CN114036846A - Pond culture dissolved oxygen deficiency data interpolation method - Google Patents

Pond culture dissolved oxygen deficiency data interpolation method Download PDF

Info

Publication number
CN114036846A
CN114036846A CN202111346609.0A CN202111346609A CN114036846A CN 114036846 A CN114036846 A CN 114036846A CN 202111346609 A CN202111346609 A CN 202111346609A CN 114036846 A CN114036846 A CN 114036846A
Authority
CN
China
Prior art keywords
data
dissolved oxygen
ssim
model
representing
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111346609.0A
Other languages
Chinese (zh)
Inventor
刘世晶
钱程
涂雪滢
李国栋
汤涛林
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Fishery Machinery and Instrument Research Institute of CAFS
Original Assignee
Fishery Machinery and Instrument Research Institute of CAFS
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Fishery Machinery and Instrument Research Institute of CAFS filed Critical Fishery Machinery and Instrument Research Institute of CAFS
Priority to CN202111346609.0A priority Critical patent/CN114036846A/en
Publication of CN114036846A publication Critical patent/CN114036846A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Activated Sludge Processes (AREA)

Abstract

The invention provides a method for interpolating dissolved oxygen deficiency data in pond culture, which comprises the following steps: s1: acquiring data; s2: preprocessing data; s3: establishing an SSIM model; s4: establishing a dissolved oxygen interpolation model based on PCA _ SSIM by using the SSIM model; s5: and evaluating the dissolved oxygen interpolation model based on the PCA _ SSIM. According to the method for interpolating the dissolved oxygen deficiency data of the pond culture, two artificial activity variable data of feeding and oxygenation are introduced into the time sequence data of the traditional meteorological environment, a PCA analysis method is used for realizing data screening, and an SSIM (structural similarity model) method is used for constructing a pond dissolved oxygen interpolation model so as to realize the interpolation of the dissolved oxygen deficiency data.

