Intelligent operation system for aquaculture management

Abstract

The invention discloses an intelligent operation system for aquaculture management, which relates to the field of aquaculture and comprises a water quality detection module for detecting water quality data in an aquaculture unit, a meteorological detection module for detecting meteorological data of the position of the aquaculture unit, and an ultrasonic detection module for detecting quantity data of fishes in the aquaculture unit, the image detection module detects weight data and growth stages of fish in the breeding unit, and the analysis control module analyzes standard water quality and standard feed feeding amount in the breeding unit through an AI analysis model according to meteorological data, quantity data, weight data and growth stages; the analysis control module compares the difference between the water quality data at the current moment and the standard water quality, controls the oxygenation equipment to oxygenate and controls the water treatment equipment to change water; according to the standard feed feeding amount, a control signal is sent to control feeding system equipment to feed. According to the invention, powerful support is provided for modernized development of aquaculture, and more efficient, real-time and intelligent aquaculture management is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of aquaculture, and more particularly to an intelligent operating system for aquaculture management. Background Art

[0002] Aquatic products are rich in nutrients such as protein, unsaturated fatty acids, vitamins, minerals, etc., and have health benefits such as improving brain function, controlling and assisting in adjusting blood pressure, blood sugar, blood lipids, and preventing cancer. They provide humans with high-quality nutritional supplements, and people's demand for aquatic products is growing.

[0003] With the continuous growth of demand for aquatic products, the aquaculture industry is facing an urgent need to improve production efficiency, ensure product quality and reduce aquaculture risks. However, traditional management methods often rely on manual management, which is only based on the personal aquaculture experience of farmers or company-related managers. The management of water quality, water quantity, feed feeding, etc. is greatly affected by personal factors. The detection of water quality, water quantity, fish quality and quantity, and weather changes is not timely or the comprehensive judgment is wrong, which often leads to poor aquaculture efficiency and feed utilization, and high aquaculture risks.

[0004] Therefore, how to introduce an intelligent aquaculture management system to improve aquaculture efficiency and reduce aquaculture risks is an urgent problem that technical personnel in this field need to solve. Summary of the invention

[0005] In view of this, the present invention provides an intelligent operating system for aquaculture management, which collects water quality sensor data, fish life status data and micro-meteorological data, and uses AI calculation to control the operation of oxygenation equipment, water treatment equipment and feeding system equipment.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] The invention discloses an intelligent operating system for aquaculture management, comprising:

[0008] Water quality detection module, meteorological detection module, ultrasonic detection module, image detection module, storage module, analysis and control module, input and output module;

[0009] The water quality detection module detects the water quality data in the breeding unit and uploads it to the storage module; the meteorological detection module detects the meteorological data of the location of the breeding unit and uploads it to the storage module; the ultrasonic detection module detects the quantity data of the fish in the breeding unit and uploads it to the storage module; the image detection module detects the weight data and growth stage of the fish in the breeding unit and uploads it to the storage module;

[0010] The analysis and control module analyzes the standard water quality and the standard feed feeding amount in the breeding unit through an AI analysis model according to the meteorological data, the quantity data, the weight data, and the growth stage;

[0011] The analysis and control module compares the difference between the water quality data at the current moment and the standard water quality. When the difference is greater than the set range, the input and output module displays the water quality abnormality and sends a control signal to control the oxygenation equipment to increase oxygen and the water treatment equipment to adjust the water quality; according to the standard feed feeding amount, a control signal is sent to control the feeding system equipment to feed the feed.

[0012] Furthermore, the water quality detection module includes: a water temperature sensor, a pH value sensor, a dissolved oxygen sensor, an ion concentration sensor, an ammonia nitrogen sensor, and a turbidity sensor;

[0013] The meteorological detection module includes: a temperature sensor, a humidity sensor, an air pressure sensor, a light intensity sensor, and a precipitation sensor.

