CN-122023058-A - Aquaculture environment abnormality early warning method and system

CN122023058ACN 122023058 ACN122023058 ACN 122023058ACN-122023058-A

Abstract

The invention relates to the technical field of aquaculture monitoring, in particular to an aquaculture environment abnormality early warning method and system, comprising the following steps: and acquiring a water quality electric signal by using a sensor, carrying out three-layer discrete wavelet denoising reconstruction to generate a pure sequence, constructing a multi-dimensional environment state matrix, distributing weights to form a weighted tensor, analyzing time sequence characteristics by using a long-term and short-term memory network, calculating an abnormal probability score, generating an alarm instruction based on a dynamic threshold comparison result, and pushing geographic coordinates. According to the invention, through establishing multi-source data space-time association mapping logic, fusing dissolved oxygen amount, temperature and pH value feature vectors for cluster analysis, eliminating data noise generated by sensor hardware deviation, constructing a dynamic evolution trend prediction model, calculating environmental safety weight in real time, realizing multi-dimensional risk index collaborative research and judgment, and shortening an abnormal state triggering feedback period.

Inventors

JING GUANGZHEN
HU BIN
TANG YUANGUI
WANG WEIGUANG
ZHENG BING
CHEN WENSHOU

Assignees

四川鹏鹞环保设备有限公司

Dates

Publication Date: 20260512
Application Date: 20260415

Claims (10)

