Measurement error data self-marking method of multi-parameter temperature-salinity-depth instrument

Abstract

The invention provides a measurement error data self-marking method of a multi-parameter temperature-salinity-depth meter, and belongs to the technical field of electrical digital data processing.The method comprises the steps that firstly, an acquisition curve function matrix is established, time window segmentation processing is conducted on newly-sampled data, and the deviation value of each parameter segment and an acquisition curve is calculated; then constructing a three-parameter association network diagram, extracting a key parameter association path, and establishing a multi-parameter relationship coupling equation set; calculating a theoretical coupling value and an actual coupling value to form a coupling deviation matrix; a mixed expert neural network model based on a multi-head sparse attention mechanism is further adopted to replace a traditional error judgment function, and the deviation value and the coupling deviation matrix are input into the model to calculate an error judgment value; and finally, comparing the error judgment value with a preset threshold value, and marking the abnormal data points. According to the method, by comprehensively considering physical correlation among parameters and applying a deep learning technology, the problem that the accuracy of automatic error identification and marking of measured data of the thermohaline depth gauge is insufficient is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of electronic digital data processing, and in particular relates to a self-marking method for measurement error data of a multi-parameter temperature-salinity-depth instrument. Background Art

[0002] In the field of ocean observation, temperature, salinity, and depth instruments are crucial for obtaining key parameters such as seawater temperature, salinity, and depth. Traditional temperature, salinity, and depth data processing techniques rely primarily on single-parameter threshold determination or simple statistical models to identify errors. These techniques typically employ methods such as historical data statistical analysis, interval determination, and variance analysis to screen outliers in the collected data. These methods are effective in identifying measurement errors that significantly deviate from normal values within a single parameter range.

[0003] However, traditional techniques have significant flaws. First, single-parameter judgment methods struggle to handle the complex coupling relationships between parameters, ignoring the physical connections between temperature, salinity, and depth. Second, simple statistical models are insensitive to marginal outliers, making it easy to misjudge normal fluctuations as anomalies or anomalous data as normal, especially in volatile ocean environments. Furthermore, fixed threshold judgments lack adaptability and struggle to cope with the changing characteristics of data across different sea areas and seasons.

[0004] As the demand for ocean observation accuracy continues to increase, existing technologies struggle to identify errors in data measured by multi-parameter temperature, salinity, and depth instruments in complex ocean environments. In particular, it's difficult to accurately identify measurement errors caused by abnormal coupling relationships between multiple parameters, hindering the efficiency and accuracy of ocean observation data quality control. There is an urgent need to develop an error identification method that comprehensively considers the physical relationships between parameters. In other words, existing technologies lack the accuracy to automatically identify and label errors in measurement data from multi-parameter temperature, salinity, and depth instruments. Summary of the Invention

[0005] In view of this, the present invention provides a self-marking method for measurement error data of a multi-parameter temperature, salinity and depth instrument, which can solve the technical problem of insufficient accuracy in automatic identification and marking of measurement data errors of a multi-parameter temperature, salinity and depth instrument in the prior art.

[0006] The present invention is implemented as follows: the present invention provides a self-marking method for measurement error data of a multi-parameter temperature-salinity-depth instrument, comprising the following steps: establishing a temperature parameter acquisition curve function matrix, a salinity parameter acquisition curve function matrix, and a depth parameter acquisition curve function matrix; performing segmented processing on newly sampled data according to a time window to form a temperature parameter segment matrix, a salinity parameter segment matrix, and a depth parameter segment matrix; respectively calculating deviation values between the temperature parameter segment matrix and the temperature parameter acquisition curve function matrix, deviation values between the salinity parameter segment matrix and the salinity parameter acquisition curve function matrix, and deviation values between the depth parameter segment matrix and the depth parameter acquisition curve function matrix; constructing a parameter association network diagram between the temperature parameter, the salinity parameter, and the depth parameter, extracting key parameter association paths using a minimum spanning tree algorithm, and establishing a multi-parameter relationship coupling equation group; calculating theoretical coupling values and actual coupling values using the multi-parameter relationship coupling equation group to obtain a coupling deviation matrix; inputting the deviation values and the coupling deviation matrix into a temperature-salinity-depth error marking neural network model to calculate an error judgment value; and comparing the error judgment value with a preset threshold to mark error data.

[0007] Among them, the parameter acquisition curve function matrix is a matrix composed of polynomial functions obtained by fitting historical measurement data through the least squares method, which is used to describe the regular characteristics of parameter changes over time or space.

[0008] Among them, the time window refers to dividing the continuously collected data into multiple data segments according to fixed time intervals. Each data segment contains several continuous sampling points, which are used for local analysis of parameter change characteristics.

[0009] The deviation value refers to the difference between the measured data and the predicted value of the theoretical model, which is quantified by calculating the Euclidean distance or Manhattan distance.

[0010] Among them, the multi-parameter relationship coupling equation group includes the temperature-salinity relationship equation, the temperature-depth relationship equation, the salt-depth relationship equation and the temperature-salinity-depth comprehensive relationship equation.

[0011] Among them, the temperature-salinity relationship equation is used to describe the physical correlation between temperature and salinity. The input includes the temperature parameter segment matrix, the salinity parameter segment matrix, the temperature-salinity history correlation database, the ambient temperature change rate and the seawater density parameter. The output is the temperature-salinity theoretical coupling value matrix.

[0012] Among them, the temperature-depth relationship equation is used to describe the physical correlation between temperature and depth. The input includes the temperature parameter segment matrix, the depth parameter segment matrix, the temperature-depth historical correlation database, the depth gradient coefficient, and the water pressure change rate. The output is the temperature-depth theoretical coupling value matrix.

[0013] Among them, the salt-depth relationship equation is used to describe the physical correlation between salinity and depth. The input includes the salinity parameter segment matrix, the depth parameter segment matrix, the salt-depth historical correlation database, the depth stratification coefficient and the ocean current influencing factor. The output is the salt-depth theoretical coupling value matrix.

[0014] Among them, the temperature-salinity-depth comprehensive relationship equation is used to integrate the comprehensive physical correlation between the three parameters. The input includes the temperature parameter segment matrix, the salinity parameter segment matrix, the depth parameter segment matrix, the temperature-salinity-depth historical correlation database and the seawater state equation parameters. The output is the three-parameter comprehensive theoretical coupling value matrix.

[0015] Among them, the theoretical coupling values include the temperature-salinity theoretical coupling value matrix, the temperature-depth theoretical coupling value matrix, the salt-depth theoretical coupling value matrix, and the three-parameter comprehensive theoretical coupling value matrix; the actual coupling values refer to the actual mutual influence values between parameters directly calculated from the actual measurement data, including the temperature-salinity actual coupling value matrix, the temperature-depth actual coupling value matrix, the salt-depth actual coupling value matrix, and the three-parameter comprehensive actual coupling value matrix.

[0016] Compared with the existing technology, the present invention provides a self-marking method for measurement error data of a multi-parameter temperature, salinity and depth instrument. The present invention proposes a self-marking method for measurement error data of a multi-parameter temperature, salinity and depth instrument. By establishing a parameter acquisition curve function matrix, constructing a parameter association network diagram, extracting key parameter association paths, establishing a multi-parameter relationship coupling equation group, and combining it with a neural network model, high-precision automatic marking of temperature, salinity and depth data errors is achieved.

[0017] This method overcomes the shortcomings of traditional technologies. By introducing a set of multi-parameter relationship coupling equations, it fully considers the physical correlation between temperature, salinity and depth, so that error identification is no longer limited to the threshold judgment of a single parameter. By adopting a hybrid expert model based on a multi-head sparse attention mechanism to replace the traditional error judgment function, the system's sensitivity to edge outliers is improved. Through the pre-trained neural network model, the method's adaptability is enhanced, and it can automatically adjust the error judgment criteria according to the data characteristics of different sea areas and seasons.