[0014] Furthermore, the ultrasonic detection module includes a plurality of ultrasonic detection units, which are evenly installed in the breeding unit to monitor the number of fish within the detection range in real time and calculate the fish density within the detection range; after calculating the average value of the fish density monitored by all the ultrasonic detection units, the breeding quantity estimate at each moment is obtained according to the area of ​​the breeding unit, and the breeding quantity estimate at multiple moments is averaged to obtain the quantity data.

[0015] Furthermore, the image detection module is arranged at the feeding system device to collect image data of the fish when eating, and use the image recognition model to identify the weight data and growth stage of the fish;

[0016] The image recognition model includes:

[0017] A first processing unit: performing segmentation processing on the image data, and segmenting the image data into a background area and a target area based on the gray value of the image data and an optimal segmentation threshold;

[0018] A second processing unit: identifying the weight and growth stage of the fish in the segmented target area based on a deep learning network to obtain the weight data and the growth stage;

[0019] The deep learning network includes an input CNN layer, a BiLSTM layer, an attention layer and an output layer.

[0020] Furthermore, the optimal partitioning threshold is obtained by:

[0021] Collect sample images containing farmed fish and mark background areas and target areas;

[0022] Divide the sample image into K gray levels, and the number of pixels with gray value k is d k The total number of pixels is The probability of each gray value appearing is in

[0023] The grayscale values ​​are divided into two categories: background O = (0, 1, ..., s) and target I = (s + 1, s + 2, ..., K-1). The probabilities of the two categories are:

[0024]

[0025] where p O 、p I are the probabilities of background O and target I appearing respectively; the grayscale means of the two categories are:

[0026]

[0027] Among them, μ O , μ I They are the grayscale mean of background O and target I respectively;

[0028] The total grayscale mean of the sample image for:

[0029]

[0030] The inter-class variance σ between the background area and the target area is:

[0031]

[0032] The optimal partition threshold is expressed as:

[0033]

[0034] Where ρ represents the optimal partition threshold, Represents a parameter-seeking function.

[0035] Furthermore, the filter of the input CNN layer is expressed as:

[0036]

[0037] Among them, X i is the i-th output feature extracted, ReLu is the correction unit, F i is the weight matrix of the i-th convolution kernel, x is the input target area image, b is the model bias, * represents the convolution operation, C i is the i-th corresponding learning filter, represents the dot multiplication operation, W is the filter kernel;

[0038] The BiLSTM layer is represented as:

[0039]

[0040] Among them, g t , g t-1 are the hidden states of the forward LSTM output at time t and t-1, respectively, t ,q t-1 are the hidden states of the reverse LSTM output at time t and t-1, respectively, t is the input at time t, h t is the hidden state of the BiLSTM layer at time t, w 1 、w 2 are the weights corresponding to the forward and reverse LSTM, respectively, t is the bias, H is the hidden layer state sequence output by the BiLSTM layer;

[0041] The attention mechanism of the attention layer is expressed as:

[0042]

[0043] Among them, att t is the output of the attention mechanism at time t, w 3 is the weight, h τ is the hidden state τ∈(1,t) of the BiLSTM layer at time τ, and A is the output matrix of the attention mechanism;

[0044] The output layer is expressed as:

[0045] y=sigmoid(w A A+b A );

[0046] Among them, y is the output vector, sigmoid is the nonlinear activation function, and w A is the weight, b A For bias.

[0047] Furthermore, the AI ​​analysis model is obtained by:

[0048] Step 1: Obtain some historical aquaculture data, select historical data whose aquaculture efficiency is greater than a set threshold, and perform a sliding window operation through a sliding window of a set size to obtain multiple sample data;

[0049] Step 2: Arrange the meteorological data, quantity data, weight data, and growth stage of the sample data to obtain a feature vector of the sample; arrange the water quality data and feed feeding amount of the sample to obtain a label vector of the sample;

[0050] Step 3: Cluster the label vectors to obtain the label vector classification data; calculate the average value of the label vectors in each classification data as the standard label vector of the corresponding classification data; replace the label vector of the sample data with the standard label vector of the corresponding classification as the training sample label, and use the feature vector of the sample data as the training sample feature to obtain a training sample set;

[0051] Step 4: Construct a decision tree classification model using the gradient boosting decision tree method, and use the training sample set to train and test to obtain the AI ​​analysis model.