Description

Pond culture dissolved oxygen deficiency data interpolation method
Technical Field
The invention relates to the field of data interpolation methods, in particular to a pond culture dissolved oxygen deficiency data interpolation method.
Background
At present, a method for realizing data screening by using a PCA analysis method, constructing a pond dissolved oxygen interpolation model by using an SSIM method and realizing the interpolation of dissolved oxygen deficiency data is lacked.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a method for interpolating dissolved oxygen deficiency data in pond culture, which introduces two artificial activity variable data of feeding and oxygenation into the time sequence data of the traditional meteorological environment, realizes data screening by using a PCA (principal component analysis) method, and realizes the interpolation of the dissolved oxygen deficiency data by constructing a pond dissolved oxygen interpolation model by using an SSIM (structural similarity model) method.
In order to achieve the aim, the invention provides a pond culture dissolved oxygen deficiency data interpolation method, which comprises the following steps:
s1: acquiring data;
s2: preprocessing data;
s3: establishing an SSIM model;
s4: establishing a dissolved oxygen interpolation model based on PCA _ SSIM by using the SSIM model;
s5: and evaluating the dissolved oxygen interpolation model based on the PCA _ SSIM.
Preferably, in the data acquisition step:
the pond is also provided with an impeller aerator and a wind power feeder;
placing a water quality sensor in a buoy form at a position 25m away from the middle line of the long axis of the short axis of the pond, wherein the water quality sensor transmits water quality data through a wireless network, and the water quality data comprises: dissolved oxygen, pH and water temperature; the sampling frequency of the water quality sensor is 5min, and the sampling water depth is 20 m;
transmitting meteorological data from a meteorological station through a wireless network and setting the meteorological data at the upper left corner of the pond; the meteorological data comprise air temperature, air pressure, wind speed, wind direction, atmospheric pressure, illumination intensity and rainfall; the sampling frequency of the weather station is 5 min;
the automatic timing control and environment online monitoring functions of the impeller aerator and the wind power feeder are realized by adopting a preset program, and control data, water quality data and meteorological data are stored in a database according to a standard timestamp.
Preferably, the step of S2 further comprises the steps of:
s21: when the impeller aerator or the wind power feeder is started, calculating an irregular time distribution state value by adopting a formula (1); when the impeller aerator and the wind power feeder are closed, calculating the distribution state value of irregular time by adopting a formula (2);
Figure BDA0003354224990000021
Figure BDA0003354224990000022
wherein, OiRepresenting the time interval state when the impeller aerator or the wind power feeder is started; ciRepresenting the time interval state when the impeller aerator and the wind power feeder are closed; t isi↑Is greater than the opening time TiAnd the sampling time point, T, of the water quality sensor closest to the opening timei↓To be less than the opening time TiAnd the sampling time point of the water quality sensor closest to the starting time; i is the switching times of the equipment;
s22: repairing the meteorological data and the water quality data with missing values less than or equal to 2 sampling time intervals by adopting a linear interpolation method;
s23: detecting and repairing abnormal values;
s24: and carrying out normalization processing on the meteorological data and the water quality data.
Preferably, in the step S24, the data is normalized by formula (3):
Figure BDA0003354224990000023
wherein x isiDenotes the feature vector at a particular time length index i, min denotes the minimum value in the data, and max denotes the maximum value in the data.
Preferably, the SSIM model uses bidirectional LSTM as an encoder;
the input to the SSIM model is a variable length sequence x ═ xiIn which xiIndexing the feature vector at i for a particular length of time; hidden states at each time index t are
Figure BDA0003354224990000031
Wherein the content of the first and second substances,
Figure BDA0003354224990000032
a forward hidden state is represented that is,
Figure BDA0003354224990000033
indicating a reverse hidden state; constructing a hidden sequence h ═ { ht };
the hidden state at time index t is updated as follows:
int＝σ(Wxinxt+Whinht-1+bin)
ft＝σ(Wxfxt+Whfht-1+bf)
Figure BDA0003354224990000034
ot＝σ(Wxoxt+Whoht-1+bo)
Figure BDA0003354224990000035
therein, intRepresenting the input gate at time index t, ftIndicating a forgetting gate at time index t, otRepresents the output gate at time index t; c. CtRepresenting a time indexA unit vector at t; sigma represents a sigmoid function;
Figure BDA0003354224990000036