1. The method for early warning of the abnormal aquaculture environment is characterized by comprising the following steps: S1, acquiring non-stationary water level fluctuation data through a multi-source physical sensor, acquiring a pipeline inner wall high-definition image sequence by utilizing mobile monitoring equipment, performing sliding window normalization on the non-stationary water level fluctuation data to generate a time sequence feature vector, and performing depth residual convolution on the pipeline inner wall high-definition image sequence to construct an image space feature map; S2, mapping the time sequence feature vector and the image space feature map to a joint high-dimensional characterization subspace, identifying a nonlinear coupling relation between the time sequence feature vector and the image space feature map by using a cross attention unit, and executing semantic alignment based on the nonlinear coupling relation to generate a multi-mode enhanced feature matrix; s3, aggregating multi-level topological features of the multi-mode enhanced feature matrix through a space-time diagram convolution network, extracting pipe network deformation sensitive factors, executing deviation quantization calculation on the pipe network deformation sensitive factors by combining historical reference parameters, and outputting state evaluation bias values; And S4, generating a real-time probability density function for the state evaluation bias value, calculating the Coebeck Leibutler divergence between the real-time probability density function and the standard working condition characteristic model, judging the structural integrity grade of the drainage pipe network based on the Coebeck Leibutler divergence, and outputting a pipe network fault identification instruction.
2. The method for pre-warning of abnormal aquaculture environment according to claim 1, wherein the step of S1 specifically comprises: S11, starting a pressure sensor array arranged at key nodes of an aquaculture area, continuously collecting non-stationary water level fluctuation data caused by tides, start-stop of a pump station and biological activities at a preset high-frequency sampling rate, constructing an original water level data stream comprising a timestamp mark, intercepting a data fragment with a target length through a sliding time window algorithm, performing self-adaptive Z-score standardization processing on the non-stationary water level fluctuation data in the window, eliminating dimension difference and baseline drift influence, and generating a time sequence feature vector; S12, controlling mobile monitoring equipment carrying a high-definition industrial camera to enter a drainage pipeline, moving at a constant speed along the axis of the pipeline in a light source intensity self-adaptive adjustment mode, continuously shooting and acquiring a pipeline inner wall high-definition image sequence, performing multi-layer convolution and pooling operation on each frame of image by using a pre-trained depth residual error network ResNet as a backbone feature extractor, extracting deep space information comprising textures, cracks and corrosion spots, and constructing an image space feature map.
3. The method for pre-warning of abnormal aquaculture environment according to claim 2, wherein the step of S2 specifically comprises: S21, acquiring the time sequence feature vector and the image space feature map, projecting the time sequence feature vector to a vector space of a target dimension through a full connection layer, and simultaneously adjusting the channel number of the image space feature map to be consistent with the vector space dimension by utilizing a 1x1 convolution kernel, so as to establish a joint high-dimensional representation subspace shared by the time sequence feature vector and the image space feature map; s22, calculating the time sequence feature vector as a query vector in the joint high-dimensional characterization subspace through a multi-head cross attention mechanism And the image space feature map is used as a key vector A correlation weight matrix between the two, and a value vector of the image space feature map is obtained by utilizing the weight matrix And carrying out weighted aggregation, identifying a nonlinear coupling relation between the water level fluctuation abnormality and the visual defect of the pipe wall, and aligning the visual features to time steps of a time sequence based on the nonlinear coupling relation to generate a multi-mode enhanced feature matrix.
4. The method for pre-warning of abnormal aquaculture environment according to claim 3, wherein the step of S3 specifically comprises: S31, constructing an adjacency matrix of a graph based on a physical connection topological structure of an aquaculture drainage pipe network, mapping the multi-mode enhanced feature matrix into attribute features of graph nodes, performing synchronous convolution operation on spatial dependence and time evolution modes among the nodes through a stacked multi-layer space-time graph convolution network ST-GCN, aggregating neighborhood node information, eliminating environmental noise interference, and extracting pipe network deformation sensitivity factors; s32, calling historical reference parameters of the drainage pipe network in a healthy state from a cloud database, calculating Euclidean distance and cosine similarity between the pipe network deformation sensitive factors and the historical reference parameters, quantifying the deviation degree of the current state relative to the healthy reference through a weighted fusion algorithm, and outputting a state evaluation bias value.
5. The method for pre-warning of abnormal aquaculture environment according to claim 4, wherein the step S4 specifically comprises: S41, collecting and storing the state evaluation bias numerical sequence in a set time window, fitting the data distribution form of the sequence by utilizing a non-parameterized kernel density estimation method KernelDensityEstimation, calculating the probability distribution condition of the numerical sequence on a continuous domain, and generating a real-time probability density function; S42, a pre-constructed standard working condition characteristic model representing a normal running state statistical rule is called, the relative entropy between the real-time probability density function and the standard working condition characteristic model is calculated by utilizing a Coebeck Brillouin scattering algorithm, the calculated scattering value is compared with a preset multi-level safety threshold, the structural integrity level of the current drainage pipe network is judged according to the comparison result, and a pipe network fault identification instruction is generated.
6. The method of claim 2, wherein the generating the timing feature vector comprises: Set length as Time window and step length of (2) Intercepting an original water level data stream by using a time window to obtain a local data fragment, calculating an arithmetic mean value and a standard deviation of the local data fragment, subtracting the arithmetic mean value from each original water level data point in the local data fragment according to a preset standardized logic, dividing the arithmetic mean value by the sum of the standard deviation and a preset tiny constant to finish normalization operation of the data, and carrying out dimension reduction coding on the normalized data sequence through a one-dimensional convolution layer to generate a time sequence feature vector.
7. A method of pre-warning of an abnormal aquaculture environment according to claim 3, wherein said performing of said semantic alignment comprises: With the cross-attention unit, according to the formula: ; calculating the fusion expression of the cross-modal characteristics, and converting the time sequence characteristic vector into a query matrix through a trained weight parameter matrix Flattening and converting the image space characteristic spectrum into a key matrix Matrix of AND values Capturing semantic association strength between water level change trend and pipe wall image characteristics through dot product operation, carrying out normalization processing by utilizing a Softmax function to obtain attention score, and utilizing the attention score to value matrix Carrying out weighted summation to generate a multi-mode enhanced feature matrix; Wherein, the The result is output representing the semantically aligned attention features, Representing a query matrix generated by linear transformation of the timing feature vector, Representing a key matrix generated by the image space feature map through linear transformation, Representing a matrix of values generated by linear transformation of the image space feature map, Representative key matrix Is a characteristic channel dimension value of (a), Representing the matrix transpose operator, Representing a matrix multiplication operation.
8. The method for pre-warning of abnormal aquaculture environment according to claim 4, wherein the extraction process of the pipe network deformation sensitive factor comprises the following steps: Constructing a space-time diagram structure comprising a space dimension and a time dimension, defining each sensor monitoring point in a drainage pipe network as a space node of a diagram, defining a space edge according to a physical connection relation of a pipeline, defining a time edge according to a front-back sequence of a time sequence, using the multi-mode enhanced feature matrix as an initial input signal of the diagram node, performing filtering operation on the signal on a diagram domain by using a chebyshev polynomial approximation diagram convolution kernel, capturing the high-order correlation of the characteristics of the node and the characteristics of neighboring nodes thereof, and outputting a pipe network deformation sensitive factor through characteristic propagation and aggregation of a multi-layer diagram convolution layer.
9. The method for pre-warning of abnormal aquaculture environment according to claim 5, wherein the outputting process of the pipe network fault identification instruction comprises: According to the formula: ; Calculating the information difference between the current state and the ideal state, and taking the probability distribution discretized by the real-time probability density function as Taking the reference probability distribution corresponding to the standard working condition characteristic model as Calculating a Coebeck Leibutil divergence value between the two, judging the first-level integrity and maintaining normal operation if the divergence value is smaller than a first threshold value, judging the second-level integrity and triggering an early warning prompt if the divergence value is between the first threshold value and a second threshold value, judging the third-level integrity and outputting a pipe network fault identification instruction comprising an emergency pump stopping control signal if the divergence value is larger than the second threshold value; Wherein, the Representing the calculated relative entropy divergence value, Representing the discretized first of the probability distribution The index of the individual section is set, Representing the current time The probability statistics of the individual intervals are calculated, Representing the standard working condition characteristic model in the first The reference probability value of each interval, Representing natural logarithm arithmetic symbols.
10. An aquaculture environment anomaly pre-warning system for implementing an aquaculture environment anomaly pre-warning method according to any one of claims 1 to 9, said system comprising: the data acquisition and feature extraction module is used for acquiring non-stationary water level fluctuation data through a multi-source physical sensor, acquiring a high-definition image sequence of the inner wall of the pipeline by using mobile monitoring equipment, performing sliding window normalization on the non-stationary water level fluctuation data to generate a time sequence feature vector, and performing depth residual convolution on the high-definition image sequence of the inner wall of the pipeline to construct an image space feature map; The multi-mode fusion module is used for mapping the time sequence feature vector and the image space feature map to a joint high-dimensional characterization subspace, identifying a nonlinear coupling relation between the time sequence feature vector and the image space feature map by using a cross attention unit, and executing semantic alignment based on the nonlinear coupling relation to generate a multi-mode enhancement feature matrix; the state evaluation module is used for aggregating multi-level topological features of the multi-mode enhanced feature matrix through a space-time diagram convolution network, extracting pipe network deformation sensitive factors, executing deviation quantization calculation on the pipe network deformation sensitive factors by combining historical reference parameters, and outputting state evaluation bias values; And the early warning decision module is used for generating a real-time probability density function for the state evaluation bias value, calculating the Kolbecchlebecker divergence between the real-time probability density function and the standard working condition characteristic model, judging the structural integrity grade of the drainage pipe network based on the Kolbecchlebecker divergence, and outputting a pipe network fault identification instruction.