[0052] Furthermore, the step 4 specifically includes:

[0053] Dividing the training sample set into a training set and a test set according to a preset ratio, retaining the sample labels of the training set and hiding the sample labels of the test set;

[0054] Initializing the decision tree model classification model parameters;

[0055] The training samples in the training set are input into the decision tree model classification model for supervised learning training to obtain the AI ​​analysis model.

[0056] Furthermore, the decision tree classification model is expressed as:

[0057]

[0058] in, represents the model prediction value, f m represents the mth decision subtree, M is the total number of decision subtrees, represents a decision classification tree, V r represents a feature vector of an input training sample, r is the number of classifications of the label vector classification data, and F represents a decision classification tree.

[0059] Furthermore, the loss function in the supervised learning training is expressed as:

[0060]

[0061] Among them, L (m) represents the loss function value, l represents the loss function that measures the estimated value and the label value, and R represents the total number of classifications of the label vector classification data. represents the decision classification prediction value before the mth decision subtree, f m (Vr ) indicates the predicted value of the mth decision subtree, y r Represents the label vector of the training sample.

[0062] Through the above technical solutions, it can be known that compared with the prior art, the present invention discloses an intelligent operating system for aquaculture management, which can collect and monitor multiple key water quality parameters in real time, effectively grasp the state of the aquaculture environment; can identify and monitor the different growth stages of fish, and provide a scientific basis for subsequent feeding, management and harvesting; can analyze data and promptly feedback to the client page when slight abnormalities occur, so that managers can understand the situation in time; combined with the feeding type, sensor data, fish quality and quantity, fish growth stage and weather changes, through AI algorithms, oxygenation equipment, water treatment equipment and feeding system equipment are controlled in real time, the automation level is improved, and through systematic management, the breeding efficiency and economic benefits are improved, and the health and growth safety of fish are guaranteed. The system integrates a number of advanced technologies such as sensor technology, Internet of Things, artificial intelligence, etc., which significantly improves the management efficiency of aquaculture, the timeliness and adaptability of fish growth status monitoring, provides strong support for the modernization of aquaculture, and realizes more efficient, real-time and intelligent breeding management. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0064] Figure 1 It is a schematic diagram of the overall structure of an embodiment of the present invention. DETAILED DESCRIPTION

[0065] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0066] The embodiment of the present invention discloses an intelligent operating system for aquaculture management, such as Figure 1 As shown, including:

[0067] Water quality detection module, meteorological detection module, ultrasonic detection module, image detection module, storage module, analysis and control module, input and output module;

[0068] The water quality detection module detects the water quality data in the breeding unit and uploads it to the storage module; the meteorological detection module detects the meteorological data of the location of the breeding unit and uploads it to the storage module; the ultrasonic detection module detects the number data of the fish in the breeding unit and uploads it to the storage module; the image detection module detects the weight data and growth stage of the fish in the breeding unit and uploads it to the storage module;

[0069] The analysis and control module analyzes the standard water quality and standard feed feeding amount in the breeding unit through the AI ​​analysis model based on meteorological data, quantity data, weight data, and growth stage;

[0070] The analysis and control module compares the difference between the current water quality data and the standard water quality. When the difference is greater than the set range, the input and output module displays the water quality abnormality and sends out a control signal to control the aeration equipment to increase oxygen and the water treatment equipment to adjust the water quality. According to the standard feed feeding amount, a control signal is sent to control the feeding system equipment to feed the feed.