[0018] The present invention successfully solves the technical problem of insufficient accuracy in automatic identification and marking of measurement data errors of multi-parameter temperature-salinity-depth instruments, significantly improves the reliability and accuracy of ocean observation data, provides a more reliable data basis for marine scientific research and marine environmental monitoring, and has important application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0020] like Figure 1FIG. 1 is a flow chart of a method for self-marking measurement error data of a multi-parameter temperature-salinity-depth instrument provided by the present invention. The method comprises the following steps: S01, establishing a temperature parameter acquisition curve function matrix, a salinity parameter acquisition curve function matrix, and a depth parameter acquisition curve function matrix; S02, segmenting the newly sampled data according to the time window to form a temperature parameter segment matrix, a salinity parameter segment matrix, and a depth parameter segment matrix; S03, respectively calculating the deviation values between the temperature parameter segment matrix and the temperature parameter acquisition curve function matrix, the deviation values between the salinity parameter segment matrix and the salinity parameter acquisition curve function matrix, and the deviation values between the depth parameter segment matrix and the depth parameter acquisition curve function matrix; S04. Construct a parameter association network diagram among temperature parameters, salinity parameters, and depth parameters. The three parameters are regarded as nodes in the diagram, and the association strength between the parameters is used as the edge weight. The minimum spanning tree algorithm is used to extract the key parameter association path, and a multi-parameter relationship coupling equation group among temperature parameters, salinity parameters, and depth parameters is established. S05, using a multi-parameter relationship coupling equation group to calculate the theoretical coupling value between the temperature parameter segment matrix, the salinity parameter segment matrix and the depth parameter segment matrix in the newly sampled data; S06, calculating the actual coupling values between the temperature parameter segment matrix, the salinity parameter segment matrix, and the depth parameter segment matrix in the newly sampled data; S07. Calculate a coupling deviation matrix between theoretical coupling values and actual coupling values; S08, using a pre-trained temperature-salinity-depth error labeling neural network model to replace the traditional error judgment function, inputting the deviation value obtained in step S03 and the coupling deviation matrix obtained in step S07 into the temperature-salinity-depth error labeling neural network model to calculate the error judgment value; S09: Compare the error judgment value with a preset threshold value. When the error judgment value is greater than the preset threshold value, mark the corresponding data point as an error.

[0021] Among them, the parameter acquisition curve function matrix is a matrix composed of polynomial functions obtained by fitting historical measurement data through the least squares method, which is used to describe the regular characteristics of parameter changes over time or space.

[0022] Among them, the time window refers to dividing the continuously collected data into multiple data segments according to fixed time intervals. Each data segment contains several continuous sampling points, which are used for local analysis of parameter change characteristics.

[0023] The deviation value refers to the difference between the measured data and the predicted value of the theoretical model, which is quantified by calculating the Euclidean distance or Manhattan distance.

[0024] Among them, the multi-parameter relationship coupling equation group includes the temperature-salinity relationship equation, the temperature-depth relationship equation, the salt-depth relationship equation and the temperature-salinity-depth comprehensive relationship equation; The temperature-salinity relationship equation is used to describe the physical correlation between temperature and salinity. The input includes a temperature parameter segment matrix, a salinity parameter segment matrix, a temperature-salinity history correlation database, an ambient temperature change rate, and a seawater density parameter. The output is a temperature-salinity theoretical coupling value matrix. The temperature-depth relationship equation is used to describe the physical correlation between temperature and depth. The input includes a temperature parameter segment matrix, a depth parameter segment matrix, a temperature-depth history correlation database, a depth gradient coefficient, and a water pressure change rate. The output is a temperature-depth theoretical coupling value matrix. The salt-depth relationship equation is used to describe the physical correlation between salinity and depth. The input includes a salinity parameter segment matrix, a depth parameter segment matrix, a salt-depth historical correlation database, a depth stratification coefficient, and an ocean current influencing factor. The output is a salt-depth theoretical coupling value matrix. The temperature-salinity-depth comprehensive relationship equation is used to integrate the comprehensive physical correlation between the three parameters. The input includes the temperature parameter segment matrix, the salinity parameter segment matrix, the depth parameter segment matrix, the temperature-salinity-depth historical correlation database and the seawater state equation parameters. The output is the three-parameter comprehensive theoretical coupling value matrix.

[0025] The theoretical coupling value is the output result of the multi-parameter relationship coupling equation group calculated in step S05, including the temperature-salinity theoretical coupling value matrix, the temperature-depth theoretical coupling value matrix, the salt-depth theoretical coupling value matrix, and the three-parameter comprehensive theoretical coupling value matrix.

[0026] Among them, the actual coupling value refers to the actual mutual influence value between parameters directly calculated from the actual measurement data, including the temperature-salinity actual coupling value matrix, the temperature-depth actual coupling value matrix, the salt-depth actual coupling value matrix, and the three-parameter comprehensive actual coupling value matrix.

[0027] Among them, the coupling deviation matrix is a numerical matrix that represents the degree of difference between the theoretical coupling relationship and the actual coupling relationship. The larger the element value, the higher the possibility of data anomaly.

[0028] Among them, the error judgment function is used to comprehensively evaluate the abnormality of data points. The input includes temperature deviation value, salinity deviation value, depth deviation value, coupling deviation matrix, historical error distribution eigenvector and environmental interference factor, and the output is the error judgment value.

[0029] Among them, error marking is to add a specific identifier to the data record. Directly using general methods such as connecting marks or appending text or using a data set to store the addresses of all data records with errors can also be used. Other general methods can be used to indicate that the data point is abnormal or unreliable, which is convenient for screening and processing in subsequent data analysis.

[0030] Among them, the specific structure of the temperature-salinity-depth error labeling neural network model is a hybrid expert model based on the multi-head sparse attention mechanism, which includes an input layer, a multi-head sparse attention layer, a parameter encoding layer, a multi-layer perceptron layer and an output layer. The multi-head sparse attention layer is used to process the relationship between different parameters, the parameter encoding layer maps the original data to a high-dimensional feature space, the multi-layer perceptron layer performs feature fusion and nonlinear mapping, and the output layer generates the final error judgment value.

[0031] Among them, the steps of establishing the training data set in the pre-training process of the temperature, salinity and depth error labeling neural network model specifically include collecting a large amount of historical temperature, salinity and depth measurement data and manually labeled error samples, cleaning and normalizing the collected data, segmenting the data according to the length of the time window and extracting features, calculating the statistical correlation between each feature and the manually labeled error, screening the feature combinations with strong correlation to form training sample pairs, and dividing the sample set into training set, validation set and test set in a ratio of 8:1:1.

[0032] Among them, the pre-training steps of the temperature-salinity-depth error labeling neural network model specifically include initializing the network parameters and setting the learning rate, batch size and number of training rounds, using the training set data for forward propagation to calculate the loss, using the adaptive moment estimation optimization algorithm for backpropagation to update the parameters, using the validation set to regularly evaluate the model performance and adjust the hyperparameters, focusing on learning samples of different error types to solve the sample imbalance problem, terminating the training early when the validation set performance no longer improves to avoid overfitting, using the test set for final performance evaluation and saving the trained model parameters.

[0033] Among them, the number of heads in the multi-head sparse attention mechanism matches the number of time windows in step S02, the sparsity of the sparse attention matrix is adapted to the sparse characteristics of the coupling bias matrix calculated in step S07, and the hidden layer dimension of the attention layer is equal to the sum of the dimensions of the three bias values calculated in step S03.

[0034] The specific implementation methods of the above steps are described in detail below. The specific implementation method of step S01 is to use historical measurement data to construct a temperature, salinity and depth parameter acquisition curve function matrix. First, a large amount of historical temperature, salinity and depth measurement data are collected and sorted according to time series or spatial distribution. The historical data of each parameter are fitted with a polynomial using the least squares method. The order of the fitting polynomial is selected according to the complexity of the data, and 3-5 orders are generally selected. For the temperature parameter, a polynomial function that characterizes the change of temperature with time or depth is fitted; for the salinity parameter, a polynomial function that characterizes the change of salinity with time or depth is fitted; for the depth parameter, a polynomial function that characterizes the change of depth with time is fitted. All fitting functions are organized into a matrix form, with each row corresponding to a time point or depth point, and each column corresponding to a polynomial coefficient. The purpose of this step is to establish a baseline model for parameter changes, provide a theoretical basis for subsequent deviation calculations, and effectively identify abnormal change patterns in new data by fitting the regular characteristics of historical data.