representing element multiplication; the first parameter W and the second parameter b map the cascade to the size of the hidden state; h istRepresenting hidden states of the model output; wxinRepresenting an input gate input weight; x is the number oftRepresenting input gate input at time t; whinRepresenting input layer hidden state weights; binRepresenting input gate bias; wxfRepresenting forgetting layer input weights; whfRepresenting a forgotten door hidden state weight; bfThe representation represents a forgetting gate bias; wxcRepresenting hidden state weights at the time index; whcRepresenting a time indexed hidden state weight; bcRepresents a bias at the time index; wxoRepresenting output gate input weights; whoRepresenting an output gate hidden state weight; boRepresents the output gate offset;
the decoder of the SSIM model is responsible for the recursive generation of the output sequence y ═ { y1, y2, …, ym-adding a fully connected layer with a linear activation function on top of the LSTM layer to generate a prediction with continuous values; for the decoder, the estimate y at time index ttThe calculation is as follows:
yt＝Linear(W[st,ct]+b)
st＝LSTM(yt-1,st-1,ct)
wherein s istIs the hidden state of the decoder at time index t, ct is the weighted sum of the hidden states passed by the encoder, [ st, ct]Is a concatenation of decoder hidden state and context vector; LSTM represents decoding operation by using LSTM network; linear denotes a recursion-based fully-connected operation.
Preferably, the step of S4 further comprises the steps of:
s41: acquiring the meteorological data, the water quality data, the data of the impeller aerator and the data of the wind power feeder by constructing a pond culture internet of things system;
s42: carrying out standardization and normalization pretreatment on the meteorological data and the water quality data according to the time series analysis requirements;
s43: selecting key influence factors based on a PCA method, screening the meteorological data and the water quality data by using the key influence factors, and constructing a training set and a testing set;
s44: training the SSIM model by using the training set until the output of the SSIM model reaches the target accuracy rate, and obtaining the PCA _ SSIM-based dissolved oxygen interpolation model;
s45: and verifying the accuracy of the PCA _ SSIM-based dissolved oxygen interpolation model by using the test set, and verifying the validity of the PCA _ SSIM-based dissolved oxygen interpolation model.
Preferably, in the step S5, the PCA _ SSIM-based dissolved oxygen interpolation model is evaluated by using a root mean square error, a mean absolute percentage error and a symmetric mean absolute percentage error.
Preferably, the root mean square error RMSE, the mean absolute error MAE, the mean absolute percent error MAPE, and the symmetric mean absolute percent error SMAPE satisfy the formula:
Figure BDA0003354224990000041
Figure BDA0003354224990000042
Figure BDA0003354224990000043
Figure BDA0003354224990000044
wherein, yiIn order to measure the dissolved oxygen content,
Figure BDA0003354224990000051
for prediction, N is the number of missing samples.
Preferably, in the step S43, a PCA method is used to implement data dimension reduction and analysis, an orthogonal rotation method based on Kaiser standardization is used to determine an influence factor load matrix, and finally the dissolved oxygen key influence factor of the pond is screened out.
Preferably, the SSIM model uses a pytouch neural network framework;
the number of encoder hidden layers of the SSIM model is 2;
the number of decoder hidden layers of the SSIM model is 2;
the number of LSTM hidden units in each layer of the SSIM model is 30;
the loss rate of the SSIM model is 0.3;
the optimizer of the SSIM model adopts Adam;
the loss function of the SSIM model adopts MSE;
the skin size of the SSIM model was 60.
Due to the adoption of the technical scheme, the invention has the following beneficial effects:
according to the method, two artificial activity variable data of feeding and oxygenation are introduced into the conventional meteorological environment time sequence data, a PCA analysis method is used for realizing data screening, and an SSIM method is used for constructing a pond dissolved oxygen interpolation model so as to realize dissolved oxygen deficiency data interpolation.
Drawings
FIG. 1 is a flow chart of a method for interpolating dissolved oxygen deficiency data in pond culture according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a network structure of an SSIM model according to an embodiment of the present invention;
fig. 3 is a flowchart for establishing a dissolved oxygen interpolation model based on PCA _ SSIM according to an embodiment of the present invention.
Detailed Description
The following description of the preferred embodiments of the present invention will be provided in conjunction with the accompanying drawings, which are set forth in detail below to provide a better understanding of the function and features of the invention.
Referring to fig. 1 to 3, a method for interpolating dissolved oxygen deficiency data in pond culture according to an embodiment of the present invention includes:
s1: acquiring data;
in the data acquisition step:
a standardized and improved pond (100 m in length, 30m in width and 1.5m in average depth) which is used for more than 5 years is selected as a test object, cement prefabricated slabs are adopted around the pond for slope protection, and the culture variety is grass carp. The pond is also provided with a 3kw impeller aerator and a 200w pneumatic feeder.