[0071] In a specific embodiment, the oxygenation equipment includes: mechanical oxygenation equipment, air oxygen oxygenation equipment, and pure oxygen system equipment; by controlling different oxygenation equipment, the aquaculture unit is oxygenated so that the oxygen content in the water is close to the dissolved oxygen concentration in the standard water quality. The water treatment equipment includes: circulation system equipment, irrigation and drainage system equipment, air energy constant temperature system equipment, disinfection system equipment, and water treatment system equipment; by controlling the circulation system equipment and cooperating with the irrigation and drainage system equipment to adjust the particle concentration of the water in the aquaculture unit, the air energy constant temperature system equipment is used to adjust the water temperature in the aquaculture unit, and the disinfection equipment and water treatment equipment are used to adjust the pH value, ion concentration, and ammonia nitrogen content of the water in the aquaculture unit, so that the water quality in the aquaculture unit is the same as the standard water quality.

[0072] In a specific embodiment, the water quality detection module includes: a water temperature sensor, a pH sensor, a dissolved oxygen sensor, an ion concentration sensor, an ammonia nitrogen sensor, and a turbidity sensor;

[0073] The meteorological detection module includes: temperature sensor, humidity sensor, air pressure sensor, light intensity sensor, precipitation sensor;

[0074] The input and output module is responsible for transmitting the control commands of the analysis and control module to the oxygenation equipment, water treatment equipment, and feeding system equipment, and also sends real-time monitoring data to the remote monitoring end; when the AI ​​analysis model detects data abnormalities, an alarm is issued.

[0075] In a specific embodiment, the ultrasonic detection module includes a plurality of ultrasonic detection units, which are evenly installed in the breeding unit to monitor the number of fish within the detection range in real time and calculate the fish density within the detection range; after calculating the average value of the fish density monitored by all ultrasonic detection units, the estimated breeding quantity at each moment is obtained according to the area of ​​the breeding unit, and the estimated breeding quantity at multiple moments is averaged to obtain the quantity data.

[0076] In a specific embodiment, the image detection module is arranged at the feeding system equipment, so as to facilitate the capture of images of fish in the breeding unit; a white background board is arranged at the bottom of the water area where the feeding system equipment is located, and a cleaning device is arranged on the background board to regularly clean the sediment on the background board to facilitate the distinction between the fish and the background in the image; the image detection module collects image data of the fish when eating, and uses the image recognition model to identify the weight data and growth stage of the fish;

[0077] Image recognition models include:

[0078] The first processing unit is used to segment the image data into a background area and a target area based on the gray value of the image data and the optimal segmentation threshold.

[0079] The second processing unit: identifies the weight and growth stage of the fish in the segmented target area based on the deep learning network to obtain the weight data and growth stage;

[0080] The deep learning network consists of an input CNN layer, a BiLSTM layer, an attention layer, and an output layer.

[0081] In a specific embodiment, the optimal partitioning threshold is obtained by:

[0082] Collect sample images containing farmed fish and mark background areas and target areas;

[0083] Divide the sample image into K gray levels, and the number of pixels with gray value k is d k The total number of pixels is The probability of each gray value appearing is in

[0084] The grayscale values ​​are divided into two categories: background O = (0, 1, ..., s) and target I = (s + 1, s + 2, ..., K-1). The probabilities of the two categories are:

[0085]

[0086] where p O 、p I are the probabilities of background O and target I appearing respectively; the grayscale means of the two categories are:

[0087]

[0088] Among them, μ O , μ I They are the grayscale mean of background O and target I respectively;

[0089] The total grayscale mean of the sample image for:

[0090]

[0091] The inter-class variance σ between the background area and the target area is:

[0092]

[0093] The optimal partition threshold is expressed as:

[0094]

[0095] Where ρ represents the optimal partition threshold, By marking the sample image and calculating the optimal segmentation threshold, the collected image can be better segmented according to the grayscale value of the collected image data and the optimal segmentation threshold, and the image target area containing fish can be selected, so as to identify the weight and growth stage of the fish in the target area containing fish.