[0035] The specific implementation method of step S02 is to perform time window segmentation processing on the newly sampled data. First, determine the appropriate time window length, which is generally selected according to the data sampling frequency and the characteristics of the marine environment change, and the typical value is 5-15 minutes. The continuously collected temperature, salinity and depth data are divided into several data segments according to the set time window. For temperature data, a temperature parameter segment matrix is formed, in which each row represents a time window and each column represents a sampling point within the window; for salinity data, a salinity parameter segment matrix is formed; for depth data, a depth parameter segment matrix is formed. Statistical features are extracted for each data segment, including mean, variance, slope, kurtosis, etc., as the feature vector of the data segment. The purpose of data segmentation processing is to decompose long time series into short time periods, facilitate local analysis of parameter change characteristics, and improve the time resolution and sensitivity of anomaly detection.

[0036] The specific implementation of step S03 is to calculate the deviation value between the parameter segment matrix and the parameter acquisition curve function matrix. For each temperature parameter segment, the Euclidean distance or Manhattan distance is calculated between it and the predicted value of the temperature parameter acquisition curve function in the corresponding time interval to obtain the temperature deviation value. During the calculation process, the actual temperature data point in each time window is firstly deviated from the theoretical model predicted value, and then the distance is calculated. For the Euclidean distance, the calculation formula is based on the square root of the sum of the squares of the deviations of each point; for the Manhattan distance, it is based on the sum of the absolute values of the deviations of each point. Similarly, the deviation value between the salinity parameter segment and the salinity parameter acquisition curve function and the deviation value between the depth parameter segment and the depth parameter acquisition curve function are calculated. The three calculated deviation values are organized into a deviation value matrix for subsequent abnormality judgment. The purpose of calculating the deviation value is to quantify the degree of difference between the newly sampled data and the historical law, and to provide a basic basis for error identification. The larger the deviation value, the further the data deviates from the historical law, and the higher the possibility of abnormality.

[0037] The specific implementation method of step S04 is to construct a parameter association network diagram and establish a multi-parameter relationship coupling equation group. First, the three parameters of temperature, salinity and depth are used as nodes of the network diagram, and the mutual information or correlation coefficient between the parameters is calculated based on historical data as the weight of the edge. For example, the Pearson correlation coefficient or mutual information value between temperature and salinity, temperature and depth, and salinity and depth are calculated. The larger the weight, the stronger the correlation between the parameters. A minimum spanning tree algorithm such as Prim's algorithm or Kruskal algorithm is used to extract key parameter association paths from the parameter association network diagram, retaining the edges with the highest association strength. Based on the extracted key association paths, a multi-parameter relationship coupling equation group is established, including the temperature-salinity relationship equation, the temperature-depth relationship equation, the salt-depth relationship equation, and the temperature-salinity-depth comprehensive relationship equation. The establishment of the coupling equation can adopt multiple regression analysis, neural network or physical oceanography model. The purpose of constructing the parameter association network is to explore the inherent physical correlation between the parameters. The minimum spanning tree algorithm helps to reduce complexity and retain the most important correlation relationships, laying the foundation for subsequent coupling deviation analysis.

[0038] The specific implementation of step S05 is to calculate the theoretical coupling value using a multi-parameter relationship coupling equation system. For the temperature-salinity relationship equation, the input includes a temperature parameter segment matrix, a salinity parameter segment matrix, a temperature-salinity historical correlation database, an ambient temperature change rate, and a seawater density parameter. The temperature-salinity theoretical coupling value matrix is calculated through the equation. For the temperature-depth relationship equation, the input includes a temperature parameter segment matrix, a depth parameter segment matrix, a temperature-depth historical correlation database, a depth gradient coefficient, and a water pressure change rate. The temperature-depth theoretical coupling value matrix is calculated. For the salt-depth relationship equation, the input includes a salinity parameter segment matrix, a depth parameter segment matrix, a salt-depth historical correlation database, a depth stratification coefficient, and an ocean current influencing factor. The salt-depth theoretical coupling value matrix is calculated. For the temperature-salinity-depth comprehensive relationship equation, the input includes three parameter segment matrices, a temperature-salinity-depth historical correlation database, and seawater state equation parameters. The three-parameter comprehensive theoretical coupling value matrix is calculated. The purpose of calculating the theoretical coupling value is to predict the ideal correlation state between parameters based on historical data and physical models, providing a benchmark for comparison with actual coupling values.

[0039] The specific implementation of step S06 is to calculate the actual coupling values between the parameter segment matrices in the newly sampled data. For the temperature and salinity parameters, the covariance or mutual correlation coefficient of temperature and salinity in each time window is directly calculated to form a temperature-salinity actual coupling value matrix. For the temperature and depth parameters, the covariance or mutual correlation coefficient of temperature and depth in each time window is calculated to form a temperature-depth actual coupling value matrix. For the salinity and depth parameters, the covariance or mutual correlation coefficient of salinity and depth in each time window is calculated to form a salt-depth actual coupling value matrix. For the comprehensive relationship between the three parameters, multivariate statistical methods such as principal component analysis or canonical correlation analysis are used to calculate the comprehensive correlation strength of the three parameters to form a three-parameter comprehensive actual coupling value matrix. The purpose of calculating the actual coupling value is to directly extract the mutual relationship between the parameters from the actual measurement data, providing a practical basis for comparison with the theoretical coupling value.

[0040] The specific implementation of step S07 is to calculate the coupling deviation matrix between the theoretical coupling values and the actual coupling values. The temperature-salinity theoretical coupling value matrix is subtracted from the actual temperature-salinity coupling value matrix to obtain the temperature-salinity coupling deviation matrix. The temperature-depth theoretical coupling value matrix is subtracted from the actual temperature-depth coupling value matrix to obtain the temperature-depth coupling deviation matrix. The salt-depth theoretical coupling value matrix is subtracted from the actual salt-depth coupling value matrix to obtain the salt-depth coupling deviation matrix. The three-parameter comprehensive theoretical coupling value matrix is subtracted from the three-parameter comprehensive actual coupling value matrix to obtain the three-parameter comprehensive coupling deviation matrix. All coupling deviation matrices are normalized to facilitate subsequent comprehensive judgment. The purpose of calculating coupling deviations is to quantify the degree of difference between the actual parameter relationship and the theoretical expectation. A larger deviation indicates a higher likelihood of data anomaly. Coupling deviation analysis can capture abnormal patterns that may be overlooked by single-parameter deviation analysis.

[0041] The specific implementation of step S08 is to calculate the error judgment value using a pre-trained temperature, salinity and depth error labeling neural network model. The temperature deviation value, salinity deviation value, depth deviation value obtained in step S03 and the coupled deviation matrices obtained in step S07 are combined into a feature vector. The feature vector is input into the pre-trained temperature, salinity and depth error labeling neural network model. The neural network model is based on a hybrid expert structure with a multi-head sparse attention mechanism. It first receives the feature vector through the input layer, and then processes the relationship between different parameters in the multi-head sparse attention layer. The number of heads matches the number of time windows. The output of the attention layer is mapped to a high-dimensional feature space through the parameter encoding layer, and then feature fusion and nonlinear mapping are performed through the multi-layer perceptron layer. The final output layer generates an error judgment value ranging from 0 to 1. The closer the value is to 1, the higher the possibility of data anomaly. The advantage of the neural network model replacing the traditional error judgment function is that it can adaptively learn complex data anomaly patterns and improve the accuracy and robustness of error recognition.