In order to avoid oxygenation and the process of throwing something and feeding to cause the influence to who water quality monitoring process, place water quality sensor at pond minor axis line distance major axis central line 25m department in adopting the buoy form, water quality sensor passes through wireless network transmission quality of water data, and the quality of water data includes: dissolved oxygen, pH and water temperature; the sampling frequency of the water quality sensor is 5min, and the sampling water depth is 20 m.
Transmitting meteorological data by a meteorological station through a wireless network and arranging the meteorological data at the upper left corner of the pond; the meteorological data comprise air temperature, air pressure, wind speed, wind direction, atmospheric pressure, illumination intensity and rainfall; the sampling frequency of the weather station is 5 min.
The automatic control of the impeller aerator and the pneumatic feeder and the on-line environment monitoring function are realized by adopting a preset program, and the control data, the water quality data and the meteorological data are stored in the database according to the standard timestamp.
S2: preprocessing data;
in the actual sampling process, the sampling frequency of the water quality sensor is artificially fixed, and the collected water quality data and the collected meteorological data have the same distribution characteristics. The impeller aerator and the wind power feeder are used as two common breeding production mechanical devices, are constrained by using conditions and price cost, generally do not use a motor assembly with a frequency conversion function, and only show an opening state and a closing state on a time axis according to data characteristics, so that the data time characteristics of the two devices are represented by adopting a 0 state and a 1 state in a data set.
The step of S2 further includes the steps of:
s21: in order to keep the generality, the switching time of the equipment is set according to the actual requirement of aquaculture production, and when an impeller aerator or a wind power feeder is started, the distribution state value of irregular time is calculated by adopting a formula (1); when the impeller aerator and the wind power feeder are closed, calculating the distribution state value of irregular time by adopting a formula (2);
Figure BDA0003354224990000061
Figure BDA0003354224990000062
wherein, OiThe time interval state when the impeller aerator or the wind power feeder is started is represented; ciThe time interval state of the impeller aerator and the wind power feeder is shown when the impeller aerator and the wind power feeder are closed; t isi↑Is greater than the opening time TiAnd the sampling time point, T, of the water quality sensor closest to the opening timei↓To be less than the opening time TiAnd the sampling time point of the water quality sensor closest to the starting time; i is the switching times of the equipment;
s22: repairing meteorological data and water quality data with missing values less than or equal to 2 sampling time intervals by adopting a linear interpolation method;
s23: detecting and repairing abnormal values;
s24: and carrying out normalization processing on the meteorological data and the water quality data.
In the step S24, the data is normalized by the formula (3):
Figure BDA0003354224990000071
wherein x isiDenotes the feature vector at a particular time length index i, min denotes the minimum value in the data, and max denotes the maximum value in the data.
S3: establishing an SSIM model;
the encoder and decoder in the SSIM model are two key functional components. The encoder processes the input time series and maps it to a high-dimensional vector. The decoder takes input from the vector and generates a target data sequence. Furthermore, the attention mechanism enables the decoder to know how to focus on a particular input sequence range of different outputs.
Unlike the general prediction problem based on the past data, data interpolation is to repair missing data and can reconstruct missing values using the past and future data. Therefore, the SSIM model uses bidirectional LSTM (BiLSTM) as an encoder to fully utilize existing data.
The input to the SSIM model is a variable length sequence x ═ xiIn which xiIndexing the feature vector at i for a particular length of time; unlike the unidirectional LSTM, the BiLSTM has two hidden states, forward and backward, with the hidden state at each time index t being
Figure BDA0003354224990000072
Wherein the content of the first and second substances,
Figure BDA0003354224990000073
a forward hidden state is represented that is,
Figure BDA0003354224990000074
indicating a reverse hidden state; constructing a hidden sequence h ═ { ht };
the hidden state at time index t is updated as follows:
int＝σ(Wxinxt+Whinht-1+bin)
ft＝σ(Wxfxt+Whfht-1+bf)
Figure BDA0003354224990000075
ot＝σ(Wxoxt+Whoht-1+bo)
Figure BDA0003354224990000081
therein, intRepresenting the input gate at time index t, ftIndicating a forgetting gate at time index t, otRepresents the output gate at time index t; c. CtRepresents the unit vector at time index t; sigma represents a sigmoid function;
Figure BDA0003354224990000082
