart. In such an embodiment, the computer program may be downloaded and installed from a network via the communication means 209, or from the storage means 208, or from the read only memory 202. The above-described functions defined in the methods of the embodiments of the present disclosure are performed when the computer program is executed by the processing device 201.

It should be noted that the computer readable medium described in the present disclosure may be a computer readable signal medium or a computer readable storage medium, or any combination of the two. The computer readable storage medium can be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples of a computer-readable storage medium may include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this disclosure, a computer-readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. In the present disclosure, however, the computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, with the computer-readable program code embodied therein. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination of the foregoing. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to electrical wiring, fiber optic cable, RF (radio frequency), and the like, or any suitable combination of the foregoing.

In some embodiments, the clients, servers may communicate using any currently known or future developed network protocol, such as HTTP (HyperText Transfer Protocol ), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), the internet (e.g., the internet), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future developed networks.

The computer readable medium may be contained in the computer device or may exist alone without being incorporated into the computer device.

The computer-readable medium carries one or more programs which, when executed by the computer device, cause the computer device to perform the steps involved in the above-described embodiments.

Or the computer-readable medium carries one or more programs which, when executed by the computer device, cause the computer device to perform the steps referred to in the embodiments.

Computer program code for carrying out operations of the present disclosure may be written in one or more programming languages, including, but not limited to, an object oriented programming language such as Java, smalltalk, C ++ and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computer (for example, through the Internet using an Internet service provider).

The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

The modules described in the embodiments of the present disclosure may be implemented in software or hardware. The name of a module does not in some cases define the module itself.

The functions described above herein may be performed, at least in part, by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic that may be used include Field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), systems-on-a-chip (SOCs), complex Programmable Logic Devices (CPLDs), and the like.

In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

The foregoing is merely illustrative of the present invention, and the present invention is not limited thereto, and any changes or substitutions easily contemplated by those skilled in the art within the scope of the present invention should be included in the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. An intelligent aquaculture regulation and control method based on machine learning, which is characterized by comprising the following steps:

Regulating and controlling the working state of an intelligent culture device according to the target regulation and control information, wherein the intelligent culture device comprises at least one of an oxygenation device, a feeding device and a water circulation treatment device;

the combined training of the target regulation model and the target water quality prediction model comprises the following steps:

2. The method of claim 1, wherein the training to obtain a target water quality prediction model based on the sample data comprises:

3. The method of claim 2, wherein the determining an intermediary feature vector from the first cascade matrix and a predefined plurality of process feature vectors comprises:

4. The method of claim 3, wherein the regulatory feature matrix comprises a plurality of first regulatory data matrices, the detection data feature matrix comprises a first detection data matrix, the feature cascading is performed according to the regulatory feature matrix and the detection data feature matrix to obtain a first cascade matrix, and the method comprises:

5. The method of claim 4, wherein the determining a model update function based on the regulatory feature matrix, the detected data feature matrix, the intermediate feature vector, the first regulatory feature matrix, and the first detected feature matrix comprises:

6. The method according to claim 4, wherein the method further comprises:

acquiring a sample data set comprising a plurality of sample data;

7. The method of claim 6, wherein the determining a model update function based on the regulatory feature matrix, the detected data feature matrix, the intermediate feature vector, the first regulatory feature matrix, and the first detected feature matrix comprises:

8. The method of claim 6, wherein the determining a model update function based on the regulatory feature matrix, the detected data feature matrix, the intermediate feature vector, the first regulatory feature matrix, and the first detected feature matrix comprises:

9. An aquaculture intelligent regulation and control system based on machine learning is characterized by comprising computer equipment and an intelligent aquaculture device;

The computer device comprises a processor and a nonvolatile memory storing computer instructions which, when executed by the processor, perform the machine learning based intelligent aquaculture regulation method of any one of claims 1-8.