[0042] The specific implementation of step S09 is to compare the error judgment value with the preset threshold and mark the error. First, set an appropriate error judgment threshold, which is generally balanced according to application requirements and acceptable false alarm rate / missing alarm rate, with a typical value of 0.7-0.85. Compare the error judgment value calculated in step S08 with the preset threshold. When the error judgment value of a certain time window or data point is greater than the preset threshold, add a specific identifier to the corresponding data record, such as setting the flag bit to 1 or adding a special marking symbol. Error marking can be divided into different levels, such as slight abnormality (0.7-0.85), moderate abnormality (0.85-0.95) and severe abnormality (0.95-1.0). The purpose of error marking is to provide a reference for subsequent data processing, so that data analysts or automated systems can perform data screening and quality control. The marked data can be processed by methods such as elimination, correction or weight reduction according to application requirements.

[0043] The specific structure and implementation of the temperature-salinity-depth error labeling neural network model are as follows: This model employs a hybrid expert model architecture based on a multi-head sparse attention mechanism. The input layer receives the feature vector, which includes three parameter bias values and elements of the coupling bias matrix. The multi-head sparse attention layer consists of multiple parallel attention sub-layers, each focusing on a different feature combination. The number of heads is designed to match the number of time windows. The attention mechanism determines feature importance by calculating the similarity between the query vector, key vector, and value vector, and introduces sparsity constraints to reduce computational complexity. The parameter encoding layer uses a fully connected network structure to map the original features into a high-dimensional latent space, enhancing feature representation. The multi-layer perceptron layer consists of multiple fully connected layers, each followed by batch normalization and ReLU activation functions, responsible for feature fusion and nonlinear mapping. The output layer is a single neuron that uses a sigmoid activation function to output an error judgment value in the range of 0-1. The model also includes multiple expert networks and a gating network that dynamically selects the most appropriate expert network for prediction based on the input features. The network training adopts the back-propagation algorithm, the loss function uses the cross entropy loss or binary focus loss, and the optimizer adopts the Adam algorithm.

[0044] Furthermore, the specific implementation process for establishing a training dataset for the temperature, salinity, and depth error labeling neural network model is as follows: First, at least 10,000 sets of historical temperature, salinity, and depth measurement data and manually labeled error samples are collected to ensure coverage across diverse sea areas, seasons, and environmental conditions. The collected data is cleaned to remove obviously unreasonable values, such as temperature, salinity, or depth records outside the physically possible range. The data is normalized using min-max or z-score normalization to scale all parameter values to a uniform range. The continuous data is segmented according to a set time window length, and statistical features such as mean, standard deviation, skew, and kurtosis, as well as frequency domain features such as power spectral density, are extracted from each segment. The mutual information or correlation coefficient between each extracted feature and the manually labeled errors is calculated. Feature combinations with mutual information or correlation coefficients exceeding 0.3 are selected to form a valid feature set. The samples are randomly divided into training, validation, and test sets in a ratio of 8:1:1, ensuring a consistent ratio of abnormal to normal samples in each set. To address sample imbalance, we use a combination of oversampling and undersampling techniques or adjust sample weights to ensure the model has equal recognition capabilities for all types of errors. We also use data augmentation techniques such as adding random noise and sliding time windows to expand the training sample, improving the model's generalization and anti-interference capabilities.

[0045] The mathematical model or calculation process involved in the present invention is described in detail below.

[0046] The process of establishing the temperature parameter acquisition curve function matrix, the salinity parameter acquisition curve function matrix, and the depth parameter acquisition curve function matrix in step S01 involves polynomial fitting calculations, which are specifically expressed as follows: For the temperature parameter acquisition curve function matrix, polynomial fitting can be expressed as: ; Where, is the fitted temperature function; is the time variable; is the temperature polynomial coefficient; is the polynomial order, usually 3-5; is the temperature fitting error term.

[0047] For the salinity parameter acquisition curve function matrix, polynomial fitting can be expressed as: ; Where, is the fitted salinity function; is the time variable; is the salinity polynomial coefficient; is the polynomial order, usually 3-5; is the salinity fitting error term.

[0048] For the depth parameter acquisition curve function matrix, polynomial fitting can be expressed as: ; Where, is the depth function of the fitting; is the time variable; is the depth polynomial coefficient; is the polynomial order, usually 3-5; is the depth fitting error term.

[0049] Temperature parameter acquisition curve function matrix It can be expressed as: ; Where, Indicates the Temperature polynomial coefficients for time periods or depth segments The value of is the number of time periods or depth segments; is the polynomial order.

[0050] Similarly, the salinity parameter acquisition curve function matrix And depth parameter acquisition curve function matrix Can be expressed as: ; ; Among them, the least squares method is used to obtain the polynomial coefficients. Taking temperature as an example, the following optimization problem is solved: ; Where, For the The actual temperature value at a time point; is the fitting temperature value at the corresponding time point; is the number of historical data points. Solve for each coefficient After that, the temperature parameter acquisition curve function matrix can be constructed.

[0051] These polynomial fits use power terms because ocean parameters often exhibit complex nonlinear variations. Power functions can approximate a variety of complex curves with simple and efficient computation. High-order terms capture subtle characteristics of parameter variations, while low-order terms reflect overall trends. The inclusion of error terms accounts for measurement errors and environmental interference, enhancing the robustness of the model.

[0052] The time window segmentation process in step S02 involves the construction of a parameter segment matrix, which is specifically expressed as follows: Temperature parameter segment matrix is expressed as: ; Where, Indicates the In the time window The temperature value of each sampling point; is the number of time windows; is the number of sampling points in each window.

[0053] Similarly, the salinity parameter segment matrix and the depth parameter segment matrix Can be expressed as: ; ; In step S03, the deviation value between the parameter segment matrix and the parameter acquisition curve function matrix is calculated, which is specifically expressed as follows: Temperature deviation The calculation uses Euclidean distance: ; Where, For the In the time window The actual temperature value of each sampling point; is the fitted temperature value at the corresponding time point.

[0054] Or using Manhattan distance: ; Similarly, the salinity deviation value and depth deviation value Can be expressed as: or ; or ; Deviation matrix It can be expressed as: ; The reason for choosing Euclidean distance or Manhattan distance as deviation metrics is that they quantify the degree of data deviation from a geometric and path perspective, respectively. Euclidean distance considers spatial straight-line distances and is suitable for detecting sudden changes, while Manhattan distance considers cumulative differences along the coordinate axes and is more sensitive to gradual anomalies.

[0055] In step S04, a parameter association network diagram is constructed, and the association strength between parameters is calculated as follows: Pearson correlation coefficient calculation between parameters: ; Where, For the Correlation coefficient between temperature and salinity within a time window; and are the average values of temperature and salinity within the window, respectively.

[0056] Similarly, and represent the correlation coefficients between temperature and depth, and between salinity and depth, respectively.

[0057] Or use mutual information to calculate the parameter correlation strength: ; Where, For the The mutual information value between temperature and salinity in a time window; is the temperature value and salinity value The joint probability of and Temperature values and salinity values The marginal probability of .

[0058] The parameter association network graph can be represented as an adjacency matrix : ; Where, 、 and are the correlation strengths between temperature and salinity, temperature and depth, and salinity and depth, respectively, and can be taken as the average or weighted sum of the correlation coefficients or mutual information in each time window.

[0059] The key parameter association path extracted based on the minimum spanning tree algorithm can be expressed as an edge set : ; Where, represents the retained edge; is the number of nodes, that is, the number of parameters 3; the minimum spanning tree contains Edge.

[0060] The correlation coefficient and mutual information are used to measure parameter association because they quantify the dependencies between variables from a linear and nonlinear perspective, respectively. The correlation coefficient is simple to calculate and is suitable for capturing linear relationships, while mutual information can detect more complex nonlinear relationships. The minimum spanning tree algorithm is used to simplify the network structure, retain the most important relationships, and reduce the complexity of subsequent calculations.

[0061] The specific expression of the multi-parameter relationship coupling equation group in step S05 is as follows: Temperature-salinity relationship equation: ; Where, For the The theoretical coupling value of temperature and salinity in a time window; and are the temperature and salinity data segments of the window respectively; is the temperature-salinity data mapping function; It is a temperature-salinity historical correlation database; is the rate of change of ambient temperature; is the seawater density parameter; It is a comprehensive mapping function of historical data and environmental parameters; and is the weight coefficient, satisfying ; is the temperature-haline coupling error term.