representing element multiplication; the first parameter W and the second parameter b map the cascade to the size of the hidden state; h istRepresenting hidden states of the model output; wxinRepresenting an input gate input weight; x is the number oftRepresenting input gate input at time t; whinRepresenting input layer hidden state weights; binRepresenting input gate bias; wxfRepresenting forgetting layer input weights; whfRepresenting a forgotten door hidden state weight; bfThe representation represents a forgetting gate bias; wxcRepresenting hidden state weights at the time index; whcRepresenting a time indexed hidden state weight; bcRepresents a bias at the time index; wxoRepresenting output gate input weights; whoRepresenting an output gate hidden state weight; boRepresents the output gate offset;
the decoder of the SSIM model is responsible for the recursive generation of the output sequence y ═ y1, y2, …, ym-adding a fully connected layer with a linear activation function on top of the LSTM layer to generate a prediction with continuous values; for the decoder, the estimate y at time index ttThe calculation is as follows:
yt＝Linear(W[st,ct]+b)
st＝LSTM(yt-1,st-1,ct)
wherein s istIs the hidden state of the decoder at time index t, ct is the weighted sum of the hidden states passed by the encoder, [ st, ct]Is a concatenation of decoder hidden state and context vector; LSTM means LiCarrying out decoding operation by using an LSTM network; linear denotes a recursion-based fully-connected operation.
S4: establishing a dissolved oxygen interpolation model based on PCA _ SSIM by using an SSIM model;
the step of S4 further includes the steps of:
s41: acquiring meteorological data, water quality data, data of an impeller aerator and data of a wind power feeder by constructing a pond culture internet of things system;
s42: carrying out standardization and normalization pretreatment on meteorological data and water quality data according to the time series analysis requirements;
s43: selecting key influence factors based on a PCA method, screening meteorological data and water quality data by using the key influence factors, and constructing a training set and a testing set;
s44: training the SSIM model by using a training set until the output of the SSIM model reaches a target accuracy rate, and obtaining a dissolved oxygen interpolation model based on PCA _ SSIM;
s45: and verifying the accuracy of the dissolved oxygen interpolation model based on the PCA _ SSIM by using the test set, and verifying the effectiveness of the dissolved oxygen interpolation model based on the PCA _ SSIM.
S5: the dissolved oxygen interpolation model based on PCA _ SSIM was evaluated.
In the step S5, the PCA _ SSIM-based dissolved oxygen interpolation model is evaluated using the root mean square error, the mean absolute percentage error, and the symmetric mean absolute percentage error.
The root mean square error RMSE, the mean absolute error MAE, the mean absolute percent error MAPE, and the symmetric mean absolute percent error SMAPE satisfy the formula:
Figure BDA0003354224990000091
Figure BDA0003354224990000092
Figure BDA0003354224990000093
Figure BDA0003354224990000094
wherein, yiIn order to measure the dissolved oxygen content,
Figure BDA0003354224990000095
for prediction, N is the number of missing samples.
In the step S43, a PCA method is used for achieving data dimension reduction and analysis, an orthogonal rotation method based on Kaiser standardization is adopted for determining an influence factor load matrix, and finally dissolved oxygen key influence factors of the pond are screened out.
The influence factors of the dissolved oxygen content of the pond are numerous, and if all data are not directly utilized, the model is huge, and the difficulty in data analysis and utilization is increased. Therefore, in order to search for the key factors influencing the change of the dissolved oxygen in the pond, the influence factors need to be subjected to composition analysis so as to reduce the associated variables and define the main influence factors. In the embodiment, the PCA method is utilized to realize data dimension reduction and analysis, the orthogonal rotation method based on Kaiser standardization is adopted to determine the influence factor load matrix, and finally the key influence factors of pond dissolved oxygen are screened out. Dissolved oxygen, pH value, water temperature, wind speed, illumination intensity, an aerator and a feeder are selected as main influence factors to construct an interpolation model, and the selection of related factors is basically consistent with the selection result based on expert experience.
In this embodiment, the SSIM model uses a pytouch neural network framework;
the number of encoder hidden layers of the SSIM model is 2;
the number of decoder hidden layers of the SSIM model is 2;
the number of LSTM hidden units in each layer of the SSIM model is 30;
the loss rate of the SSIM model is 0.3;
an optimizer of the SSIM model adopts Adam;
the loss function of the SSIM model adopts MSE;
the skin size of the SSIM model was 60.
While the present invention has been described in detail and with reference to the embodiments thereof as illustrated in the accompanying drawings, it will be apparent to one skilled in the art that various changes and modifications can be made therein. Therefore, certain details of the embodiments are not to be interpreted as limiting, and the scope of the invention is to be determined by the appended claims.