[0096] In a specific embodiment, the filter of the input CNN layer is represented as:

[0097]

[0098] Among them, X i is the i-th output feature extracted, ReLu is the correction unit, F i is the weight matrix of the i-th convolution kernel, x is the input target area image, b is the model bias, * represents the convolution operation, C i is the i-th corresponding learning filter, represents the dot multiplication operation, W is the filter kernel;

[0099] The BiLSTM layer is represented as:

[0100]

[0101] Among them, g t , g t-1 are the hidden states of the forward LSTM output at time t and t-1, respectively, t ,q t-1 are the hidden states of the reverse LSTM output at time t and t-1, respectively, tis the input at time t, h t is the hidden state of the BiLSTM layer at time t, w 1 、w 2 are the weights corresponding to the forward and reverse LSTM, respectively, t is the bias, H is the hidden layer state sequence output by the BiLSTM layer;

[0102] The attention mechanism of the attention layer is expressed as:

[0103]

[0104] Among them, att t is the output of the attention mechanism at time t, w 3 is the weight, h τ is the hidden state τ∈(1,t) of the BiLSTM layer at time τ, and A is the output matrix of the attention mechanism;

[0105] The output layer is represented as:

[0106] y=sigmoid(w A A+b A );

[0107] Among them, y is the output vector, sigmoid is the nonlinear activation function, and w A is the weight, b A For bias.

[0108] The deep learning network collects images of fish at different growth stages and weights and marks the weight and growth stage to obtain training image samples to form a training set; then supervised training is used to obtain a trained deep learning network to identify the target area of ​​the segmented image, confirm the growth stage and weight of the fish, and after identifying multiple fish, take the average to obtain the weight data and growth stage data of the fish in the breeding unit.

[0109] In a specific embodiment, the AI ​​analysis model is obtained by:

[0110] Step 1: Obtain some historical aquaculture data, select historical data whose aquaculture efficiency is greater than a set threshold, and perform a sliding window operation through a sliding window of a set size to obtain multiple sample data;

[0111] Step 2: Arrange the meteorological data, quantity data, weight data, and growth stage of the sample data to obtain the sample's feature vector; Arrange the sample's water quality data and feed feeding amount to obtain the sample's label vector;

[0112] Step 3: Cluster the label vectors to obtain label vector classification data; calculate the average value of the label vectors in each classification data as the standard label vector of the corresponding classification data; replace the label vector of the sample data with the standard label vector of the corresponding classification as the training sample label, and use the feature vector of the sample data as the training sample feature to obtain the training sample set;

[0113] Step 4: Use the gradient boosting decision tree method to construct a decision tree classification model, and use the training sample set to train and test to obtain the AI ​​analysis model.

[0114] Among them, meteorological data include: air temperature, air humidity, atmospheric pressure, light intensity, precipitation; water quality data include: water temperature pH value, dissolved oxygen content, ion concentration, ammonia nitrogen content, turbidity value (particle concentration value);

[0115] The feature vector determines the time length according to the sliding window size, and is obtained by averaging the meteorological data, quantity data, weight data, and growth stage data. The label vector is also obtained by averaging the water quality data and feed feeding amount.

[0116] In a specific embodiment, step 4 specifically includes:

[0117] The training sample set is divided into a training set and a test set according to a preset ratio, the sample labels of the training set are retained and the sample labels of the test set are hidden;

[0118] Initialize the decision tree model classification model parameters;

[0119] The training samples in the training set are input into the decision tree model classification model for supervised learning training to obtain the AI ​​analysis model.

[0120] In a specific embodiment, the decision tree classification model is expressed as:

[0121]

[0122] in, represents the model prediction value, f m represents the mth decision subtree, M is the total number of decision subtrees, represents the decision classification tree, V r represents the feature vector of the input training sample, r is the number of classifications of the label vector classification data, Represents a decision classification tree.