[0062] Temperature-depth relationship equation: ; Where, For the Theoretical temperature-depth coupling value of a time window; and are the temperature and depth data segments of the window respectively; is the temperature-depth data mapping function; It is the Wenshen historical correlation database; is the depth gradient coefficient; is the water pressure change rate; It is a comprehensive mapping function of historical data and environmental parameters; and is the weight coefficient, satisfying ; is the temperature-depth coupling error term.

[0063] Salt-depth relationship equation: ; Where, For the Theoretical coupling value of salt depth in a time window; and are the salinity and depth data segments of the window respectively; is the salt depth data mapping function; It is a salt depth historical correlation database; is the depth stratification coefficient; is the ocean current influence factor; It is a comprehensive mapping function of historical data and environmental parameters; and is the weight coefficient, satisfying ; is the salt-depth coupling error term.

[0064] The comprehensive relationship equation of temperature, salinity and depth: ; Where, For the The three-parameter comprehensive theoretical coupling value of the time window; 、 and They are the temperature, salinity and depth data segments of the window respectively; is a three-parameter data mapping function; It is a historical correlation database of temperature, salinity and depth; is the parameter set of the seawater equation of state; It is a comprehensive mapping function between historical data and state equation; and is the weight coefficient, satisfying ; is the three-parameter comprehensive coupling error term.

[0065] The method for obtaining each parameter is as follows: Ambient temperature change rate Calculation of temperature differences over successive time windows: , For the The average temperature of the windows, is the time window interval.

[0066] Seawater density parameters Calculated based on the international seawater state equation: ,in is the temperature, is salinity, For pressure.

[0067] Depth gradient coefficient Calculate the depth change rate of continuous sampling points: .

[0068] Water pressure change rate Calculated based on depth change: ,in is the acceleration due to gravity.

[0069] Depth stratification coefficient Calculated by the stratified characteristics of salinity with depth: .

[0070] Ocean current influencing factors Based on regional ocean current databases or measured by acoustic Doppler current profilers.

[0071] Seawater equation of state parameter set Contains various thermodynamic coefficients, obtained based on the International Maritime Convention standards.

[0072] The historical correlation database is established through long-term observation data and contains parameter correlation characteristics of different sea areas and seasons.

[0073] These equations are formulated as linear combinations to account for both the practical characteristics of current data (the first term) and the constraints of historical patterns and physical models (the second term). Weighting coefficients adjust the importance of each according to actual needs. The inclusion of error terms accounts for model uncertainty and the influence of random factors. The overall equation design adheres to fundamental principles of ocean physics and effectively captures the complex interrelationships between parameters.

[0074] The calculation of the actual coupling value in step S06 is specifically expressed as follows: The actual temperature-salinity coupling value is calculated using covariance: ; Where, For the Actual temperature-salinity coupling value in a time window; and The first Temperature and salinity values at each sampling point; and are the average values of temperature and salinity within the window, respectively; is the number of sampling points in the window.

[0075] Or using the cross-correlation coefficient: ; Similarly, the actual coupling value of temperature and depth and actual coupling value of salt depth Can be calculated similarly.

[0076] The actual coupling value of the three parameters is calculated using the standard correlation analysis: ; Where, For the The actual coupling value of the three parameters in a time window; and are the parameter grouping matrices within the window respectively; and is the norm vector; Represents the correlation coefficient function.

[0077] Covariance and correlation coefficients are chosen to calculate actual coupling values because they directly reflect the statistical correlation between parameters, are computationally simple, and have clear physical meaning. Covariance reflects the absolute strength of correlation, while the correlation coefficient provides a normalized correlation measure. For a three-parameter comprehensive relationship, canonical correlation analysis can identify the maximum correlation pattern between multiple variables and is suitable for dealing with high-dimensional data association problems.

[0078] The calculation of the coupling deviation matrix in step S07 is specifically expressed as follows: Calculation of temperature-salinity coupling deviation matrix: ; Where, For the Thermohaline coupling deviation in each time window; and are the theoretical and actual coupling values of temperature and salinity in this window respectively.

[0079] Similarly, the temperature-depth coupling deviation , salt-depth coupling deviation and three-parameter comprehensive coupling deviation Can be calculated similarly.

[0080] Coupling deviation matrix It can be expressed as: ; Normalization uses the maximum-minimum normalization method: ; Where, is the normalized coupling deviation value; is the original coupling deviation value; and are the minimum and maximum values of all coupling deviations, respectively.

[0081] The absolute difference in coupling bias calculations is used because it directly quantifies the degree of inconsistency between theoretical expectations and actual observations, is computationally simple, and is easy to understand. Normalization brings different types of biases to the same scale, facilitating subsequent comprehensive analysis and comparison.

[0082] In step S08, the traditional error judgment function is replaced by the pre-trained temperature-salinity-depth error labeling neural network model, and the deviation value obtained in step S03 and the coupling deviation matrix obtained in step S07 are input into the temperature-salinity-depth error labeling neural network model to calculate the error judgment value.

[0083] The error judgment threshold in step S09 is generally set in the range of 0.7-0.85. The specific value should be balanced according to the application requirements and the acceptable false alarm rate / missing alarm rate. The error judgment rule can be expressed as: ; Where, For the Error markers for each time window or data point; is the error judgment value; is the preset threshold.

[0084] Different levels of error marking can be divided into: ; Where, For the The error level for each time window or data point; 1, 2, and 3 represent slight anomaly, moderate anomaly, and severe anomaly, respectively.

[0085] The reason for using the threshold method for error determination is that it implements a simple and clear binary classification strategy, which facilitates automated processing. Multi-level error marking provides a more detailed classification of abnormality levels, which facilitates subsequent differentiated processing.

[0086] Optionally, for a set of parameters, the deviation value can be quantified by Euclidean distance or Manhattan distance. For example, the temperature deviation value is calculated as follows: Euclidean distance calculation method: ; Where, For the In the time window The actual temperature value of each sampling point; is the fitting temperature value at the corresponding time point; is the number of sampling points in the window.

[0087] Manhattan distance calculation method: ; The theoretical basis for the deviation value is to quantify the degree of difference between actual measured data and theoretical model predictions, providing a basic metric for subsequent error identification. Euclidean distance considers spatial straight-line distance and is sensitive to outliers, while Manhattan distance reflects cumulative deviation and is more sensitive to persistent drift.

[0088] Specifically, the present invention's principle is based on a data error identification method that combines parameter association networks and deep learning. First, by establishing a collection curve function matrix for temperature, salinity, and depth, the regularity of each parameter's temporal or spatial variations is captured. Then, newly sampled data is segmented by time window, and the deviation between each parameter segment and the corresponding collection curve function matrix is calculated to preliminarily screen for data intervals that may contain anomalies.

[0089] The core innovation of this invention lies in constructing a correlation network diagram between the three parameters of temperature, salinity, and depth, using the minimum spanning tree algorithm to extract the correlation paths of key parameters, and establishing a set of multi-parameter relationship coupling equations. This step fully considers the physical coupling relationship between temperature, salinity, and depth in seawater. For example, the temperature-salinity relationship equation describes the density correlation between temperature and salinity, the temperature-depth relationship equation reflects the stratification characteristics of temperature with depth, and the salt-depth relationship equation reflects the change pattern of salinity with depth. Through these equations, the system can calculate the theoretical coupling value and compare it with the actual coupling value calculated from the actual measurement data to generate a coupling deviation matrix.

[0090] Furthermore, this invention innovatively introduces a hybrid expert neural network model based on a multi-head sparse attention mechanism, replacing the traditional error determination function. Pre-trained on a large amount of historical data, this model can learn the characteristic patterns of different error types while addressing the problem of sample imbalance. The multi-head sparse attention mechanism enables the model to simultaneously focus on data features from different time windows and adaptively adjust based on the sparse nature of the coupled bias matrix, significantly improving the model's ability to identify subtle anomalies.