Claims (10)

1. A method for interpolating dissolved oxygen deficiency data of pond culture comprises the following steps:
s1: acquiring data;
s2: preprocessing data;
s3: establishing an SSIM model;
s4: establishing a dissolved oxygen interpolation model based on PCA _ SSIM by using the SSIM model;
s5: and evaluating the dissolved oxygen interpolation model based on the PCA _ SSIM.
2. The pond culture dissolved oxygen deficiency data interpolation method according to claim 1, wherein in the data acquisition step:
the pond is also provided with an impeller aerator and a wind power feeder;
placing a water quality sensor in a buoy form at a position 25m away from the middle line of the long axis of the short axis of the pond, wherein the water quality sensor transmits water quality data through a wireless network, and the water quality data comprises: dissolved oxygen, pH and water temperature; the sampling frequency of the water quality sensor is 5min, and the sampling water depth is 20 m;
transmitting meteorological data from a meteorological station through a wireless network and setting the meteorological data at the upper left corner of the pond; the meteorological data comprise air temperature, air pressure, wind speed, wind direction, atmospheric pressure, illumination intensity and rainfall; the sampling frequency of the weather station is 5 min;
the automatic timing control and environment online monitoring functions of the impeller aerator and the wind power feeder are realized by adopting a preset program, and control data, water quality data and meteorological data are stored in a database according to a standard timestamp.
3. The method for interpolating dissolved oxygen deficiency data in pond culture according to claim 2, wherein the step of S2 further comprises the steps of:
s21: when the impeller aerator or the wind power feeder is started, calculating an irregular time distribution state value by adopting a formula (1); when the impeller aerator and the wind power feeder are closed, calculating the distribution state value of irregular time by adopting a formula (2);
Figure FDA0003354224980000011
Figure FDA0003354224980000012
wherein, OiRepresenting the time interval state when the impeller aerator or the wind power feeder is started; ciRepresenting the time interval state when the impeller aerator and the wind power feeder are closed; t isi↑Is greater than the opening time TiAnd the sampling time point, T, of the water quality sensor closest to the opening timei↓To be less than the opening time TiAnd the sampling time point of the water quality sensor closest to the starting time; i is the switching times of the equipment;
s22: repairing the meteorological data and the water quality data with missing values less than or equal to 2 sampling time intervals by adopting a linear interpolation method;
s23: detecting and repairing abnormal values;
s24: and carrying out normalization processing on the meteorological data and the water quality data.
4. The method for interpolating dissolved oxygen deficiency data in pond culture according to claim 3, wherein in the step S24, the data is normalized by the formula (3):
Figure FDA0003354224980000021
wherein x isiDenotes the feature vector at a particular time length index i, min denotes the minimum value in the data, and max denotes the maximum value in the data.
5. The pond culture dissolved oxygen deficiency data interpolation method according to claim 4, wherein the SSIM model employs bidirectional LSTM as an encoder;
the input to the SSIM model is a variable length sequence x ═ xiIn which xiIndexing the feature vector at i for a particular length of time; hidden states at each time index t are
Figure FDA0003354224980000022
Wherein the content of the first and second substances,
Figure FDA0003354224980000023
a forward hidden state is represented that is,
Figure FDA0003354224980000024
indicating a reverse hidden state; constructing a hidden sequence h ═ { ht };
the hidden state at time index t is updated as follows:
int＝σ(Wxinxt+Whinht-1+bin)
ft＝σ(Wxfxt+Whfht-1+bf)
Figure FDA0003354224980000025
ot＝σ(Wxoxt+Whoht-1+bo)
Figure FDA0003354224980000026
therein, intRepresenting the input gate at time index t, ftIndicating a forgetting gate at time index t, otRepresents the output gate at time index t; c. CtRepresents the unit vector at time index t; sigma represents a sigmoid function;
Figure FDA0003354224980000031