Description

Aquaculture environment abnormality early warning method and system Technical Field The invention relates to the technical field of aquaculture monitoring, in particular to an aquaculture environment abnormality early warning method and system. Background The technical field of aquaculture monitoring relates to systematic engineering which utilizes various sensors to collect physical and chemical parameters of water in real time and carries out automatic evaluation on the safety state of the aquaculture environment. The traditional aquaculture environment abnormality early warning method is characterized in that single indexes such as dissolved oxygen, pH value and water temperature are subjected to timing sampling test by means of manual inspection and a handheld detection instrument, and risk prompt signals are sent to management staff through physical alarm lamps or broadcasting when observed values exceed fixed experience thresholds. The existing monitoring equipment is complicated in deployment environment, high-frequency noise interference exists in collected data, a traditional data processing mode only carries out linear threshold judgment aiming at a single water quality parameter, hidden risks generated by coupling of multiple physical quantities cannot be identified, early warning instructions are delayed due to limited link stability in an information transmission process, timely intervention actions are difficult to be executed due to delayed acquisition of risk feedback by management personnel, an effective filtering mechanism for false alarm data generated by sensor offset is lacked, applicability of static judgment standards is poor due to environmental dynamic change characteristics, and finally the farmed organisms cannot be protected when the environment is deteriorated. Disclosure of Invention The invention aims to solve the defects in the prior art and provides an abnormal aquaculture environment early warning method and system. In order to achieve the purpose, the invention adopts the following technical scheme that the method for early warning the abnormal aquaculture environment comprises the following steps: S1, acquiring non-stationary water level fluctuation data through a multi-source physical sensor, acquiring a pipeline inner wall high-definition image sequence by utilizing mobile monitoring equipment, performing sliding window normalization on the non-stationary water level fluctuation data to generate a time sequence feature vector, and performing depth residual convolution on the pipeline inner wall high-definition image sequence to construct an image space feature map; S2, mapping the time sequence feature vector and the image space feature map to a joint high-dimensional characterization subspace, identifying a nonlinear coupling relation between the time sequence feature vector and the image space feature map by using a cross attention unit, and executing semantic alignment based on the nonlinear coupling relation to generate a multi-mode enhanced feature matrix; s3, aggregating multi-level topological features of the multi-mode enhanced feature matrix through a space-time diagram convolution network, extracting pipe network deformation sensitive factors, executing deviation quantization calculation on the pipe network deformation sensitive factors by combining historical reference parameters, and outputting state evaluation bias values; And S4, generating a real-time probability density function for the state evaluation bias value, calculating the Coebeck Leibutler divergence between the real-time probability density function and the standard working condition characteristic model, judging the structural integrity grade of the drainage pipe network based on the Coebeck Leibutler divergence, and outputting a pipe network fault identification instruction. As a further scheme of the present invention, step S1 specifically includes: S11, starting a pressure sensor array arranged at key nodes of an aquaculture area, continuously collecting non-stationary water level fluctuation data caused by tides, start-stop of a pump station and biological activities at a preset high-frequency sampling rate, constructing an original water level data stream comprising a timestamp mark, intercepting a data fragment with a target length through a sliding time window algorithm, performing self-adaptive Z-score standardization processing on the non-stationary water level fluctuation data in the window, eliminating dimension difference and baseline drift influence, and generating a time sequence feature vector; S12, controlling mobile monitoring equipment carrying a high-definition industrial camera to enter a drainage pipeline, moving at a constant speed along the axis of the pipeline in a light source intensity self-adaptive adjustment mode, continuously shooting and acquiring a pipeline inner wall high-definition image sequence, performing multi-layer convolution and pooling operation on each