[0091] This method, which combines parameter physical correlation and deep learning, enables the system to distinguish between normal environmental fluctuations and instrument measurement errors, thereby achieving high-precision automatic error identification and marking, and solving the problem of multi-parameter coupling anomalies that are difficult to handle with traditional methods.

[0092] A specific embodiment 1 of the present invention is provided below. The specific implementation of each step in this embodiment 1 is described in detail as follows.

[0093] The specific implementation of step S01 is to use historical measurement data to construct a temperature, salinity, and depth parameter acquisition curve function matrix. First, a large amount of historical temperature, salinity, and depth measurement data is collected and sorted according to time series or spatial distribution to ensure the continuity and representativeness of the data. The least squares method is used to perform polynomial fitting on the historical data of each parameter. The order of the fitting polynomial is selected according to the complexity of the data, and is generally selected from 3 to 5 orders. For the temperature parameter acquisition curve function matrix, the polynomial fitting can be expressed as: Where, is the fitted temperature function; is the time variable; is the temperature polynomial coefficient; is the polynomial order, usually 3-5; is the temperature fitting error term. For the salinity parameter acquisition curve function matrix, polynomial fitting can be expressed as: Where, is the fitted salinity function; is the time variable; is the salinity polynomial coefficient; is the polynomial order, usually 3-5; is the salinity fitting error term. For the depth parameter acquisition curve function matrix, the polynomial fitting can be expressed as: Where, is the depth function of the fitting; is the time variable; is the depth polynomial coefficient; is the polynomial order, usually 3-5; It is the depth fitting error term. All fitting functions are organized into a matrix form, and the temperature parameter acquisition curve function matrix It can be expressed as: Where, Indicates the Temperature polynomial coefficients for time periods or depth segments The value of is the number of time periods or depth segments; is the polynomial order. Similarly, the salinity parameter acquisition curve function matrix And depth parameter acquisition curve function matrix Can be expressed as: and Among them, the polynomial coefficients are obtained using the least squares method. Taking temperature as an example, the following optimization problem is solved: Where, For the The actual temperature value at a time point; is the fitting temperature value at the corresponding time point; is the number of historical data points. The principle of fitting a polynomial function using the least squares method is to obtain the optimal fit coefficient by minimizing the sum of squares between the actual observed values and the fitted values. This method effectively captures both the overall trends and local variations in the data, providing a reliable baseline model for subsequent deviation analysis. The purpose of this step is to establish a baseline model for parameter variation, providing a theoretical basis for subsequent deviation calculations. By fitting the regularities of historical data, anomalous patterns in new data can be effectively identified.

[0094] The specific implementation of step S02 is to perform time window segmentation processing on the newly sampled data. First, determine the appropriate time window length, which is generally selected based on the data sampling frequency and the characteristics of the ocean environment changes. The typical value is 5-15 minutes. The continuously collected temperature, salinity and depth data are divided into several data segments according to the set time window. For temperature data, a temperature parameter segment matrix is formed. Where, Indicates the In the time window The temperature value of each sampling point; is the number of time windows; is the number of sampling points in each window. Similarly, the salinity parameter segment matrix and the depth parameter segment matrix . Statistical features are extracted for each data segment, including mean, variance, skew, kurtosis, etc., as the feature vector of the data segment. The principle of time window segmentation is to decompose long time series into short time periods to facilitate local analysis of parameter change characteristics. The selection of window length needs to balance resolution and stability. If it is too short, it cannot reflect the complete change trend, and if it is too long, it may cover up local anomalies. The purpose of this step is to decompose long time series into short time periods to facilitate local analysis of parameter change characteristics and improve the time resolution and sensitivity of anomaly detection.

[0095] The specific implementation of step S03 is to calculate the deviation value between the parameter segment matrix and the parameter acquisition curve function matrix. For each temperature parameter segment, the Euclidean distance or Manhattan distance is calculated between it and the predicted value of the temperature parameter acquisition curve function in the corresponding time interval. The temperature deviation value is calculated using the Euclidean distance as follows: Where, For the In the time window The actual temperature value of each sampling point; is the fitted temperature value at the corresponding time point. Or it can be calculated using Manhattan distance: Similarly, the salinity deviation value and depth deviation value Can be calculated separately. Organize the three calculated deviation values into a deviation value matrix The principle of calculating the deviation value is to quantify the degree of difference between the newly sampled data and historical patterns through a metric function. Euclidean distance and Manhattan distance measure the degree of deviation from the geometric and path perspectives, respectively. The former is more sensitive to outliers, while the latter is more sensitive to persistent deviations. The purpose of this step is to quantify the degree of difference between the newly sampled data and historical patterns, providing a basic basis for error identification. The larger the deviation value, the further the data deviates from the historical pattern and the higher the possibility of anomaly.

[0096] The specific implementation of step S04 is to construct a parameter association network diagram and establish a multi-parameter relationship coupling equation system. First, the three parameters of temperature, salinity and depth are used as nodes of the network diagram, and the mutual information or correlation coefficient between the parameters is calculated based on historical data as the edge weight. The Pearson correlation coefficient between the parameters is calculated as follows: Where, For the Correlation coefficient between temperature and salinity within a time window; and are the average values of temperature and salinity in the window respectively. and Or use mutual information to calculate the parameter correlation strength: Where, For the The mutual information value between temperature and salinity in a time window; is the temperature value and salinity value The joint probability of and Temperature values and salinity values The marginal probability of the parameter association network graph can be represented as the adjacency matrix Where, 、 and The correlation strengths between temperature and salinity, temperature and depth, and salinity and depth can be calculated by taking the average or weighted sum of the correlation coefficients or mutual information of each time window. A minimum spanning tree algorithm such as Prim's algorithm or Kruskal's algorithm is used to extract the key parameter association paths from the parameter association network graph, and the edges with the highest association strength are retained. The key parameter association paths extracted based on the minimum spanning tree algorithm can be represented as an edge set Where, represents the retained edge; is the number of nodes, that is, the number of parameters 3; the minimum spanning tree contains Based on the extracted key correlation paths, a set of multi-parameter relationship coupling equations was established, including the temperature-salinity relationship equation, the temperature-depth relationship equation, the salt-depth relationship equation, and the combined temperature-salinity-depth relationship equation. The principle of constructing the parameter correlation network diagram is to visualize the relationships between parameters based on graph theory, while the minimum spanning tree algorithm extracts the most important relationships by retaining the edges with the smallest total weight, thereby reducing complexity. The purpose of this step is to explore the inherent physical correlations between parameters and lay the foundation for subsequent coupling deviation analysis.

[0097] The specific implementation of step S05 is to calculate the theoretical coupling value using a multi-parameter relationship coupling equation group. The temperature-salinity relationship equation is expressed as: Where, For the The theoretical coupling value of temperature and salinity in a time window; and are the temperature and salinity data segments of the window respectively; is the temperature-salinity data mapping function; It is a temperature-salinity historical correlation database; is the rate of change of ambient temperature; is the seawater density parameter; It is a comprehensive mapping function of historical data and environmental parameters; and is the weight coefficient, satisfying ; is the temperature-salinity coupling error term. The temperature-depth relationship equation is expressed as: Where, For the Theoretical temperature-depth coupling value of a time window; and are the temperature and depth data segments of the window respectively; is the temperature-depth data mapping function; It is the Wenshen historical correlation database; is the depth gradient coefficient; is the water pressure change rate; It is a comprehensive mapping function of historical data and environmental parameters; and is the weight coefficient, satisfying ; is the temperature-depth coupling error term. The salt-depth relationship equation is expressed as: Where, For the Theoretical coupling value of salt depth in a time window; and are the salinity and depth data segments of the window respectively; is the salt depth data mapping function; It is a salt depth historical correlation database; is the depth stratification coefficient; is the ocean current influence factor; It is a comprehensive mapping function of historical data and environmental parameters; and is the weight coefficient, satisfying ; is the salt-depth coupling error term. The comprehensive relationship equation between temperature, salt and depth is expressed as: Where, For the The three-parameter comprehensive theoretical coupling value of the time window; 、 and They are the temperature, salinity and depth data segments of the window respectively; is a three-parameter data mapping function; It is a historical correlation database of temperature, salinity and depth; is the parameter set of the seawater equation of state; It is a comprehensive mapping function between historical data and state equation; and is the weight coefficient, satisfying ; is the three-parameter comprehensive coupling error term. Ambient temperature change rate Calculation of temperature differences over successive time windows: , For the The average temperature of the windows, is the time window interval. Seawater density parameter Calculated based on the international seawater state equation: ,in is the temperature, is salinity, is the pressure. Depth gradient coefficient Calculate the depth change rate of continuous sampling points: Water pressure change rate Calculated based on depth change: ,in is the acceleration due to gravity. Depth stratification coefficient Calculated by the stratified characteristics of salinity with depth: . Ocean current influence factors Obtained from regional ocean current database or measured by acoustic Doppler current profiler. Seawater state equation parameter set This includes various thermodynamic coefficients, obtained based on International Maritime Convention standards. The principle behind the multi-parameter coupled equation system is to combine physical models with data-driven methods to establish mapping relationships between parameters. The equations use a linear combination format, taking into account both current data characteristics and historical constraints. This step aims to predict the ideal correlation between parameters based on historical data and physical models, providing a benchmark for comparison with actual coupling values.