[0123] In a specific embodiment, the loss function in supervised learning training is expressed as:

[0124]

[0125] Among them, L (m)represents the loss function value, l represents the loss function that measures the estimated value and the label value, and R represents the total number of classifications of the label vector classification data. represents the decision classification prediction value before the mth decision subtree, f m (V r ) indicates the predicted value of the mth decision subtree, y r Represents the label vector of the training sample.

[0126] The loss function is Taylor expanded and constants are derived to obtain the following formula:

[0127]

[0128] Among them, J is the total number of leaf nodes, w j is the weight of the jth leaf node, and the optimal weight of the jth leaf node The formula is:

[0129]

[0130] The corresponding optimal solution of the loss function is expressed as:

[0131]

[0132] Among them, g r 、h r ,I j are all process parameters of the expansion; λ and γ are constant terms that are optimized by the particle swarm algorithm to obtain the optimal value, thereby determining the weight value of each leaf node.

[0133] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0134] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. Intelligent operating system for aquaculture management, characterized by: include: Water quality detection module, meteorological detection module, ultrasonic detection module, image detection module, storage module, analysis and control module, input and output module; The water quality detection module detects the water quality data in the breeding unit and uploads it to the storage module; the meteorological detection module detects the meteorological data of the location of the breeding unit and uploads it to the storage module; the ultrasonic detection module detects the quantity data of the fish in the breeding unit and uploads it to the storage module; the image detection module detects the weight data and growth stage of the fish in the breeding unit and uploads it to the storage module; The analysis and control module analyzes the standard water quality and the standard feed feeding amount in the breeding unit through an AI analysis model according to the meteorological data, the quantity data, the weight data, and the growth stage; The analysis and control module compares the difference between the water quality data at the current moment and the standard water quality. When the difference is greater than the set range, the input and output module displays the water quality abnormality and sends a control signal to control the oxygenation equipment to increase oxygen and the water treatment equipment to adjust the water quality; according to the standard feed feeding amount, a control signal is sent to control the feeding system equipment to feed the feed.

2. The intelligent operating system for aquaculture management according to claim 1, characterized in that: The water quality detection module includes: a water temperature sensor, a pH value sensor, a dissolved oxygen sensor, an ion concentration sensor, an ammonia nitrogen sensor, and a turbidity sensor; The meteorological detection module includes: a temperature sensor, a humidity sensor, an air pressure sensor, a light intensity sensor, and a precipitation sensor.

3. The intelligent operating system for aquaculture management according to claim 1, characterized in that: The ultrasonic detection module includes a plurality of ultrasonic detection units, which are evenly installed in the breeding unit to monitor the number of fish within the detection range in real time and calculate the fish density within the detection range; after calculating the average value of the fish density monitored by all the ultrasonic detection units, the breeding quantity estimation value at each moment is obtained according to the area of ​​the breeding unit, and the breeding quantity estimation values ​​at multiple moments are averaged to obtain the quantity data.

4. The intelligent operating system for aquaculture management according to claim 1, characterized in that: The image detection module is arranged at the feeding system device to collect image data of the fish when eating, and uses an image recognition model to identify the weight data and growth stage of the fish; The image recognition model includes: A first processing unit: performing segmentation processing on the image data, and segmenting the image data into a background area and a target area based on the gray value of the image data and an optimal segmentation threshold; A second processing unit: identifying the weight and growth stage of the fish in the segmented target area based on a deep learning network to obtain the weight data and the growth stage; The deep learning network includes an input CNN layer, a BiLSTM layer, an attention layer and an output layer.