representing element multiplication; the first parameter W and the second parameter b map the cascade to the size of the hidden state; h istRepresenting hidden states of the model output; wxinRepresenting an input gate input weight; x is the number oftRepresenting input gate input at time t; whinRepresenting input layer hidden state weights; binRepresenting input gate bias; wxfRepresenting forgetting layer input weights; whfRepresenting a forgotten door hidden state weight; bfThe representation represents a forgetting gate bias; wxcRepresenting hidden state weights at the time index; whcRepresenting a time indexed hidden state weight; bcRepresents a bias at the time index; wxoRepresenting output gate input weights; whoRepresenting an output gate hidden state weight; boRepresents the output gate offset;
the decoder of the SSIM model is responsible for the recursive generation of the output sequence y ═ { y1, y2m-adding a fully connected layer with a linear activation function on top of the LSTM layer to generate a prediction with continuous values; for the decoder, the estimate y at time index ttThe calculation is as follows:
yt＝Linear(W[st，ct]+b)
st＝LSTM(yt-1，st-1，ct)
wherein s istIs the hidden state of the decoder at time index t, ct is the weighted sum of the hidden states passed by the encoder, [ st, ct]Is the decoder hidingConcatenation of state and context vectors; LSTM represents decoding operation by using LSTM network; linear denotes a recursion-based fully-connected operation.
6. The method for interpolating dissolved oxygen deficiency data in pond culture according to claim 1, wherein the step of S4 further comprises the steps of:
s41: acquiring the meteorological data, the water quality data, the data of the impeller aerator and the data of the wind power feeder by constructing a pond culture internet of things system;
s42: carrying out standardization and normalization pretreatment on the meteorological data and the water quality data according to the time series analysis requirements;
s43: selecting key influence factors based on a PCA method, screening the meteorological data and the water quality data by using the key influence factors, and constructing a training set and a testing set;
s44: training the SSIM model by using the training set until the output of the SSIM model reaches the target accuracy rate, and obtaining the PCA _ SSIM-based dissolved oxygen interpolation model;
s45: and verifying the accuracy of the PCA _ SSIM-based dissolved oxygen interpolation model by using the test set, and verifying the validity of the PCA _ SSIM-based dissolved oxygen interpolation model.
7. The method for interpolating dissolved oxygen deficiency data in pond culture according to claim 1, wherein the step S5 is performed to evaluate the PCA _ SSIM-based dissolved oxygen interpolation model by using root mean square error, mean absolute percentage error and symmetric mean absolute percentage error.
8. The method for interpolating dissolved oxygen deficiency data in pond culture according to claim 7, wherein the root mean square error RMSE, the mean absolute error MAE, the mean absolute percent error MAPE and the symmetric mean absolute percent error SMAPE satisfy the following formula:
Figure FDA0003354224980000041
Figure FDA0003354224980000042
Figure FDA0003354224980000043
Figure FDA0003354224980000044
wherein, yiIn order to measure the dissolved oxygen content,
Figure FDA0003354224980000045
for prediction, N is the number of missing samples.
9. The method for interpolating dissolved oxygen deficiency data in pond culture according to claim 6, wherein in the step S43, a PCA method is used for realizing data dimension reduction and analysis, an orthogonal rotation method based on Kaiser standardization is used for determining an influence factor load matrix, and finally the key influence factors of the dissolved oxygen in the pond are screened out.
10. The pond culture dissolved oxygen deficiency data interpolation method according to claim 1, wherein the SSIM model uses a pytouch neural network framework;
the number of encoder hidden layers of the SSIM model is 2;
the number of decoder hidden layers of the SSIM model is 2;
the number of LSTM hidden units in each layer of the SSIM model is 30;
the loss rate of the SSIM model is 0.3;
the optimizer of the SSIM model adopts Adam;
the loss function of the SSIM model adopts MSE;
the skin size of the SSIM model was 60.
CN202111346609.0A 2021-11-15 2021-11-15 Pond culture dissolved oxygen deficiency data interpolation method Pending CN114036846A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111346609.0A CN114036846A (en) 2021-11-15 2021-11-15 Pond culture dissolved oxygen deficiency data interpolation method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111346609.0A CN114036846A (en) 2021-11-15 2021-11-15 Pond culture dissolved oxygen deficiency data interpolation method