[0098] The specific implementation of step S06 is to calculate the actual coupling value between the parameter segment matrices in the new sampled data. The actual coupling value of temperature and salinity is calculated using covariance: Where, For the Actual temperature-salinity coupling value in a time window; and The first Temperature and salinity values at each sampling point; and are the average values of temperature and salinity within the window, respectively; is the number of sampling points in the window. Or use the cross-correlation coefficient to calculate: . Similarly calculate the actual coupling value of temperature and depth and actual coupling value of salt depth The actual coupling value of the three parameters is calculated using the standard correlation analysis: Where, For the The actual coupling value of the three parameters in a time window; and are the parameter grouping matrices within the window respectively; and is the norm vector; Represents the correlation coefficient function. The principle behind calculating actual coupling values is to extract the statistical correlations between parameters directly from the measured data. Covariance reflects the absolute strength of correlation, while the correlation coefficient provides a normalized correlation measure. Normalized correlation analysis can identify the maximum correlation pattern between multiple variables. This step aims to directly extract the inter-parameter relationships from the actual measured data, providing a practical basis for comparison with theoretical coupling values.

[0099] The specific implementation of step S07 is to calculate the coupling deviation matrix between the theoretical coupling value and the actual coupling value. Temperature-salinity coupling deviation calculation: Where, For the Thermohaline coupling deviation in each time window; and are the theoretical and actual coupling values of temperature and salinity in this window respectively. Similarly, the temperature-depth coupling deviation is calculated , salt-depth coupling deviation and three-parameter comprehensive coupling deviation . Coupling deviation matrix . Normalize all coupling deviations: Where, is the normalized coupling deviation value; is the original coupling deviation value; and are the minimum and maximum values of all coupled deviations, respectively. The principle behind calculating coupled deviations is to directly quantify the degree of inconsistency between theoretical expectations and actual observations through absolute differences. Normalization unifies different types of deviations to the same scale, facilitating subsequent comprehensive analysis. This step aims to quantify the degree to which the actual parameter relationship differs from theoretical expectations. Larger deviations indicate a higher likelihood of data anomalies. Coupled deviation analysis can capture unusual patterns that may be overlooked by single-parameter deviation analysis.

[0100] The specific implementation of step S08 is the same as above and will not be repeated here.

[0101] The specific implementation of step S09 is to compare the error determination value with a preset threshold and mark the error. First, set an appropriate error determination threshold, generally based on the application requirements and the acceptable false alarm rate / missing rate. The typical value is 0.7-0.85. The error determination value calculated in step S08 is compared with the preset threshold. The determination rule can be expressed as: Where, For the Error markers for each time window or data point; is the error judgment value; is the preset threshold. Different levels of error marking can be divided into: Where, For the The error level for each time window or data point; 1, 2, and 3 represent mild, moderate, and severe anomalies, respectively. The principle of using thresholds for error determination is to implement a simple and clear binary classification strategy, facilitating automated processing. Multi-level error marking provides a more detailed classification of anomaly levels, facilitating subsequent differentiated processing.

[0102] To better understand and implement the present invention, Example 2 of a specific application scenario is provided below: Researchers conducted a 30-day ocean observation mission in a certain sea area, using a multi-parameter temperature-salinity-depth instrument for data collection. The instrument collected data every 10 seconds, including three parameters: temperature, salinity, and depth. During the observation process, the researchers found that the traditional single-threshold judgment method was unable to accurately identify measurement errors in complex ocean environments. Therefore, they decided to apply a self-labeling method for the measurement error data of the multi-parameter temperature-salinity-depth instrument.

[0103] First, the researchers constructed a function matrix for collecting temperature, salinity, and depth parameters based on historical observational data. They collected 84,320 sets of valid data from historical temperature, salinity, and depth measurements taken over the same sea area over the past five years. They used the least squares method to fit polynomials to these data, using fourth-order polynomials for temperature and salinity and a third-order polynomial for depth. The resulting polynomial coefficients are shown in Table 1: Table 1 Parameter acquisition curve function polynomial coefficient table

[0104] According to the above coefficients, the temperature parameter acquisition curve function matrix, salinity parameter acquisition curve function matrix and depth parameter acquisition curve function matrix were constructed, each containing 24 rows (corresponding to 24 hours in a day) and the number of columns of the corresponding order.

[0105] Next, the researchers segmented the newly sampled data into time windows. They set the time window length to 10 minutes, meaning each window contained 60 sampling points. The 30 days of continuous observation data resulted in 4,320 time windows, within which they generated a temperature, salinity, and depth parameter segment matrix.

[0106] Then, the deviation between the parameter segment matrix and the parameter acquisition curve function matrix was calculated. For each time window, the Euclidean distance was used to calculate the deviation between the actual temperature data and the theoretical prediction value, the deviation between the actual salinity data and the theoretical prediction value, and the deviation between the actual depth data and the theoretical prediction value. Some of the calculation results are shown in Table 2: Table 2 Parameter deviation values for some time windows

[0107] To construct the parameter association network diagram, the researchers calculated the correlation coefficients between the parameters. Taking the first time window as an example, the correlation coefficient between temperature and salinity was -0.78, the correlation coefficient between temperature and depth was -0.92, and the correlation coefficient between salinity and depth was 0.85. Based on the average correlation coefficients of all time windows, the adjacency matrix of the parameter association network diagram was constructed: Table 3 Parameter association network graph adjacency matrix table

[0108] The minimum spanning tree was extracted using the Prim algorithm, retaining the temperature-depth and depth-salinity edges with edge weights of 0.89 and 0.82, respectively.

[0109] Based on the extracted key parameter correlation paths, the researchers established a multi-parameter coupled equation system. For the temperature-salinity relationship equation, weight coefficients α1=0.7 and α2=0.3 were set; for the temperature-depth relationship equation, weight coefficients β1=0.8 and β2=0.2 were set; for the salt-depth relationship equation, weight coefficients θ1=0.75 and θ2=0.25 were set; and for the combined temperature-salinity-depth relationship equation, weight coefficients φ1=0.6 and φ2=0.4 were set. Environmental parameter measurements revealed an ambient temperature change rate of 0.026°C / minute, a seawater density parameter of ρ=1025.8 kg / m³, a depth gradient coefficient of 0.178 m / minute, a water pressure change rate of 1.794 kPa / minute, a depth stratification coefficient of 0.023 / m, and an ocean current influence factor of 0.135.

[0110] The theoretical coupling values are calculated using the above equations and parameters. Some of the calculation results are shown in Table 4: Table 4 Theoretical coupling values for some time windows

[0111] Next, the researchers calculated the actual coupling values. They used the covariance method to calculate the actual coupling values of temperature and salinity, temperature and depth, and salt and depth, and used canonical correlation analysis to calculate the combined actual coupling value of the three parameters. Some of the calculation results are shown in Table 5: Table 5 Actual coupling values for some time windows

[0112] Based on the theoretical coupling value and the actual coupling value, the coupling deviation matrix is calculated. Some of the calculation results are shown in Table 6: Table 6 Coupling deviation table of some time windows

[0113] All deviation values were normalized and formed into feature vectors, which were then fed into a pretrained temperature-salinity-depth error labeling neural network model. This model uses a hybrid expert architecture with a multi-head sparse attention mechanism, consisting of 12 attention heads, corresponding to the number of time windows. The model was trained using 15,000 sets of historical data and manually labeled error samples. Using the Adam optimizer with a learning rate of 0.001, the model achieved an F1 score of 0.92 on the test set after 350 epochs of training.

[0114] The error judgment value and marking results calculated by the model are shown in Table 7: Table 7 Error judgment results of some time windows

[0115] The researchers set the error threshold at 0.75. When the error threshold is greater than 0.75, the corresponding data point is marked as an error. The error levels are divided into: 0.75-0.85 is a slight anomaly (Level 1), 0.85-0.95 is a moderate anomaly (Level 2), and 0.95-1.0 is a severe anomaly (Level 3).

[0116] By applying this method, a total of 87 abnormal windows were detected in 4320 time windows, including 35 mild abnormalities, 42 moderate abnormalities, and 10 severe abnormalities.

[0117] To verify the effectiveness of their method, researchers manually inspected 10 time windows marked as abnormal and found that nine of them contained measurement anomalies, for an accuracy rate of 90%. They also randomly selected 100 unmarked time windows for inspection and found that two of them had missed detections, for a miss detection rate of 2%.

[0118] Traditional methods for detecting errors in temperature, salinity, and depth measurements rely primarily on threshold judgments for a single parameter. For example, if the temperature, salinity, or depth value falls outside a preset range, it is considered an anomaly. This simple and straightforward approach, however, has significant drawbacks: first, it cannot detect data that is within a reasonable range but is actually abnormal; second, it cannot comprehensively judge interrelated parameters; third, it cannot adapt to environmental changes in different sea areas and seasons; and fourth, it is prone to misjudgment in highly volatile ocean environments.

[0119] The method of the present invention achieves automatic error marking of temperature, salinity and depth measurement data by establishing a parameter acquisition curve function matrix, calculating parameter deviation values, constructing a parameter association network diagram, establishing a multi-parameter relationship coupling equation group, calculating theoretical and actual coupling values, and applying a neural network model. Compared with traditional methods, the present invention has the following significant advantages: First, it takes into account the physical correlation between parameters and can detect correlation anomalies that cannot be discovered by the single-parameter threshold method; second, it introduces the constraints of historical data and physical models to improve the accuracy of judgment; third, it uses the neural network model to adaptively learn complex anomaly patterns, reducing the false alarm rate and missed alarm rate; fourth, it achieves a subdivision of error levels, facilitating subsequent differentiated processing.

[0120] It should be noted that the variables involved in the present invention are explained in detail as shown in Tables 8, 9 and 10 below.

[0121] Table 8 Variable Explanation Table (Part 1)

[0122] Table 9 Variable Explanation Table (Part 2)

[0123] Table 10 Variable Explanation Table (Part 3)

[0124] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A self-labeling method for measurement error data of a multi-parameter temperature-salinity-depth instrument, characterized in that: include: Establishing a temperature parameter acquisition curve function matrix, a salinity parameter acquisition curve function matrix, and a depth parameter acquisition curve function matrix; The newly sampled data are segmented according to the time window to form a temperature parameter segment matrix, a salinity parameter segment matrix, and a depth parameter segment matrix. The deviation values between the temperature parameter segment matrix and the temperature parameter acquisition curve function matrix, the deviation values between the salinity parameter segment matrix and the salinity parameter acquisition curve function matrix, and the deviation values between the depth parameter segment matrix and the depth parameter acquisition curve function matrix are calculated respectively. A parameter association network diagram between the temperature parameters, salinity parameters, and depth parameters is constructed. The minimum spanning tree algorithm is used to extract the key parameter association path and establish a multi-parameter relationship coupling equation group. The theoretical coupling value and the actual coupling value are calculated using the multi-parameter relationship coupling equation group to obtain the coupling deviation matrix. The deviation value and the coupling deviation matrix are input into the temperature-salinity-depth error marker neural network model to calculate the error judgment value. Compare the error judgment value with the preset threshold and mark the error data.

2. The method for self-marking measurement error data of a multi-parameter temperature-salinity-depth instrument according to claim 1, characterized in that: The parameter acquisition curve function matrix is a matrix composed of polynomial functions obtained by fitting historical measurement data through the least squares method, which is used to describe the regular characteristics of parameter changes over time or space.

3. The self-marking method for measurement error data of a multi-parameter temperature-salinity-depth instrument according to claim 2, characterized in that: Time window refers to dividing the continuously collected data into multiple data segments according to fixed time intervals. Each data segment contains several consecutive sampling points, which are used for local analysis of parameter change characteristics.

4. The self-marking method for measurement error data of a multi-parameter temperature-salinity-depth instrument according to claim 3 is characterized in that: The deviation value refers to the difference between the measured data and the predicted value of the theoretical model, which is quantified by calculating the Euclidean distance or Manhattan distance.

5. The method for self-marking measurement error data of a multi-parameter temperature-salinity-depth instrument according to claim 4, characterized in that: The multi-parameter relationship coupling equation group includes the temperature-salinity relationship equation, the temperature-depth relationship equation, the salt-depth relationship equation and the temperature-salinity-depth comprehensive relationship equation.

6. The self-marking method for measurement error data of a multi-parameter temperature-salinity-depth instrument according to claim 5, characterized in that: The temperature-salinity relationship equation is used to describe the physical correlation between temperature and salinity. The input includes the temperature parameter segment matrix, the salinity parameter segment matrix, the temperature-salinity history correlation database, the ambient temperature change rate, and the seawater density parameter. The output is the temperature-salinity theoretical coupling value matrix.

7. The method for self-marking measurement error data of a multi-parameter temperature-salinity-depth instrument according to claim 6, characterized in that: The temperature-depth relationship equation is used to describe the physical correlation between temperature and depth. The input includes the temperature parameter segment matrix, the depth parameter segment matrix, the temperature-depth historical correlation database, the depth gradient coefficient, and the water pressure change rate. The output is the temperature-depth theoretical coupling value matrix.

8. The method for self-marking measurement error data of a multi-parameter temperature-salinity-depth instrument according to claim 7, characterized in that: The salt-depth relationship equation is used to describe the physical correlation between salinity and depth. The input includes the salinity parameter segment matrix, the depth parameter segment matrix, the salt-depth historical correlation database, the depth stratification coefficient, and the ocean current influencing factor. The output is the salt-depth theoretical coupling value matrix.

9. The method for self-marking measurement error data of a multi-parameter temperature-salinity-depth instrument according to claim 8, characterized in that: The temperature-salinity-depth comprehensive relationship equation is used to integrate the comprehensive physical correlation between the three parameters. The input includes the temperature parameter segment matrix, the salinity parameter segment matrix, the depth parameter segment matrix, the temperature-salinity-depth historical correlation database, and the seawater state equation parameters. The output is the three-parameter comprehensive theoretical coupling value matrix.

10. The method for self-marking measurement error data of a multi-parameter temperature-salinity-depth instrument according to claim 9, characterized in that: The theoretical coupling values include the temperature-salinity theoretical coupling value matrix, the temperature-depth theoretical coupling value matrix, the salt-depth theoretical coupling value matrix, and the three-parameter comprehensive theoretical coupling value matrix; The actual coupling value refers to the actual mutual influence value between parameters directly calculated from the actual measurement data, including the temperature-salinity actual coupling value matrix, the temperature-depth actual coupling value matrix, the salt-depth actual coupling value matrix, and the three-parameter comprehensive actual coupling value matrix.

Images