5. The intelligent operating system for aquaculture management according to claim 4, characterized in that: The optimal partitioning threshold is obtained by: Collect sample images containing farmed fish and mark background areas and target areas; Divide the sample image into K gray levels, and the number of pixels with gray value k is d k The total number of pixels is The probability of each gray value appearing is in The grayscale values ​​are divided into two categories: background O = (0, 1, ..., s) and target I = (s + 1, s + 2, ..., K-1). The probabilities of the two categories are: where p O 、p I are the probabilities of background O and target I appearing respectively; the grayscale means of the two categories are: Among them, μ O , μ I They are the grayscale mean of background O and target I respectively; The total grayscale mean of the sample image for: The inter-class variance σ between the background area and the target area is: The optimal partition threshold is expressed as: Where ρ represents the optimal partition threshold, Represents a parameter-seeking function.

6. The intelligent operating system for aquaculture management according to claim 4, characterized in that: The filter of the input CNN layer is represented as: Among them, X i is the i-th output feature extracted, ReLu is the correction unit, F i is the weight matrix of the i-th convolution kernel, x is the input target area image, b is the model bias, * represents the convolution operation, C i is the i-th corresponding learning filter, represents the dot multiplication operation, W is the filter kernel; The BiLSTM layer is represented as: Among them, g t , g t-1 are the hidden states of the forward LSTM output at time t and t-1, respectively, t ,q t-1 are the hidden states of the reverse LSTM output at time t and t-1, respectively, t is the input at time t, h t is the hidden state of the BiLSTM layer at time t, w1 and w2 are the weights corresponding to the forward and reverse LSTM respectively, and b t is the bias, H is the hidden layer state sequence output by the BiLSTM layer; The attention mechanism of the attention layer is expressed as: Among them, att t is the output of the attention mechanism at time t, w3 is the weight, h τ is the hidden state τ∈(1,t) of the BiLSTM layer at time τ, and A is the output matrix of the attention mechanism; The output layer is expressed as: y=sigmoid(w A A+b A ); Among them, y is the output vector, sigmoid is the nonlinear activation function, and w A is the weight, b A For bias.

7. The intelligent operating system for aquaculture management according to claim 1, characterized in that: The AI ​​analysis model is obtained in the following way: Step 1: Obtain some historical aquaculture data, select historical data whose aquaculture efficiency is greater than a set threshold, and perform a sliding window operation through a sliding window of a set size to obtain multiple sample data; Step 2: Arrange the meteorological data, quantity data, weight data, and growth stage of the sample data to obtain a feature vector of the sample; arrange the water quality data and feed feeding amount of the sample to obtain a label vector of the sample; Step 3: Clustering the label vectors to obtain the label vector classification data; Calculate the average value of the label vector in each classification data as the standard label vector of the corresponding classification data; replace the label vector of the sample data with the standard label vector of the corresponding classification as the training sample label, and use the feature vector of the sample data as the training sample feature to obtain a training sample set; Step 4: Construct a decision tree classification model using the gradient boosting decision tree method, and use the training sample set to train and test to obtain the AI ​​analysis model.

8. The intelligent operating system for aquaculture management according to claim 7, characterized in that: The step 4 specifically includes: Dividing the training sample set into a training set and a test set according to a preset ratio, retaining the sample labels of the training set and hiding the sample labels of the test set; Initializing the decision tree model classification model parameters; The training samples in the training set are input into the decision tree model classification model for supervised learning training to obtain the AI ​​analysis model.

9. The intelligent operating system for aquaculture management according to claim 7, characterized in that: The decision tree classification model is expressed as: in, represents the model prediction value, f m represents the mth decision subtree, M is the total number of decision subtrees, represents a decision classification tree, V r represents the feature vector of the input training sample, r is the number of classifications of the label vector classification data, Represents a decision classification tree.

10. The intelligent operating system for aquaculture management according to claim 8, characterized in that: The loss function in the supervised learning training is expressed as: Among them, L (m) represents the loss function value, l represents the loss function that measures the estimated value and the label value, and R represents the total number of classifications of the label vector classification data. represents the decision classification prediction value before the mth decision subtree, f m (V r ) indicates the predicted value of the mth decision subtree, y r Represents the label vector of the training sample.

Images