Publications (1)

Publication Number Publication Date
CN114036846A true CN114036846A (en) 2022-02-11

Family

ID=80144349

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111346609.0A Pending CN114036846A (en) 2021-11-15 2021-11-15 Pond culture dissolved oxygen deficiency data interpolation method

Country Status (1)

Country Link
CN (1) CN114036846A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116992295A (en) * 2023-09-26 2023-11-03 北京宝隆泓瑞科技有限公司 Reconstruction method and device for machine pump equipment monitoring missing data for machine learning

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116992295A (en) * 2023-09-26 2023-11-03 北京宝隆泓瑞科技有限公司 Reconstruction method and device for machine pump equipment monitoring missing data for machine learning

Similar Documents

Publication Publication Date Title
CN110084367B (en) Soil moisture content prediction method based on LSTM deep learning model
CN112633604B (en) Short-term power consumption prediction method based on I-LSTM
CN103023065B (en) Wind power short-term power prediction method based on relative error entropy evaluation method
CN109143408B (en) Dynamic region combined short-time rainfall forecasting method based on MLP
CN108428017A (en) Wind power interval prediction method based on core extreme learning machine quantile estimate
CN114707688A (en) Photovoltaic power ultra-short-term prediction method based on satellite cloud chart and space-time neural network
CN113255972B (en) Short-term rainfall prediction method based on Attention mechanism
CN113837499A (en) Ultra-short-term wind power prediction method and system
CN114077931A (en) Method for predicting content of dissolved oxygen in aquaculture pond
CN109816167A (en) Runoff Forecast method and Runoff Forecast device
CN114154619A (en) Ship track prediction method based on CNN and BILSTM
CN113469427A (en) Convolution LSTM-based photovoltaic power station day-ahead power generation amount prediction method and system
CN115237896A (en) Data preprocessing method and system for forecasting air quality based on deep learning
CN114036846A (en) Pond culture dissolved oxygen deficiency data interpolation method
CN116757057A (en) Air quality prediction method based on PSO-GA-LSTM model
CN117313990A (en) Air conditioner load prediction method and system based on meta-integrated learning and electronic equipment
CN116470486A (en) Ultra-short-term photovoltaic power generation power prediction method, device and storage medium
CN115587713A (en) Marine ranch disaster decision method based on reinforcement learning
CN112819244B (en) Meteorological factor-based RF-HW water quality index hybrid prediction method
CN113762591B (en) Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning
CN115759343A (en) E-LSTM-based user electric quantity prediction method and device
CN115730744A (en) Water consumption prediction method and system based on user mode and deep learning combined model
CN114648095A (en) Air quality concentration inversion method based on deep learning
CN109840613B (en) Short-term wind speed prediction method integrating coding and decoding and linear regression
CN114372615A (en) Short-term photovoltaic power prediction method and system considering spatial correlation

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination